Sleepwalking Into a Crisis

David Friedberg, CEO of Ohalo, & Co-Host of All-In joins Sourcery to share his take on America's 250th anniversary, reflecting on where does the country actually stand? And how do the bottom 50% get left behind?

This is a real damn masterclass.

→ Listen on X, Spotify, YouTube, Apple

He argues the wealth gap is widely misunderstood, pointing to roughly $183 trillion in total US household net worth, compared to about $8 trillion held by billionaires and $4 trillion by the bottom half of Americans. His view is that the rich didn't take everything. Instead, the bottom 50% never got to own the productive assets that created the last 50 years of wealth, as Friedberg puts it: "The truth is that the bottom 50% were left behind because of bad tax policy, because of spending in healthcare, housing, and education, & because we didn’t give them access to owning capital.”

“I don’t blame Americans for finding themselves in the situation that they’re in. I blame the bad policy that brought us to this moment.”

“This can’t just be a CCP psyop. It’s coming from a place of hurt, of turmoil, of fundamental struggle.”

“I’m saying everyone is right, but we need to be honest about what has happened to get us here and what the right path is to get us out of it.”

Friedberg breaks down the policy decisions he believes created today's affordability crisis, including a tax code that taxes capital below labor, a Social Security system invested in Treasury bonds instead of equities since 1982, the shift away from pensions after ERISA, & government policies that drove housing, healthcare, & education costs higher. He argues those are the failures that have created the conditions driving the rise of socialism in America.

"Americans should feel good about America. We just have to fix the policies."

BIG thank you to David for coming on to share what he’s been working on behind the scenes, this was INSANE. So much research & data. A lot to learn from.

Subscribe now

Share

𝐓𝐈𝐌𝐄𝐒𝐓𝐀𝐌𝐏𝐒

(00:00) David Friedberg, CEO at Ohalo Genetics & Co-Host at All-In
(01:11) Is California functionally bankrupt?
(02:28) The real reason Americans can't afford anything anymore
(12:13) Why a wealth tax is really government theft
(21:22) Why blaming billionaires is a political strategy, not a solution
(25:33) How Socialism grows like a virus
(30:40) Inside Trump Accounts: a 401(k) for every American
(34:22) Why rational arguments can't compete with viral outrage
(40:56) Jeff Bezos's radical fix for the wealth gap
(45:47) 5 years till Social Security is bankrupt
(46:35) Is AI actually coming for your job?
(50:29) The K-shaped economy
(55:39) Five things America needs to fix to survive
(59:21) David's research process
(1:02:54) "It's about making Americans feel good again"

Brought to you by:

Brex—The intelligent finance platform: cards, expenses, travel, bill pay, banking—wrapped into a high-performance stack. Built for scale. Trusted by OpenAI, Anthropic, Vercel, Granola, Deepgram, & Sourcery.. teams that move fast AF. visit → brex.com/sourcery

Turing—Turing partners with frontier AI labs to improve model capabilities in coding, reasoning, tool use, & multimodality, as well as with Fortune 500 enterprises to build & deploy end-to-end agentic AI systems in mission-critical workflows Visit: turing.com/sourcery

VCX—VCX is the public ticker for private tech, allowing investors of all sizes to invest in venture capital. View The Portfolio at GetVCX.com

Deel—Deel is the global people platform that helps startups hire, manage, pay, and equip anyone, anywhere. Trusted by more than 35,000 fast-growing companies, Deel is the people platform that just works, so teams can scale without the chaos. Visit: deel.com/sourcery

Public-–Investing platform Public just launched Generated Assets, which lets you turn any idea into an investable index with AI. With Generated Assets, you can build, backtest, refine, and invest in any thesis with AI. Gone are the days of one-size-fits-all ETFs. Try it today: public.com/sourcery

David Friedberg on America’s Future

David Friedberg joined Sourcery to discuss the state of the US economy in the country’s 250th year. The conversation covered the structure of American wealth, California’s fiscal position, the policy decisions he identifies as the root of the affordability crisis, and the reforms he argues should replace the wealth tax proposals now circulating in California & Washington.

→ Listen on X, Spotify, YouTube, Apple

The Wealth Distribution Data

Friedberg’s starting point is the wealth data he says most people get wrong. Citing Federal Reserve figures, total US household net worth stands at $183 trillion. Billionaires hold roughly $8 trillion of that. The bottom 50% of Americans hold roughly $4 trillion. Extending the top cohort down to households worth $50 million or more raises its share to about $23 trillion, which still leaves roughly $158 to $170 trillion, the large majority, with the middle class.

The standard concentration narrative misses where American wealth actually sits. The middle class with ~$158T, largely through equity ownership in 401(k)s and home equity, captured most of the gains of the last half century. The failure is concentrated in the bottom 50% at $4T, which was never positioned to own productive assets at all.

“It’s not like the rich have taken everything from the poor. That’s not true. The truth is that the bottom 50% were left behind because of bad tax policy, because of spending in healthcare, housing, and education, & because we didn’t give them access to owning capital.”

That exclusion, in his view, is the material basis for the socialist momentum in current politics, and government spending has functioned as fuel on the fire rather than a corrective.

California’s Fiscal Position

We started this conversation all the way back at the Hill and Valley Forum in March, with Friedberg’s hot take that “California is functionally bankrupt.” The evidence lays in the current state budget, which he puts $30 to $40 billion short and bridged with an accounting maneuver, plus another $30 to $40 billion in new spending proposed through a constitutional amendment on this year’s ballot. Layered on top of that is a pension cliff and what he described as the largest tax increase in California history, in a state that already carries the highest tax rates in the country.

While California is actually not an isolated case, it’s certainly the big bus hurdling to the edge of the cliff. The state is running ahead of a national trajectory shaped by policy decisions made in the mid-20th century, and the symptoms are the same everywhere. Costs for housing, education, and healthcare have outpaced incomes for most of the population, while asset owners have pulled away.

“The cost of everything has run away from the majority of Americans. They can’t afford stuff. The majority of Americans can’t afford to pay housing, they can’t afford their student loan, they can barely afford food, they definitely can’t afford healthcare, and they don’t know if they’ll ever be able to buy a home.”

3 Structural Missteps in Tax Code, Social Security, & Pensions

The first misstep in Friedberg’s account is the inversion of capital and labor taxation. Capital gains are taxed at 15% to 20% while labor income is taxed at roughly 40%, despite capital compounding on its own once accumulated. He argues the rates should have been flipped from the start, with labor always taxed less than capital, and that the inversion is a direct driver of the wealth gap.

The second is the 1982 change to Social Security, which directed the trust fund’s assets exclusively into US Treasury bonds yielding an average of 3.5% per year. The fund holds $2.7 trillion today, all of it lent to a government that has already spent the proceeds. “If you had put all the extra money that was put into Social Security since 1982 into the S&P 500, the Social Security Trust Fund would have an extra $37 trillion in assets sitting in it today.” Those assets would have belonged to the bottom 50%, the same cohort that today holds $4 trillion.

The third is the retirement divergence that followed the 1974 ERISA law. In 1980, 40% of Americans had a pension plan. Today it is under 8%. Government workers held at 90% coverage across the same period. Private-sector workers were pushed into 401(k)s, which turned out to be the better structure because they held equities, and that equity ownership is how the middle class built its $170 trillion. The bottom 50%, relying on Social Security, received a 3.5% Treasury yield instead of a share of American business. He also cited the Congressional Budget Office timeline that Social Security reaches insolvency within 5 years.

Affordability Crisis in Education, Healthcare, & Housing

Friedberg attributes the cost explosions in all 3 categories to the structure of federal involvement. In education, unlimited federal student lending removed every pricing constraint. Loans flowed to any student, at any school, for any degree, regardless of price or likely income outcome, and institutions responded by expanding administrative staff 6x over 30 years against flat enrollment while raising tuition continuously. The result is 45 million Americans who graduated with student loan debt in the last 10 years. “If you end the federal student loan program, I think you solve 90% of the problem in education.”

Healthcare follows the same mechanism. With the federal government as the dominant payer, providers price against a counterparty with no discipline, and costs rise across the system.

In housing, postwar policy pushed universal homeownership and concentrated most household net worth in a single asset. Once that happened, sustaining middle-class wealth required policies that push home prices up every year, which priced the next generation out. “Because we made everyone put all of their net worth in their home, and therefore we had to keep home prices going up to keep everyone’s net worth going up, people got screwed.”

His broader claim is that without reforming the federal role in housing, education, and healthcare, costs continue to compound and no tax change matters.

Wealth Tax Debacle & Property Rights

Friedberg opposed California’s proposed wealth tax on constitutional rather than personal grounds, noting he is not a billionaire & would not be directly affected. His objection rests on private property as a founding principle, and he anchored it in John Adams, who wrote that property must be secured or liberty cannot exist, & who warned that once society admits property is not sacred, anarchy & tyranny commence.

“Property must be secured, or liberty cannot exist.”

“The moment the idea is admitted into society, that property is not as sacred as the laws of God, & that there is not a force of law & public justice to protect it, anarchy & tyranny commence.

If ‘Thou shalt not covet,’ & ‘Thou shalt not steal,’ were not commandments of Heaven, they must be made inviolable precepts in every society, before it can be civilized or made free.” — John Adams

His practical concern is threshold creep. Current proposals have variously targeted net worths above $50 million, then $10 million, and $1 million, and the floor keeps moving lower & lower. “Once you open the door that says the government can take any of your property, suddenly all property is the government’s property, which leads to mob rule.” Beyond the principle, he argues the tax fails on its own terms. It drives capital flight, deepens discontent, and does nothing to address the actual problem, which is that the majority of Americans own no assets.

The episode connected this to Brian Singerman’s earlier appearance on Sourcery, where the former Founders Fund partner argued a wealth tax would ultimately hit the middle class harder than the top 1%. On why the proposals persist anyway, Friedberg pointed to electoral incentives. “It’s a lot easier to claim an enemy, get elected, than it is to fix a difficult policy problem.”

Proposed Reforms on Taxes & Universal Ownership

Friedberg’s alternative agenda starts with inverting the tax code he criticized. Capital gains rise to 40%, income taxes fall, borrowing against unrealized gains becomes a taxable realization event, tax-free asset transfer loopholes close, and the step-up in basis at inheritance ends. He also endorsed eliminating income tax entirely for the bottom 50% of earners, roughly those under $75,000 per year, a proposal put forward by Jeff Bezos, noting that cohort contributes only about 3% of total tax revenue. On the argument that a 40% capital gains rate would deter investment, “What are you gonna do with your excess capital if the tax rate is 40% versus 20%? You’re not gonna invest it now? Of course you’re gonna invest it.”

The ownership side of the agenda builds on the structure behind the new Trump Accounts, also called Invest America Accounts. Friedberg wants Social Security converted to the same model immediately. Sell the trust fund’s $2.7 trillion in Treasuries, buy the S&P 500, and give every American an individual account holding equities. He would apply the same conversion to government pension plans, which he called poison on the grounds that they either bankrupt the states or fail the retirees depending on them.

The unifying metric he proposes for the country is economic mobility, defined concretely as converting 2% of Americans from labor to capital each year. At that rate, within 50 years every American would hold enough invested capital to make work a choice rather than a requirement. His argument against socialism runs through the same metric. Redistribution locks the population into labor permanently, while ownership is the only mechanism that moves people out of it.

The Invest America Accounts for Children

The Invest America accounts we pointed to launched on July 4, 2026, timed to the country’s 250th anniversary. Created under the One Big Beautiful Bill Act through the Invest America Act, with Senator Ted Cruz as lead author and Altimeter’s Brad Gerstner as the originator of the concept dating to 2020, the program seeds $1,000 from the US Treasury into an account for every US citizen child born between January 1, 2025 and December 31, 2028. The default investment is an ETF tracking the S&P 500, funds are locked until 18, fees are capped at 0.10%, & contributions are capped at $5,000 per year, with employers able to add up to $2,500 within that limit. At 18 the money carries traditional IRA tax treatment & can go toward education, a first home, or starting a business.

Invest America’s calculator, which assumes a hypothetical 10.5% annual return in line with long-run S&P 500 total returns, shows the $1,000 seed plus $25 per month reaching $22,466 by age 18. The accounts have no spending deadline. Left invested with the same $25 monthly contribution continuing, the balance reaches $2,040,857 by age 60. With an $100/mo contribution that account can hit over $6.5M!

Comparatively, an account funded at the $5,000 annual maximum is estimated by the administration to reach roughly $271,000 by 18 alone. Set against the 3.5% average Treasury yield Social Security contributions have earned since 1982, the same $1,000 seed would reach roughly $1,860 over 18 years.

Early traction has been substantial. Per the Treasury Department, more than 6M accounts were opened in the first weeks, 1.4 million of them receiving the seed contribution, and 86% of opened accounts belong to families earning under $200,000 per year. The Michael and Susan Dell Foundation pledged $6.25 billion to deposit $250 into the accounts of 25 million children age 10 and under in qualifying ZIP codes, Micron announced employee matching up to $1,000 per child, and Gerstner has predicted more than $100 billion in additional private commitments over the next 12 months, saying tens of billions are already committed but unannounced.

However, Gerstner & Friedberg describe the relationship to Social Security differently. Gerstner has called Social Security a sacred promise and described the accounts as “additive & create positive optionality for future generations” in a direct reply to Ro Khanna. Friedberg wants Social Security converted into the same structure. “So we’re at day one of Trump Accounts, or Invest America Accounts, or whatever they end up being called. Obviously, it turns a lot of people off by calling them Trump Accounts. But the idea is right. It’s a 401(k) for everyone. Social Security needs to flip into being a 401(k) for everyone tomorrow.” Critics point out that Social Security is structured as a guaranteed benefit while index accounts carry market risk, & a stretch of weak returns would land directly on account holders.

AI, Jobs, & the K-Shaped Economy

Friedberg rejected the claim that AI is currently eliminating jobs. “As soon as AI starts deleting jobs from the economy, I will change my position on this, but I need to see it factually happen first.” A 1963 Newsweek article predicted mainframes would eliminate American jobs, after which mathematician, statistician, accounting, and service employment all grew. The 1980s desktop era produced the same magazine covers and the same outcome, a boom in desk jobs.

In his framing, each computing wave raised the leverage of the worker operating it. A person using the mainframe, the desktop, or AI produces 10-100x more, becomes worth more as the operator of the tool, and growing businesses hire more of them. That is productivity and growth, not displacement. He paired the argument with skepticism about the current cycle, referencing the token bubble conversation and questioning whether enterprises are getting the value they are paying for.

On the K-shaped economy Ray Dalio raised in their March conversation, Friedberg reframed the divergence as wealth rather than jobs. The bottom 50% cannot save because costs consume their income, so they cannot accumulate assets, while savers compound wealth and spending on the other side of the K. The fix runs through lower costs and rising incomes rather than redistribution.

While on the income side, he pointed to company-led workforce training already underway, including Meta’s programs for software engineers, and plumbers, electricians, and contractors. Companies profiled on Sourcery run the same playbook, including Senra Systems, the SpaceX-alumni wire harness company in LA that trains workers with no prior background in 4 weeks, down from a 2-year timeline, and Hadrian in manufacturing. Friedberg’s view is that this is the market working as intended, with the caveat that the trend needs to be tracked.

A Libertarian Arguing Against His Own Interests

Friedberg was direct that this agenda cuts against his own politics. He identifies as a hardcore libertarian, a political philosophy that prioritizes individual liberty, private property rights, & minimal government taxation & intervention in markets & personal life. A reform agenda built on higher capital gains rates, closed transfer loopholes, & an ended step-up at inheritance runs opposite to that framework, and it would fall hardest on people in his position.

“I’m a hardcore libertarian, so all of this is foreign to me. What am I doing talking about charging more taxes? At the end of the day, we do live in a society with other people. Libertarianism works well if you’re willing to leave people behind and you don’t expect that the social structure will become disordered over time, but it does become disordered over time.”

Joe Lonsdale has made a version of the same point, that he is rich but would pay 90% in taxes if it actually worked. Friedberg agreed on the same condition.. i.e. if the money actually produces results.

“Right, if your money goes to something good. And this is why the money that’s being paid into Social Security & pension plans should not go into the government coffers. It should go into your own account. If you’re an individual putting money from your paycheck into something, why are you putting it into the government? You should have a healthcare account. You should have an education account. You should have a retirement account. These should be your accounts to manage, not the government.”

→ Listen on X, Spotify, YouTube, Apple

Subscribe now

Share

The material presented on Molly O’Shea’s website are my opinions only and are provided for informational purposes and should not be construed as investment advice. It is not a recommendation of, or an offer to sell or solicitation of an offer to buy, any particular security, strategy, or investment product. Any analysis or discussion of investments, sectors or the market generally are based on current information, including from public sources, that I consider reliable, but I do not represent that any research or the information provided is accurate or complete, and it should not be relied on as such. My views and opinions expressed in any website content are current at the time of publication and are subject to change. Past performance is not indicative of future results.

Paid Endorsement. Brokerage services by Open to the Public Investing Inc, member FINRA & SIPC. Advisory services by Public Advisors LLC, SEC-registered adviser. Crypto trading provided by Zero Hash LLC, licensed by the NYSDFS. Generated Assets is an interactive analysis tool by Public Advisors. Output is for informational purposes only and is not an investment recommendation or advice. See disclosures at public.com/disclosures/ga. Matched funds must remain in your account for at least 5 years. Match rate and other terms are subject to change at any time.

AI’s Biggest Downstream Winner

CJ Desai, CEO of MongoDB, joins Sourcery at the RAISE Summit in Paris. We cover why data is the downstream winner of the AI cycle, how hyperscaler capacity limits are pushing workloads back on-prem, the return of data sovereignty, & what a bank's real agentic architecture actually looks like.

→ Listen on X, Spotify, YouTube, Apple

"Data is the unsung hero and data is back. You cannot create an AI application without a great data layer, and your AI application is as good as your data."

"You see these hyperscalers, some of them are running out of capacity… the hyperscaler said, 'Sorry, we don't have a capacity.' And they are one of the top 50 customers for that hyperscaler."

CJ took over MongoDB (NASDAQ: MDB) in November 2025 after Dev Ittycheria's 11-year run, with a mandate to reposition the company as the default modern database for AI applications. MongoDB carries a market cap of roughly ~$25 billion. CJ runs on a cadence of speaking to 10 to 12 customers a week, and in this conversation he brings that view straight from frontier labs, AI-native startups, and Global 2000 enterprises.

Plus prediction markets under World Cup load, the open vs closed source model debate, data centers in space, and the mentors who shaped him.

We cover:

› The 3 classes of AI customers MongoDB serves
› Why hyperscalers are telling top-50 accounts "no capacity"
› On-prem, sovereign AI, and the data-center comeback
› ElevenLabs running 50M+ agents on MongoDB
› MongoDB (OLTP) vs Snowflake and Databricks (OLAP)
› Why there is no standardization in enterprise AI models
› Auto-scaling and the death of the expensive DBA

Special thank you to Brex, MongoDB, & AssemblyAI for helping make this RAISE AI Summit mini-series in Paris, France happen.

Subscribe now

Share

𝐓𝐈𝐌𝐄𝐒𝐓𝐀𝐌𝐏𝐒

(00:00) CJ Desai, CEO of MongoDB
(01:12) Top CEOs at the Raise AI Summit
(03:24) The 3 types of companies powering the AI boom
(07:25) Why on-prem is making a shocking comeback
(09:01) Hyperscalers are quietly running out of capacity
(11:06) What actually separates MongoDB from Snowflake and Databricks
(13:18) Why Frontier Labs treat MongoDB as their memory layer
(15:23) The truth about data labeling companies
(16:22) From Oracle intern to first-time CEO
(19:19) Becoming CEO during the AI chaos
(23:53) The World Cup crisis that tested MongoDB's scale
(28:19) The real complexity behind simple AI agents
(31:22) Open source vs. closed models: what customers actually pick
(34:36) CJ's honest take on data centers in space
(36:50) The mentors who shaped a first-time CEO
(40:32) MongoDB's next big bets
(41:36) Data is back?

Brought to you by:

Brex—The intelligent finance platform: cards, expenses, travel, bill pay, banking—wrapped into a high-performance stack. Built for scale. Trusted by OpenAI, Anthropic, Vercel, Granola, Deepgram, & Sourcery.. teams that move fast AF. visit → brex.com/sourcery

MongoDB–Millions of developers and more than 65,200+ customers across industries – including ~75% of the Fortune 100 – rely on MongoDB for their most important applications. With integrated capabilities for operational data, search, real-time analytics, & AI-powered data retrieval, MongoDB helps organizations everywhere move faster, innovate more efficiently, & simplify complex architectures. visit → mongodb.com/ai

AssemblyAI–Millions of developers use AssemblyAI to power their voice ai apps & features. One API gives you access to best-in-class speech-to-text, voice agent, & speech understanding models for both pre-recorded and real-time audio. Granola, ClickUp & HeyGen are scaling with AssemblyAI - get $50 of free credits today → AssemblyAI.com/sourcery

CJ Desai on the Surge in AI Database Demand

Desai got his start in databases at Oracle, went on to help scale ServiceNow from $1.5 billion in revenue past $10 billion as its COO, then ran product & engineering at Cloudflare. In November 2025 he took over as CEO of MongoDB.

We spoke with CJ at the RAISE Summit in Paris about 8 months in.

Hyperscalers are running out of capacity

Desai’s customers are telling him the big cloud providers are running low on capacity. “You see these hyperscalers, some of them are running out of capacity.” One case he brings up is a Fortune 100 company in Texas that went to move more workloads, AI included, into public cloud and got turned down. The provider said no. “Sorry, we don’t have a capacity.” The company is one of that provider’s top 50 customers.

So it kept a data center it had been planning to shut down and now runs those workloads itself. Another customer, a US telecom, was refused more regional capacity by one provider and signed with a second, and now juggles two clouds. Different models tend to live in different clouds, so a single app can end up with one sub-agent on GCP and another on Azure.

To Desai, this flips the usual assumption. “Most people get that wrong because everybody thought all the workloads are just gonna move to these hyperscalers.” Between the capacity crunch and the pull to keep proprietary data in-house, he sees on-prem and private data centers coming back, and not only at regulated companies.

Sovereignty adds to it. A large customer in Paris tells him it is keeping everything off public cloud. “Data sovereignty is highest priority. French regulations forces us to have that as the highest priority.” Working across clouds is part of why some customers choose MongoDB to begin with. “They chose MongoDB because we work in multiple clouds. We are not just a database that works in only one cloud, and they really like that for resiliency purposes.”

Databases as downstream winners

A lot of the AI attention goes to the models, but Desai sees MongoDB winning off the data underneath them. Every model and every agent throws off data that has to be stored and served, whatever model a customer picks. “The data layer is typically the unsung hero.” He does not separate the two. “We feel that models and data, both are needed to create a great agentic application.”

MongoDB handles that data at scale. “We do unstructured data really well. We are a scale out architecture.” Mongo does stands for Humongous.

The rest of the category is moving too. Databricks, still private, signed a term sheet in July 2026 for a Coatue-led round at a $188 billion valuation, about $3 billion per the Wall Street Journal, closing later this summer, up from $134 billion in February. Its revenue was running at about $5.4 billion a year in early 2026, growing more than 65%, with positive free cash flow. Snowflake (NYSE, SNOW), which is public, is worth around $93 billion, with roughly $5 billion in trailing revenue and guidance of $5.84 billion in product revenue for fiscal 2027, about 31% growth.

MongoDB is a different kind of database from those two. It is built for live, operational work rather than analytics, though the line is blurring as Databricks pushes Lakebase, a Postgres database aimed at AI agents, into operational territory Oracle has long held. In July 2026, Citi named MongoDB, Snowflake, and Palantir its top three software picks for the year.

Inside enterprise agent architectures

The agent systems Desai sees inside big companies are getting more complicated, not less. He asked one large New York bank to walk him through its whole setup, from the cloud provider down through databases, agent runtime, and models. “I wanna say 55 different boxes for the agentic architecture.” Half of them, he says, did not exist a year earlier, things like observability, security, evaluations, guardrails, and getting agents into production. The bank expects to add more.

Customers are not settling on one model either, and Desai admits he expected them to. “I had a very simple viewpoint of the world from a customer perspective. Oh, are you standardizing on this particular model.” Instead it is a mix. “There is no standardization even when it comes to these models. It goes anywhere from open source to closed source, small to large, horizontal to domain specific based on use cases.” For coding, customers reach for proprietary tools like Claude Code, Codex, Cognition, & Cursor. Some pull domain-specific open-source models off Hugging Face, but still.. “I have not seen a complete pivot towards open source or a complete pivot towards proprietary.”

Where the money is going in the AI stack

Menlo Ventures’ enterprise survey puts AI spending at $37 billion in 2025, up from $1.7 billion in 2023, about 6% of the global software market. In Menlo’s map, the data layer, the pipelines, databases, vector stores, and caches that feed models the right context, sits next to deployment and observability, not beneath the model. Other reads of the agent stack, like AImultiple’s, land in a similar place, that value spreads out more than it did in older software, where most of it sat at the app layer, and that pieces like reasoning, orchestration, and governance are the hard ones to copy.

Memory is becoming its own layer. Research from groups like mem0 sees dedicated memory, storage that carries context across sessions, turning into standard infrastructure in 2026 the way vector databases did in 2024. The recurring finding is that the top reason agents break in production is bad data, not the model or the framework. Gartner figures 40% of enterprise applications will use task-specific agents by the end of 2026, up from under 5%.

This is roughly where Desai puts MongoDB. The frontier labs are the ones he watches most. “Frontier labs, I would consider the holy grail for us,” with growth he calls faster than linear. They lean on MongoDB as a memory layer alongside inference and voice, image, and video work.

What fast-growing customers want

The fastest-growing customers want infrastructure that runs itself, Desai says, and they have stopped building things and waiting for demand. “You can’t come and think of, if I build it, they will come. That era is gone.” He points to a prediction-market company that ran low on capacity during the World Cup, with predictions landing every minute. The CEO texted him directly. “It looks like we may be out of capacity in few days on MongoDB, and the World Cup soccer is still couple of weeks away.”

Companies like that rarely staff for infrastructure, he says, since their focus is the models or the product, so they go direct to him. “Why am I calling you and telling you that I’m gonna run out of capacity? Why couldn’t you see it?” He compares it to the old Oracle database administrator, once the most expensive hire a company made, a job AI companies now skip. “We don’t have time to hire these people. If we are using your software, it should have its own machine-based database administration.”

Keeping up with all of that has no fixed answer. "There is no solution. You constantly have to pivot or change the roadmap." He ties the pace to the falling cost of building software. "When software writes software, the incremental cost of innovation has gone down," and the cost of trying things has dropped with it. "This is the slowest. I think it is going to get even faster."

3 kinds of AI customers

Desai splits AI demand into 3 groups. First are the frontier labs, more than one of them, which use MongoDB for inference, voice, image, video, and as a memory layer. He will not name them. Their usage, he says, climbs almost straight up on weekly or daily active users.

Second are the AI-native companies. Vibe-coding platforms like Emergent and Base44 are built on MongoDB, and robotics and physical-AI companies use it to hold what their machines throw off. “Every robot is generating data as they sense thing, as they act on things, and then all that data gets stored in MongoDB.” For scale, he points to ElevenLabs. “They have north of 50 million agents, depending on when you look at it, all running on MongoDB. That also gives us a lot of confidence that we have the right architecture for agentic workloads.”

Third are the big enterprises building agent workloads, and there Desai says it is still early. He has not seen customer-facing agents running at scale in airline or banking apps yet, and most of what he sees is prototyping. The customers run across travel, entertainment, and media, and include the food-delivery company Zomato, on MongoDB Atlas.

MongoDB’s talent density

Mati, founder of ElevenLabs told Desai that MongoDB’s staff are a hiring target. “Everybody told me we should hire salespeople from MongoDB because you guys do a really good job of training them. Same thing with our engineers because they understand distributed systems & data really well.” ..Which he obviously has mixed feelings about it..

TLDR: Data is back

“Data is the unsung hero & data is back. You cannot create an AI application without a great data layer, & your AI application is as good as your data.”

“Regardless of the era.. i.e. internet era, mobile era, after iPhone, now the AI era.. you always need a data layer.”

Subscribe now

Share

The material presented on Molly O’Shea’s website are my opinions only and are provided for informational purposes and should not be construed as investment advice. It is not a recommendation of, or an offer to sell or solicitation of an offer to buy, any particular security, strategy, or investment product. Any analysis or discussion of investments, sectors or the market generally are based on current information, including from public sources, that I consider reliable, but I do not represent that any research or the information provided is accurate or complete, and it should not be relied on as such. My views and opinions expressed in any website content are current at the time of publication and are subject to change. Past performance is not indicative of future results.

Paid Endorsement. Brokerage services by Open to the Public Investing Inc, member FINRA & SIPC. Advisory services by Public Advisors LLC, SEC-registered adviser. Crypto trading provided by Zero Hash LLC, licensed by the NYSDFS. Generated Assets is an interactive analysis tool by Public Advisors. Output is for informational purposes only and is not an investment recommendation or advice. See disclosures at public.com/disclosures/ga. Matched funds must remain in your account for at least 5 years. Match rate and other terms are subject to change at any time.

DAAAAAAMN

Day 2 of RAISE Summit in Paris, the biggest AI summit in Europe. 12 of the biggest names in AI, CEOs, CTOs, & Investors give us their hottest takes on the debates splitting the room right now: is token maxing genius or a waste, is open source dying or about to take over, and are we about to explode? (I’m just being dramatic)

→ Listen on X, Spotify, YouTube, Apple

The disagreements are the best part. Dylan Patel (SemiAnalysis) loves token maxing and says open source is dying quickly, while others call token maxing the wrong approach entirely. Qasar Younis (Applied Intuition) on the opulence phase of the cycle and the pullback he sees coming. Apoorv Agrawal (Altimeter) on the four seasons of AI and planning for the climate, not the weather. Plus enterprise ROI, physical AI, petabyte-scale search, and why data is back.

"Open is dying quickly." - Dylan Patel, SemiAnalysis.

"There are 4 seasons in AI. They're called OpenAI, Anthropic, SpaceX, Google." - Apoorv Agrawal, Altimeter.

Guest lineup:

• Dylan Patel, CEO, SemiAnalysis

• Qasar Younis, CEO, Applied Intuition

• Apoorv Agrawal, Partner, Altimeter Capital

• Laura Diorio, NYSE

• Arvind Jain, CEO, Glean

• Ariel Cohen, CEO, Navan

• CJ Desai, CEO, MongoDB

• Gil Feig, CTO, Merge

• Nikhil Benesch, CTO, TurboPuffer

• Barak Kaufman, Chief Strategy Officer, Wonderful

• Max Junestrand, CEO, Legora

• Marc Boroditsky, CRO, Nebius

Special thank you to Brex, MongoDB, & AssemblyAI for helping make this RAISE AI Summit mini-series in Paris, France happen.

Subscribe now

Share

𝐓𝐈𝐌𝐄𝐒𝐓𝐀𝐌𝐏𝐒

00:00 RAISE Summit Day 2
00:12 Qasar Younis (Applied Intuition)
03:46 AI Trends at RAISE
05:29 Applied Intuition Updates
09:46 Bubble Risks and Books
15:37 Self-Driving Future
17:29 Dylan Patel (SemiAnalysis)
22:05 Tokenmaxxing Debate
24:31 Marc Boroditsky (Nebius)
28:37 Arvind Jain (Glean)
32:02 Laura Diorio (NYSE)
35:03 Apoorv Agrawal (Altimeter): Four Seasons of AI
38:23 Nikhil Benesch (TurboPuffer)
42:18 CJ Desai (MongoDB): Agent Data Layer
43:00 Barak Kaufman (Wonderful)
45:34 Max Junestrand (Legora)
49:35 Gil Feig (Merge)
51:58 Ariel Cohen (Navan): Bullish on Humans
54:54 CJ Desai (MongoDB): Data Is Back

Brought to you by:

Brex—The intelligent finance platform: cards, expenses, travel, bill pay, banking—wrapped into a high-performance stack. Built for scale. Trusted by OpenAI, Anthropic, Vercel, Granola, Deepgram, & Sourcery.. teams that move fast AF. visit → brex.com/sourcery

MongoDB–Millions of developers and more than 65,200+ customers across industries – including ~75% of the Fortune 100 – rely on MongoDB for their most important applications. With integrated capabilities for operational data, search, real-time analytics, & AI-powered data retrieval, MongoDB helps organizations everywhere move faster, innovate more efficiently, & simplify complex architectures. visit → mongodb.com/ai

AssemblyAI–Millions of developers use AssemblyAI to power their voice ai apps & features. One API gives you access to best-in-class speech-to-text, voice agent, & speech understanding models for both pre-recorded and real-time audio. Granola, ClickUp & HeyGen are scaling with AssemblyAI - get $50 of free credits today → AssemblyAI.com/sourcery

The 12 Hottest Takes, RAISE Day 2

At the RAISE Summit in Paris, we pulled 12 of the biggest names in AI, CEOs, CTOs, & Investors aside for their hottest takes on where AI is heading.

Open Source vs Closed Models

Arvind Jain, CEO, Glean

Jain works across both closed and open models and expects the balance to flip inside two years. Glean routes to the cheapest capable model for each task through its context graph.

“Open source models are gonna dominate AI inferencing. You’ll see in the next two years, we’ll shift to where it’s almost no open source to where it’s gonna be almost all open source.”

“All these agents that you want to automate the work that humans do, they need that context, that data, that information that humans use to do the same work. Being the leader in context graphs is helping create a massive demand for Glean.”

“We can pick the right model. Since we work with all the closed domain and open source models, we can pick the right model for the right task, which is cheaper but still gets the work done.”

Dylan Patel, CEO, SemiAnalysis

Patel reads the direction differently, pointing to Chinese labs shifting from open releases to licensing.

“The theme here is that there’s multiple Chinese model labs who are telling all the inference guys, hey, our next model’s not gonna be open source, we’re gonna license it to you. So open is dying quickly, unfortunately.”

Is Tokenmaxxing Working?

Dylan Patel, CEO, SemiAnalysis

Patel treats token budgeting as a drag on adoption, with a caveat on ROI.

“I love tokenmaxxing. I think anyone who doesn’t tokenmaxx is gonna get left behind. I think all this token budgeting stuff is loser mentality.”

“If you’re token budgeting hardcore, then your people are not gonna learn the new workflows, and they’re gonna get left behind. They’re not gonna reshape your company and make it more efficient.”

“I got mad at someone ‘cause their model was set to Haiku, and they didn’t even realize. They spent like $1,000 on Haiku. They changed to Opus or Fable now, and they’re so much more productive.”

Gil Feig, CTO, Merge

Feig takes the opposite read, tying output to iteration speed rather than token volume.

“Tokenmaxxing, not working. You’re getting zero results from using more tokens. It was cool to encourage your employees to use more AI, but what we’re seeing now is we’re not getting more output.”

“The only thing that people are seeing a real connection between productivity and usage of AI is how you use AI to bring your cycle times down, get feedback, and iterate really quickly.”

The Infrastructure Build-Out

Dylan Patel, CEO, SemiAnalysis

Patel warns that building three-year infrastructure around today’s workloads leads to waste, and that memory costs are already flowing into token prices.

“A lot of people are trying to build all these optimized solutions for data centers, and they’re just looking at a backward-looking view. And then when they actually end up building that infra, a lot of it may be useless.”

“It feels like a lot of people are trying to optimize on the current rather than think about where the workload is heading. And that’s gonna lead to a lot of wasted infra spend.”

“We’ve seen next generation hardware receive price increases before they’ve even started production. We see that with cost of tokens not falling as fast as they were previously, because the cost of memory is flowing through.”

Marc Boroditsky, CRO, Nebius

Boroditsky points to enterprise adoption as the next stage and a diversified data center portfolio as Nebius’s edge.

“The discussion is shifting from what are AI natives building to when are enterprises adopting. It’s really incredible to see the bridges being built to see enterprise adoption take place.”

“The biggest problem with the build-out of data centers is something that Nebius is not experiencing, which is that a lot of people are focused on single individual projects. Today, we have 20 data centers.”

“The biggest challenge that many of our competitors are facing is being single-threaded, which is strange when you’re in a multi-threaded parallel processing industry and getting stuck with a single project.”

Storytelling & Public Markets

Laura Diorio, NYSE

Diorio points to partnerships & storytelling as what separates companies from overused mainstream buzzwords.

“The thing that’s making these companies stand out is their access and visibility. It’s their partnerships. We’re seeing it in real time, watching real C-suite founders, entrepreneurs actually meeting in person to close deals.”

“It is about the story you’re telling, how well you’re telling it, and where you’re gonna go from there, how it’s gonna set you apart.”

“If you have a clear message, and you have a clear goal with really good people, that’s what helps with being in person. You meet the actual, real good people.”

The Cycle: Bubble Risk & What Lasts

Qasar Younis, CEO, Applied Intuition

Younis flags cost structures and a coming pullback, draws a parallel to the collapse of LTCM, and argues physical AI is the longer story. Applied Intuition’s mission is to put intelligence on a billion machines.

“We’re at that phase of the bubble where all the opulence in the companies is what people talk about rather than the products. And so I think there’s gonna be a little bit of a pullback.”

“LTCM was this hedge fund in the late ‘90s that was a bunch of savants, geniuses, and they kinda do what’s happening now with compute, which is they just used leverage. Is there something that happens more peripherally in the AI business, and then it has a domino effect into our business?”

“Our mission is to get intelligence onto a billion machines. I think the physical world and AI going into that, when we look back, that’s gonna be the bigger story.”

4 Seasons of AI

Apoorv Agrawal, Partner, Altimeter Capital

Agrawal frames model choice as a climate to plan for rather than a single bet.

“There’s four seasons in AI. They’re called OpenAI, Anthropic, SpaceX, and Google. So if you’re about to make a multimillion, billion-dollar decision of which lab to build your intelligence on, plan for the climate, not for weather.”

“That’s multi-model routing. That’s evals. That’s thinking about post-train custom models. It’s not a state. It’s a four seasons of the year.”

“You’ll have a portion of your workloads going through open source, and then the frontier stuff, discovering new use cases, will always go through the most intelligent models. Planning to not get stuck in one regime is the climate.”

Search, Data, & Retrieval

Nikhil Benesch, CTO, TurboPuffer

Benesch powers retrieval for Cursor, Notion, Legora, and Anthropic, and argues search economics still cap what products can be built. TurboPuffer’s largest workloads cost tens of millions of dollars to search over.

“I think search is still too expensive, and TurboPuffer was founded because we thought search was an order of magnitude too expensive at the time. We’ve brought it down by an order of magnitude, but it kills us that there are still products out there that are limited in their ambition because search is still too expensive.”

“If you’re spending $5 on TurboPuffer per user, but you’re only charging your users $5 a month, the economics just don’t work. But if we can bring that cost down so you’re only spending 50 cents a user, suddenly you’re able to build a product where you couldn’t before.”

“We used to think in terms of gigabytes or terabytes. Folks are coming up to me with not just one, but 10 or hundreds of petabytes that they wanna search over.”

CJ Desai, CEO, MongoDB

Desai puts the data layer at the center of every AI application.

“Data is the unsung hero and data is back. You cannot create an AI application without a great data layer, and your AI application is as good as your data.”

“Number one is about customer-focused innovation and making sure we are the foundational layer for all AI applications and agentic applications.”

Going Global

Barak Kaufman, Chief Strategy Officer, Wonderful

Kaufman argues geography beats vertical specialization, run on an Uber-style go-to-market. Wonderful operates in over 30 countries.

“I believe geographies will be even bigger than verticals when it comes to AI. When you look at actual labor and labor displacement and where it’s likely to go, countries are way more interesting from an expansion perspective than verticals.”

“You can go horizontal without a vertical specialization with AI, as long as you’re able to solve the go-to-market strategy with it. It’s the enterprise applied AI playbook with the Uber go-to-market strategy.”

“What it requires is a non-consensus insight that you need to actually be right about. For Wonderful, that was the geographic focus, to go horizontal in terms of the use cases, but to focus on rest of world versus US.”

Max Junestrand, CEO, Legora

Junestrand calls out European startups for complaining rather than building for global scale. Legora was the first legal AI company to move to consumption-based pricing.

“My hottest take is that there is a lot of complaining from European startups rather than just locking in and building for the global stage. I think there’s a bit of laziness.”

“It’s nice in the summers here to go to Italy or to go to France. But if you wanna build the biggest companies in the world, you’re competing with the US, you’re competing with China. If you wanna win globally, you need to work as hard as they do.”

“We were the first legal AI company to move into consumption-based pricing. Token consumption is going to be the pricing model in legal, and I expect many of the other companies to follow us.”

Integrations, Security, & Sovereignty

Gil Feig, CTO, Merge

Feig locates the real risk at the integration layer & expects enterprises to demand control.

“Integrations are the point where an LLM actually becomes dangerous. Something isolated, an LLM, it might insult you, but it’s not gonna do much worse than that. But the second that thing can send your data elsewhere, that’s where all the problems come in.”

“There’s never been a good position in business to hand over the fate of your company, your future, your development, to another company. So people are gonna want sovereignty. They want control over their own systems. They do not wanna be locked in.”

Humans in the Loop

Ariel Cohen, CEO, Navan

Cohen argues humans stay central, especially in travel, where hallucinations carry real consequences. Navan grew usage 50% and revenue 40% last quarter and passed $10 billion in annual bookings.

“How humans are still so important, both for meeting and being here, but also supporting all of this. You cannot have an AI agent do everything, especially with travel, because it’s so complex.”

“We’ve built our own platform, our own model to prevent that, and we are supporting most of our customers by using our own AI platform that actually does not hallucinate.”

“We can do all of this AI infrastructure, the AI fun stuff, but we need to remember that we are all humans and this needs to stay.”

Subscribe now

Share

The material presented on Molly O’Shea’s website are my opinions only and are provided for informational purposes and should not be construed as investment advice. It is not a recommendation of, or an offer to sell or solicitation of an offer to buy, any particular security, strategy, or investment product. Any analysis or discussion of investments, sectors or the market generally are based on current information, including from public sources, that I consider reliable, but I do not represent that any research or the information provided is accurate or complete, and it should not be relied on as such. My views and opinions expressed in any website content are current at the time of publication and are subject to change. Past performance is not indicative of future results.

Paid Endorsement. Brokerage services by Open to the Public Investing Inc, member FINRA & SIPC. Advisory services by Public Advisors LLC, SEC-registered adviser. Crypto trading provided by Zero Hash LLC, licensed by the NYSDFS. Generated Assets is an interactive analysis tool by Public Advisors. Output is for informational purposes only and is not an investment recommendation or advice. See disclosures at public.com/disclosures/ga. Matched funds must remain in your account for at least 5 years. Match rate and other terms are subject to change at any time.

RAISE in Paris was HOT AF

Day 1 of RAISE Summit in Paris, the most high-profile AI summit in Europe. We pulled 10 of the biggest names in AI, CEOs, CTOs, & Investors aside for their hottest takes on the debates defining AI right now.

→ Listen on X, Spotify, YouTube, Apple

Open source vs closed models dominated the room, alongside agents rewriting how work gets done, the death of the keyboard and mouse, the economics of inference, and physical AI moving from data centers into cars, devices, and even orbit. Real numbers, real disagreements, and a few pieces of breaking news, straight from the people building it.

"The keyboard and mouse are slowly dying." - Max Cook, Coatue

"In the next 12 months, 90% of tokens will be going to open models." - Matan Grinberg, Factory

"Most of AI is just glorified data science." - Pim de Witte, General Intuition

"The SpaceX IPO is going to be viewed historically as the most undervalued IPO of all time." - Philip Johnston, Starcloud

"People don't even understand how transformational agentic AI is." - Mark Papermaster, AMD

Guest lineup:

• Max Cook, Sector Head, Coatue

• Matan Grinberg, CEO, Factory

• Rodrigo Liang, CEO, SambaNova

• Philip Johnston, CEO, Starcloud

• Mark Papermaster, CTO, AMD

• Pim de Witte, CEO, General Intuition

• Ramin Hasani, CEO, Liquid AI

• Robin Rombach, CEO, Black Forest Labs

• Chris Madden, CEO, Good Future Media

• Henri Delahaye, CEO, RAISE Summit

Special thank you to Brex, MongoDB, & AssemblyAI for helping make this RAISE AI Summit mini-series in Paris, France happen.

Subscribe now

Share

𝐓𝐈𝐌𝐄𝐒𝐓𝐀𝐌𝐏𝐒

00:00 RAISE in Paris
00:16 Max Cook (Coatue)
01:01 Keyboard and Mouse Dying
03:42 Future Interaction Models
06:32 Open Source vs Closed
12:20 Henri Delahaye Hosts the Night
13:21 Pim de Witte (General Intuition)
14:46 Mark Papermaster (AMD) on Agentic Workflows
16:38 Robin Rombach (Black Forest Labs)
20:31 Matan Grinberg (Factory): Droids in Europe
21:03 Open Models Token Shift
23:17 Philip Johnston (StarCloud): Space Bets
24:01 SpaceX IPO Bull Case
25:13 Chips for Space Compute
26:58 Chris Madden (Good Future Media): Clipping
28:23 Viral Clip Hook Secrets
32:03 Ramin Hasani (Liquid AI): Physical Models
34:50 Three-Axis Foundation Models
38:43 Rodrigo Liang (SambaNova): $1B Raise

Brought to you by:

Brex—The intelligent finance platform: cards, expenses, travel, bill pay, banking—wrapped into a high-performance stack. Built for scale. Trusted by OpenAI, Anthropic, Vercel, Granola, Deepgram, & Sourcery.. teams that move fast AF. visit → brex.com/sourcery

MongoDB–Millions of developers and more than 65,200+ customers across industries – including ~75% of the Fortune 100 – rely on MongoDB for their most important applications. With integrated capabilities for operational data, search, real-time analytics, & AI-powered data retrieval, MongoDB helps organizations everywhere move faster, innovate more efficiently, & simplify complex architectures. visit → mongodb.com/ai

AssemblyAI–Millions of developers use AssemblyAI to power their voice ai apps & features. One API gives you access to best-in-class speech-to-text, voice agent, & speech understanding models for both pre-recorded and real-time audio. Granola, ClickUp & HeyGen are scaling with AssemblyAI - get $50 of free credits today → AssemblyAI.com/sourcery

The 10 Hottest Takes, RAISE Day 1:

At the RAISE Summit in Paris, we pulled 10 of the biggest names in AI, CEOs, CTOs, Investors aside for their hottest takes on where AI is heading.

Open Source vs Closed Models

Matan Grinberg, CEO, Factory

Grinberg tracks open vs closed usage across Factory’s enterprise customers and reports open-model token share moving from under 1% in January to 10% in May.

“My hottest take is that in the next 12 months, 90% of tokens will be going to open models. Now, not all the tokens, and maybe not the most important tokens, but these open models are incredibly performant, incredibly cheap, incredibly fast, and there is gonna be an important use for them in the portfolio of models.”

“At the beginning of the year, we saw enterprises were using less than 1% of their tokens were going to open models. Then in around March, it crossed that 1% threshold, and then in May, it crossed the 10% threshold.”

“By the end of the year, at least in the enterprise, we’ll probably cross the 50% threshold by the end of this calendar year. Not every task you need Opus 4.8 or GPT 5.6 Ultra High, and many tasks day-to-day you do are very simple.”

Robin Rombach, CEO, Black Forest Labs

Rombach makes the case for accessibility as the deciding factor, not the open weights vs open source distinction.

“Open innovation is good for the world. And fear-mongering around AI models is only going to lead to them being more closed, and that’s ultimately going to lead to a slowdown of all AI progress.”

“It’s super important that AI is actually an accessible technology. It’s also gonna make it more safe in the end. I think that’s something that is a pretty fundamental principle.”

“Is it open weights? Is this open source? It doesn’t really matter in the end, but having accessible AI models is key.”

Max Cook, Sector Head, Coatue

Cook reframes the debate around workload type, & flags an unasked question about the frontier labs offering open source alternatives themselves.

“A lot of the discussion I see is sort of there’s either one is a winner and one is a loser, or the other’s a winner and the other’s a loser. I think that’s totally the wrong way to think about this.”

“Open source is good for production scale workflows, yet we’re so early in this AI journey that there’s not that many production grade workflows. And so if you are trying to discover a new use case, you want the smartest model you can possibly get, you use the frontier.”

“I don’t think anyone’s asking the question of what happens when Anthropic & OpenAI decide we can offer that as a product. It seems plausible that they would see this opportunity to take their older, cheaper, more efficient, smaller models & sell that as a product.”

Agents & the End of Apps

Max Cook, Sector Head, Coatue

Cook argues the shift from dozens of apps to a handful of agentic interfaces changes how people interact with compute.

“Hottest take might be that the keyboard and mouse are slowly dying. We talk about big tech waves. You went from mainframe to PCs to networking PCs to desktop internet, which is where you get a mouse and keyboard, to mobile internet, which is where the iPhone supremacy began.”

“We’re going from a world in which you have 60 to 80 apps to an agentic AI world where you’ve got a lot fewer than that, two, three, four apps, where everything else plugs in as an API or an MCP. And in that world, it would only make sense that we reinvent how we interact with compute. It’s not the keyboard and the mouse. It’s probably natural language.”

“We’re gonna look back in five years and say, of course we’re not using keyboards and mice anymore. We’ve come up with a better way to interact with intelligence.”

Agentic AI in the Enterprise

Mark Papermaster, CTO, AMD

Papermaster describes agentic workflows reshaping both enterprise work and AMD’s own chip design, with the CPU-to-GPU ratio moving toward parity.

“Hottest take is people don’t even understand how transformational agentic AI is. It’s just transforming how work gets done, and entire end-to-end workflows are being reimagined.”

“AMD, we got ahead of it. We’ve worked to getting our CPUs and GPUs ready for this, and now with these agentic workflows, you actually need both. In fact, the ratio of CPU to GPU is becoming one to one.”

“Every day I start with agents. They run my day, look at what’s coming up. But what’s really cool is how we do our chip design. We have thousands of sub-agents just doing incredibly complex chip design, speeding how fast we can get the next generation to market.”

Economics of Inference & Chips

Rodrigo Liang, CEO, SambaNova

Liang announced a $1 billion Series F at an $11 billion valuation led by General Atlantic, alongside the new SN50 chip for premium inference.

“We’re announcing the first close. Our series F is a $1 billion raise at 11 billion valuation. The round was led by General Atlantic, and significant investment from Seligman Ventures, from T. Rowe Price, and Capital Group.”

“The world of inference is running into an economics problem. And what we’re able to do with premium inference is drive the performance of the largest models up so you can actually deliver high-quality models, these trillion parameter models, at their full precision at incredible speeds.”

“When you’re running faster with the high accuracy models, you can charge more at a lower cost, allowing their service providers to generate more margins for their services.”

Efficiency & Physical AI

Ramin Hasani, CEO, Liquid AI

Hasani frames efficiency and deployment substrate as first-class design constraints, not afterthoughts. i.e. Liquid AI powers in-car intelligence for Mercedes-Benz.

“Efficiency is not an afterthought. Energy is not abundant. There’s not that many chips available in the world right now, & supply chains are getting disrupted on the memory side of things.”

“We’re bringing the cost of tokens to zero.

& we are just bringing the value of the AI for enterprises. We’re building foundation models that are so cheap that you can bring them on Raspberry Pis. So you can host them on any kind of device that is on the planet.”

“The third axis I would say is substrate. Where does this intelligence system go? AI is majorly getting hosted in data centers, but you could also bring intelligence on phones, on laptops, on airplanes, on cars.”

AI, Public Markets, & Space

Philip Johnston, CEO, Starcloud

Johnston builds orbital data centers and ties his bull case to SpaceX’s launch cost advantage.

“My hottest take right now is that the SpaceX IPO is gonna be viewed historically as the most undervalued IPO of all time. I think they will tear through 10 trillion within the next couple of years, and then have almost unlimited time for what they’re building.”

“They own what will be by far the most cost-effective launch vehicle, and that opens up every industry in space that will be possible beyond that. So that’s all of asteroid mining, lunar resource mining, all of the comms businesses that are gonna be built. Everything else is gonna have to go through SpaceX.”

“I’m not regulated by anybody, so I can tell people, just buy the SpaceX stock. I think it’s a great stock to buy.”

Data as the Foundation

Pim de Witte, CEO, General Intuition

De Witte trains models that predict a range of actions beyond text, and puts data quality ahead of model complexity.

“As humans, we can talk, or choose to output text as one of the actions we can take, whereas an LLM’s kind of the only thing they can do. So we train models that can predict lots of different actions, where text is just a subset, which is a much more general approach than LLMs.”

“Most of AI is just glorified data science.”

“Good models mostly are downstream from good data and simple things.”

Media & Distribution

Chris Madden, CEO, Good Future Media

Madden runs a clipping agency producing thousands of clips a month and puts the value on the opening seconds.

“Clipping is the new ad. Whatever the long form is, whatever message you have, whatever idea or thing you’re trying to sell, you need to get it into a short amount of time to be able to grab somebody’s attention.”

“It’s really the first three seconds. The three seconds need to grab the viewer in. If the spoken line isn’t catching you, you need to have a visual hook on the screen that will tell the viewer what is the payoff for them.”

“200 bucks a clip is pretty standard in our agency. Minimum, and then you wanna go clips daily, so a clip a day at least. If you can do two or three a day, that’s even better.”

The View from Europe

Henri Delahaye, CEO, RAISE Summit

Delahaye hosts the summit & frames the event around pairing AI with history.

“We are all in our tech bubble, token maxing on a daily basis, but it’s good to have a bit of a history in our sight.”

Subscribe now

Share

The material presented on Molly O’Shea’s website are my opinions only and are provided for informational purposes and should not be construed as investment advice. It is not a recommendation of, or an offer to sell or solicitation of an offer to buy, any particular security, strategy, or investment product. Any analysis or discussion of investments, sectors or the market generally are based on current information, including from public sources, that I consider reliable, but I do not represent that any research or the information provided is accurate or complete, and it should not be relied on as such. My views and opinions expressed in any website content are current at the time of publication and are subject to change. Past performance is not indicative of future results.

Paid Endorsement. Brokerage services by Open to the Public Investing Inc, member FINRA & SIPC. Advisory services by Public Advisors LLC, SEC-registered adviser. Crypto trading provided by Zero Hash LLC, licensed by the NYSDFS. Generated Assets is an interactive analysis tool by Public Advisors. Output is for informational purposes only and is not an investment recommendation or advice. See disclosures at public.com/disclosures/ga. Matched funds must remain in your account for at least 5 years. Match rate and other terms are subject to change at any time.

Premium Inference 101

Economics & Infrastructure Behind Running Trillion-Parameter Models

Rodrigo Liang is the CEO and Co-Founder of SambaNova. The company just announced a first close on a $1B round at an $11B valuation, led by General Atlantic with T. Rowe Price and Capital Group participating.

→ Listen on X, Spotify, YouTube, Apple

Rodrigo has spent 32 years building chips. We covered the 101 of semiconductors and data centers: what inference is, why it is a different problem than training, and why inference will require orders of magnitude more chips than training.

He walks through SambaNova's chip lineup from SN10 to the new SN50, why a 10 kilowatt air-cooled rack changes where AI can be deployed, and why running a trillion parameter model in a single rack matters for agents, latency, and edge.

We also covered the era of premium inference, how providers measure revenue per rack, coopetition and traffic routing across Nvidia and AMD, sovereign models, the move back to on-prem, and why tokenmaxxing is the wrong goal.

Special thank you to Brex, MongoDB, & AssemblyAI for helping make this RAISE AI Summit mini-series in Paris, France happen.

Subscribe now

Share

𝐓𝐈𝐌𝐄𝐒𝐓𝐀𝐌𝐏𝐒

(00:00) Rodrigo Liang , Co-Founder & CEO at SambaNova Systems
(00:59) SambaNova’s Series F: $1B raise at an $11 billion valuation
(03:00) The Inference problem nobody saw coming
(04:52) SambaNova's chip evolution
(07:19) Running a trillion-parameter model on a single rack
(11:00) Do $100 billion data centers actually make sense?
(14:14) What "premium inference" really means
(18:28) Speed is about to become AI's biggest price tag
(20:43) Starlink, edge computing, and AI reaching every corner of the planet
(24:27) Working alongside NVIDIA and rival chipmakers
(27:49) How customers actually measure inference performance
(32:07) The biggest bottlenecks in AI's global land grab
(35:12) Justifying the billion-dollar AI valuations
(37:53) Why SambaNova refuses to build its own cloud
(41:03) The "AI sovereignty" debate
(43:48) Data privacy fears are driving the return to on-prem AI
(47:55) How to actually get ROI out of AI spend
(51:16) The one question every business should be asking about AI
(56:09) The mentors and lessons behind a 32-year career in chips
(58:02) Unveiling SambaNova's newest chip, the SN50

Brought to you by:

Brex—The intelligent finance platform: cards, expenses, travel, bill pay, banking—wrapped into a high-performance stack. Built for scale. Trusted by OpenAI, Anthropic, Vercel, Granola, Deepgram, & Sourcery.. teams that move fast AF. visit → brex.com/sourcery

MongoDB–Millions of developers and more than 65,200+ customers across industries – including ~75% of the Fortune 100 – rely on MongoDB for their most important applications. With integrated capabilities for operational data, search, real-time analytics, & AI-powered data retrieval, MongoDB helps organizations everywhere move faster, innovate more efficiently, & simplify complex architectures. visit → mongodb.com/ai

AssemblyAI–Millions of developers use AssemblyAI to power their voice ai apps & features. One API gives you access to best-in-class speech-to-text, voice agent, & speech understanding models for both pre-recorded and real-time audio. Granola, ClickUp & HeyGen are scaling with AssemblyAI - get $50 of free credits today → AssemblyAI.com/sourcery

Premium Inference 101: The Economics & Infrastructure Behind Running Trillion Parameter Models

I sat down with Rodrigo Liang, co-founder and CEO of SambaNova, in Paris at the RAISE Summit, on the announcement of the first close of their $1B Series F at an $11B valuation.. just 5 months after their last round.

→ Listen on X, Spotify, YouTube, Apple

The $11B Valuation Round

SambaNova completed the first close of $1 billion in strategic financing as part of a Series F round, valuing the company at $11 billion post money, led by General Atlantic with significant investment from Seligman Ventures, T. Rowe Price Associates, & Capital Group. New & existing investors include A&E Investment, Assam Ventures, Battery Ventures, Cambium Capital, BlackRock, Intel Capital, Kabila Capital, QFO Capital, Qatar Investment Authority, Vista Equity Partners, & Volantis.

“I’ve been in this industry for 32 years, building high performance chips for a long time. I’ve never seen the interest in semiconductors higher.”

SambaNova raised a $676 million Series D led by SoftBank Vision Fund 2 in April 2021, past a $5 billion valuation. The company reportedly reached roughly $100 million in annual revenue in June 2025 against more than $1.13 billion raised. Intel was rumored to be trying to buy it for about $1.6 billion. The $350 million Series E closed in February 2026, bringing Intel in as strategic partner and investor. Five months later the company is at $11B. Rodrigo puts total funding at $2.5 billion.

Liang said the capital is going to secure the supply chain, which he called the key constraint on fulfilling orders over the next 12 months, with the SN50 shipping in the second half of 2026.

Why Inference Is a Different Problem Than Training

Training is the one-time cost of creating a model.

Inference is the cost of running it every time somebody sends it a prompt.

“There’s no point of training a model if you aren’t gonna inference it, i.e. if you’re not gonna use it, right? You don’t invent a search algorithm if you’re never gonna do search.”

SambaNova started in 2017 with the goal of lowering training cost, back when the workloads were image recognition & voice. Inference wasn’t a problem then because the only people running the models were the people who had just trained them.

“Now what you’re seeing at scale with Anthropic, OpenAI, & with Gemini, you’ve got millions & millions of people using it every day. And so now you have the problem that SambaNova was originally focused on, which is around efficiency.”

“How do you actually deploy at scale so the whole planet can use it without burning up the planet, without running out of data center space, without blowing up your infrastructure cost? Because at scale, the number of chips deployed for inferencing will be orders of magnitude greater than whatever you’re doing for training.”

Inference has moved from a lab expense to a product gross-margin line, with per-token unit price down roughly an order of magnitude over two years while consumption rose more than 100x. SemiAnalysis reports inference providers including Fireworks, Baseten, & Fal are seeing widening margins on hyper-growth revenue.

The Rack Is the Unit of Measure

“Instead of 130, 140 kilowatt rack of Nvidia GPU, we were outperforming them with a 10 kilowatt SN40 rack.. air-cooled. You didn’t need liquid cooling upgrades.”

By comparison, the Nvidia GB200 NVL72 draws roughly 120 kW nominal with 130-132 kW observed at full load, against a 7.6 kW industry-average rack per the Uptime Institute in 2025. Air cooling isn’t viable at that density, and rear-door heat exchangers designed for 30 to 40 kW racks can’t handle the load.

Most data centers can handle 30 kW racks, more modern facilities 60 kW, and 120-140 kW almost requires new facilities. A rack that runs on air at 10 kW can go into a building that already exists. A 130 kW rack needs a building designed around it.

The SN40L rack uses 16 chips in a standard 19-inch air-cooled form factor delivering 10.2 PFLOPs. Each socket is a 5nm TSMC dual-die design with 102 billion transistors, 640 BF16 TFLOPs, and 520MB of on-chip SRAM, using a three-tier memory system of SRAM, HBM, and off-package DRAM.

“We’re using standard Kubernetes, standard Red Hat Linux, standard Ethernet at the top for networking.. And so that allows people to go in to existing data centers, roll this thing in, pull out the old gear, & you’re up & running with new services, which otherwise might take you nine months to a year, maybe as long as 18 months, to build a gigawatt data center.”

The Minimum Quantum

“With SambaNova, that minimum quantum is down to one rack. Where if you have other service providers, you just run, say, a DeepSeek model, which is now one and a half trillion parameters, just to run that, the minimum for some of the other providers might be 10 to 20 racks.”

The minimum quantum is the smallest amount of hardware you have to buy before you can serve a single customer. If it takes 20 racks to load the model at all, that’s the cost of entry regardless of how many users you have.

16 SN40L RDUs make up a single rack running models such as DeepSeek R1 671B and Llama 4 Maverick. Fifth-generation SN50 RDUs scale to 256 chips across racks, running models up to 10 trillion parameters with context length up to 10 million tokens.

“We take the biggest models in the world & we run them in the original precision. We don’t quantize. Quantizing is, you know, you chop half the weights off. So we don’t chop the model down. We just run original precision, full precision, run it faster than anybody else.”

Quantization compresses a model so it fits on less hardware, at some cost to accuracy. For Llama 3.3 70B, the same weights span roughly $0.12 to $1.05 per million tokens, about 9x, depending only on who serves them, with quantized tiers at the cheap end. Two providers can list wildly different prices for what looks like the same model because they aren’t running the same thing.

What Premium Inference Means: 2 Axes

1. Model size, which today is a proxy for accuracy

“It’s not that people wanna spend the hundreds of millions of dollars to train. ... They’re fighting for that bit of accuracy.”

His example was code generation. If the output is trustworthy, you don’t spend engineering hours checking it, and that saved time is what the buyer is paying for.

2. Speed, & the argument for it is arithmetic

“If each of those 20 agents took two seconds, that’s 40 seconds, you’ve already given up on that prompt, right? So the response time by the end user’s expectation is, say, 1-2 seconds. Divide that by 20, it’s 0.1 seconds per.”

Two seconds is fine for one person talking to one model. In an agent workflow where 20 models call each other before anything comes back, 2 seconds each is 40 seconds of silence.

“When 5G showed up, nobody’s signing up for 2G..

It’s never been the case, ‘Let me pay more for the slow.’ Right? Or even, ‘Let me pay a little bit less for the slow.’ Most people over time are gonna say, ‘No, I want the fastest.’”

“It’s whether today we can offer at low enough cost that everybody can afford it.”

Forbes reported in June 2026 that OpenAI and Anthropic are on the brink of a token price war ahead of anticipated IPOs. Custom chips including TPUs, Trainium, Maia, and MTIA cut inference costs between 30% and 50% versus Nvidia GPUs, with TrendForce projecting custom silicon runs 40% of AI servers by 2030.

Open No Longer Means Cheap

Two weeks after this conversation, Moonshot announced Kimi K3, a 2.8 trillion parameter mixture-of-experts model with 16 of 896 experts active per token, which would make it the largest open model ever released, ahead of DeepSeek V4-Pro at 1.6 trillion. Weights are promised by July 27.

Jamin Ball of Altimeter ran the pricing in Clouded Judgement. K3’s blended price is $5.40 per million tokens, against $2 for GLM 5.2, roughly $9 for Opus 4.8, and $10 for GPT 5.5. His framing is that “open weights were a cost lever and the ‘license’ was the discount.” At 2.8T parameters that lever mostly disappears. The weights alone come to roughly 1.4TB, which means 10-plus H200s just to load the model before you touch the KV cache, and Moonshot’s own launch materials recommend supernode configurations of 64 or more accelerators.

Open models were cheap because they were small enough that anyone could serve them and compete each other’s margins down. K3 is the first one big enough that only a handful of operators can serve it at all.

Gavin Baker of Atreides added the second variable. Per Artificial Analysis, Kimi K3 is 50-70% more expensive to run than GPT 5.6 per task, because what you pay is cost per token multiplied by how many tokens the model burns to finish the job, & K3 burns a lot. Baker calls it a “token wastrel.”

His larger point is that a market with 2 to 3 dominant frontier labs at 90% inference margins turns those labs into monopsonies for power, data centers, & semiconductors, so anything that compresses margin at the model layer pushes margin dollars into every infrastructure layer beneath it. That’s his explanation for why Jensen Huang champions open source. The same logic applies to SambaNova.

The K3 numbers line up with both of Rodrigo’s axes. The open frontier went from 1.6T to 2.8T parameters in one release, so model size is still climbing. Moonshot recommending 64-plus accelerators is the minimum quantum problem, stated by the model maker rather than by a competitor. And if open models are no longer cheap to serve, then whoever runs them at full precision on the smallest footprint captures the difference. What we find out on July 27 is whether third-party servers can undercut Moonshot’s $5.40 at all, given that the spread on serving identical weights already runs about 9x across providers for smaller models.

An open model needs the same compute as a closed model of the same size and architecture. The license discounts the lab’s margin and nothing else. That’s good for whoever sells the infrastructure and bad for the assumption that open weights deflate inference spend.

The Demo, Nvidia for Prefill & SambaNova for Decode

At RAISE, SambaNova ran MiniMax M2.7 on a setup that split the work between two kinds of chips. One Nvidia H200 rack with four GPUs handled prefill, and one SambaRack SN50 with 16 RDU chips handled decode. As benchmarked by Artificial Analysis, decode speeds reached up to 850 tokens per second on short-context workloads and over 450 tokens per second on long-context workloads.

Every request has two phases. Prefill is the model reading everything you gave it, which is compute-heavy and highly parallel, so GPUs are good at it. Decode is the model writing the answer one token at a time, which is limited by memory bandwidth and gets worse as context grows. Sending each phase to the chip built for it is what disaggregated inference means. The demo ran on vLLM, so providers don’t have to change the serving software they already use.

Agents spend nearly all their time in decode, since planning, tool calls, validation, and revision are all generation. Long-context decode speed is what decides whether an agent feels useful or stuck, which is the same problem as the 40-second math above.

The COMPUTEX version of this, with Nvidia B200 for prefill and SN40 RDUs for decode, delivered 2x the inference speed of B200-only configurations as verified by Artificial Analysis, is running live at Vector Core Compute’s data center, and Together.ai is the first commercial customer. MiniMax is the open-weights coding model Rodrigo names twice in our conversation. Artificial Analysis’ public MiniMax M2.7 provider page shows SambaNova leading measured output speed among tracked API providers, at a higher blended price than lower-cost GPU-backed providers. Fastest, not cheapest.

Calculating ROI: Revenue Per Rack

“[Inference providers] purchase per rack, they operate per rack, so they want to generate revenue per rack.

The revenue’s generated per token.. If I put a rack of hardware, I’m just seeing how many tokens am I generating in a particular model, and that model has a price per token, right? Multiply that by 30 days per month, 24 hours per day, number of tokens per second, & you can figure how much money that rack is generating, & you look at how much it’s costing you to operate.”

“We’re very focused on making sure that when you deploy a rack of SambaNova, you generate great margins relative to the model. And you can change the model, different models have different pricing per token, but you’re still generating significant number of tokens per month so that you’re making profit on that rack.”

“For them to sustain themselves, as you know today, inference services, they’re not making enough margin. They’re generating lots of revenue, but you’re not generating enough margin.”

SambaNova sells providers on routing inference traffic to its racks rather than replacing what they already own. In a facility where 70% to 80% of racks are running inference, “Why would you run inference on those racks when you can run it at a fraction of the cost at a higher performance on SambaNova? And so route that traffic to SambaNova, frees up all these racks.. And so the economics starts getting much better because now without buying more hardware, they generate more revenue.” SambaNova’s SN50 target puts numbers on it, claiming that at 500 tokens per second per user, B300 plus SN50 lowers output cost-to-serve versus GPU-only decode, improving provider margin at the same API price.

“People forget, as much as Nvidia costs, it’s commodity. Because what you offer is the same as what your neighbor offers.. and your differentiation is, ‘I can save you a little bit of money because maybe I got a discount from Nvidia.’”

“It’s not gonna be 100 different chips. Right? It might be two or three. Maybe three or four. That’s as heterogeneous as AI infrastructure is gonna get.”

Ok so, Who Buys This?

SambaNova doesn’t build data centers and doesn’t run its own cloud. “Many of the chip companies have chosen to go build their own cloud and compete with the AWSs of the world. We have chosen not to do that.” The company sells to three customer types, sovereign clouds, neoclouds, and enterprises building private AI deployments.

• JPMorgan Chase will deploy SN40 and SN50 systems to run AI workloads on its own premises rather than through a third-party cloud provider. Rodrigo told CNBC that JPMorgan selected SambaNova to be the inference provider for the bank. The quote in the release comes from the CIO of Infrastructure Platforms, and the language is about testing speed and security, which reads as evaluation rather than deployment.

• SoftBank Corp. is the first deployment partner for the SN50, shipping in the second half of 2026. SoftBank Corp. already hosts SambaCloud for developers in the region, and by anchoring new clusters on SN50 it positions SambaNova as the inference backbone for its sovereign AI initiatives and future agentic services. Its affiliate SoftBank Vision Fund 2 led the 2021 Series D. Intel entered a planned multi-year strategic collaboration in February 2026 alongside the SN50 launch.

• Three sovereign agreements were announced in October 2025 with SCX in Australia, Infercom in Europe, and Argyll in the UK. Others include stc Group in Saudi Arabia, OVHcloud, over 30 enterprise customers including multiple US Department of Energy national laboratories, AWS Marketplace availability from May 2025, and an Accenture integrator partnership. Saudi Aramco & several Japanese firms are also named.

• Vista Equity and Cambium announced Vector Core Compute, an agentic neocloud built on SambaNova hardware, in June 2026, and both are investors in this round.

Most companies that want fast inference buy an API, not a rack. SambaNova reaches them only if the model they want runs on its hardware, which in practice means open weights, or if their provider routes to it underneath. Baseten resells SambaNova as an Nvidia alternative in multi-tenant environments, and Together.ai is the first commercial customer of the disaggregated architecture at VC2.

Sovereignty & the Return to On-Prem

“I was talking to a CIO recently of a big bank.. they were never gonna go to a cloud, and it’s like, ‘See, I knew all along.’ I was like, ‘Yeah, it’s cyclical.’ A 20-year cycle, wait long enough, it’ll come back.”

He puts the shift down to margin rather than security. “If you actually transfer all of those services, that differentiation to all using the same exact model that’s in the community, where does the differentiation come? And so what you find yourself is in a place of low margin for all these enterprises that historically had been able to enjoy much better margins.” He thinks companies realize this within two years.

On national models, “I don’t want my data trained into a model and have that model shipped worldwide. Can you imagine if your bank account information starts showing up in ChatGPT in some other place in the world without your permission?” SambaNova sells a full stack of chips, systems, software, and cloud directly to enterprises, sovereign programs, and regulated industries, which its chip-only competitors don’t.

Asked whether $50B and $100B data centers make sense, “I think you’re gonna have some data centers like that, because I think there’s still going to be large scale deployments.. And I think you’re gonna find that the world’s gonna be heterogeneous.” His argument is about where the big builds can’t go. “In terms of ultra-low latency in the big cities, you’re gonna have to find smaller quantums.”

What People Should Be Asking

“Just using AI doesn’t differentiate. How do you differentiate?”

“It’s kinda like early days of internet. If we say, ‘What is the ROI for email? Well, show me.’” On where that goes wrong, “If you aren’t paying attention, it becomes token maximizer, just like, ‘Hey, just spend.’”

On scale, “The hints that you’re seeing today with energy constraints, data center constraint, chip availability constraints, cost constraints, all of those things are only getting exacerbated.” CoreWeave crossed 1 GW of active power in the quarter and expanded contracted capacity past 3.5 GW, with capex raised to as much as $35 billion for 2026 citing infrastructure component inflation.

“Business at scale is gonna come with the highest highs and sometimes the lowest lows, and you’ve gotta fight through all of it. It’s never a straight line.” SambaNova was worth $5B in 2021, was reportedly the subject of a $1.6B acquisition conversation, and is worth $11B now. “You can’t control what the world wants at any given point in time. You can’t control what the economy does. You can’t even control what the politics do. What you can control is your conviction around what you’re building and being resilient and staying with it.”

→ Listen on X, Spotify, YouTube, Apple

Subscribe now

Share

The material presented on Molly O’Shea’s website are my opinions only and are provided for informational purposes and should not be construed as investment advice. It is not a recommendation of, or an offer to sell or solicitation of an offer to buy, any particular security, strategy, or investment product. Any analysis or discussion of investments, sectors or the market generally are based on current information, including from public sources, that I consider reliable, but I do not represent that any research or the information provided is accurate or complete, and it should not be relied on as such. My views and opinions expressed in any website content are current at the time of publication and are subject to change. Past performance is not indicative of future results.

Paid Endorsement. Brokerage services by Open to the Public Investing Inc, member FINRA & SIPC. Advisory services by Public Advisors LLC, SEC-registered adviser. Crypto trading provided by Zero Hash LLC, licensed by the NYSDFS. Generated Assets is an interactive analysis tool by Public Advisors. Output is for informational purposes only and is not an investment recommendation or advice. See disclosures at public.com/disclosures/ga. Matched funds must remain in your account for at least 5 years. Match rate and other terms are subject to change at any time.

WIRE HARNESSES

Senra Systems closes $65 million Series B, bringing total funding to over $112 million. The round was co-led by Lowercarbon Capital and Interlagos, with participation from General Catalyst, Sequoia Capital, Andreessen Horowitz, Founders Fund, Dylan Field, CIV, 8VC, The Friedkin Group, Jaws Estates Capital, Sozo Ventures, and Alumni Ventures.

→ Listen on X, Spotify, YouTube, Apple

Jordan Black, co-founder and CEO of Senra Systems, and Ken Venner, Senra's new Chief Technology & Product Officer, join Sourcery for a full breakdown plus a walkthrough of the new 80,000 sq ft Cypress factory.

Jordan Black co-founded Senra after managing the R&D team for electronics at SpaceX, where wire harnessing was the bane of everyone's existence and 95% of it was outsourced for Falcon and Dragon. He started the company on his apartment carpet with co-founder Ben Shanahan, who helped write the application software that scaled Starlink production from day one.

Ken Venner joins as CTPO after a legendary career in manufacturing. He scaled Broadcom from 1,000 people and $400M in revenue to 10,000 employees and $8.6B across 52 acquisitions, then served as CIO of SpaceX, where he built Warp Drive, the custom operating system that helped scale production from 1 booster a year to 40.

Wire harnesses are the nervous system behind every advanced platform, from rockets and satellites to cars, missiles, and generators. They can account for 25% of a vehicle's cost, and they are still built almost entirely by hand. Senra is digitizing the process with Amp, its proprietary manufacturing operating system, and training a new workforce in weeks instead of years.

We cover the funding environment, the Redondo vs Cypress factory split, the reindustrialization macro, vertical integration, the Ken Venner hire, and the plan for Factory 3.

Subscribe now

Share

𝐓𝐈𝐌𝐄𝐒𝐓𝐀𝐌𝐏𝐒

(00:00) Jordan Black, Co-Founder & CEO & Ken Venner, CTPO at Senra Systems
(00:57) Senra raises a $65M Series B
(01:36) What Senra Systems actually builds
(05:18) Why they built two very different factories
(07:00) The reindustrialization of America
(09:09) Commercial vs. Government: who's really buying?
(09:52) The origin story
(17:35) Turning a 2-year training pipeline into 4 weeks
(25:06) Bringing on SpaceX legend Ken Venner
(29:23) Why Jordan became obsessed with wire harnessing
(31:46) The investors betting big on wire harnesses
(34:16) The real story behind the name "Senra"
(36:49) Will the SpaceX IPO fuel Senra's growth?
(39:32) Ken Venner on why he joined Senra
(41:40) Scaling lessons from Broadcom and SpaceX
(43:21) Why Ken left Broadcom for SpaceX
(44:40) Life as SpaceX's CIO
(46:11) The secret to Elon Musk's management style
(47:15) What it's really like working with Gwynne Shotwell
(51:28) Scaling with AI vs. pure headcount
(55:04) Factory tour with Jordan

Brought to you by:

Brex—The intelligent finance platform: cards, expenses, travel, bill pay, banking—wrapped into a high-performance stack. Built for scale. Trusted by OpenAI, Anthropic, Vercel, Granola, Deepgram, & Sourcery.. teams that move fast AF. visit → brex.com/sourcery

Turing—Turing partners with frontier AI labs to improve model capabilities in coding, reasoning, tool use, & multimodality, as well as with Fortune 500 enterprises to build & deploy end-to-end agentic AI systems in mission-critical workflows Visit: turing.com/sourcery

VCX—VCX is the public ticker for private tech, allowing investors of all sizes to invest in venture capital. View The Portfolio at GetVCX.com

Deel—Deel is the global people platform that helps startups hire, manage, pay, and equip anyone, anywhere. Trusted by more than 35,000 fast-growing companies, Deel is the people platform that just works, so teams can scale without the chaos. Visit: deel.com/sourcery

Public-–Investing platform Public just launched Generated Assets, which lets you turn any idea into an investable index with AI. With Generated Assets, you can build, backtest, refine, and invest in any thesis with AI. Gone are the days of one-size-fits-all ETFs. Try it today: public.com/sourcery

SpaceX Alumni Raise $65M for Senra Systems

The $65M Round

Senra Systems has closed a $65 million Series B co-led by Lowercarbon Capital and Interlagos. It brings Senra’s total funding to over $112 million. The existing cap table came along for the ride, including Sequoia, Founders Fund, General Catalyst, Andreessen Horowitz, 8VC, Dylan Field, CIV, The Friedkin Group, Jaws Estates Capital, Sozo Ventures, and Alumni Ventures.

→ Listen on X, Spotify, YouTube, Apple

I sat down with co-founder and CEO Jordan Black at the company’s new Cypress facility to talk through the raise, and the framing Jordan keeps coming back to is that this is one of the least glamorous and most consequential problems in hardware. As Jordan put it in the announcement, “Wire harnesses are the nervous system behind every advanced platform, yet they’re still built on PDFs, spreadsheets and tribal knowledge. It’s a 100% manual assembly process.”

Lowercarbon is Chris Sacca’s firm, best known for climate and now pushing hard into industrials and advanced manufacturing, with well over $2 billion in assets under management. The firm was founded in 2018 by Chris Sacca, formerly of Lowercase Capital, and manages over $2 billion across clean energy, materials, and advanced manufacturing.

Lowercarbon GP Caie Kelley “Wire harnesses sit behind everything that turns on, & they’re still built by hand. Senra automates production & trains the workforce to run it, which turns a chokepoint into capacity the country can build on.”

Interlagos is a Los Angeles deep tech fund founded in 2024 by SpaceX alumni, and it has quietly become one of the most active checkbooks in the El Segundo hard tech scene. Interlagos has backed companies including Apex, Castelion, Neros, Impulse Space, and Cowboy Space across aerospace and defense.

What a Wire Harness Actually Is

If you have never thought about a wire harness, don’t fret. “It’s like the nervous system of anything that requires electricity.. it’s what sends power and data in your rockets, in your cars, in your generators, in your electric planes and everything.” It is the bundle that moves power and signal through a machine, and almost none of it is standardized. Every harness is custom, flexible, and built to fit one specific platform.

The scale is the surprising part. Jordan notes there are hundreds of miles of harnessing in a single plane, miles in a single car, and that harnessing can run up to 25% of a vehicle’s cost at any given moment. And it is still, in Jordan’s words, essentially “arts and crafts.” The wire is flexible, the routing is bespoke, and the entire thing is assembled by hand.

The global wire harness market is large and growing steadily. The global wire harness market was valued at roughly $95.9 billion in 2025 and is estimated at about $100.5 billion in 2026, growing at more than 5.3% annually. The aerospace and defense slice is a specialized, higher-spec segment within that, and estimates for it vary widely by definition, but the direction is consistent. Aerospace wiring harness demand is being driven by the shift to more electric aircraft, next-generation avionics, and military modernization programs, with North America the dominant region.

The catch is who makes them. The incumbent landscape is a mix of large primes and a long tail of small suppliers, and the work is copper-heavy, labor-intensive, and hard to automate. Copper dominates wire harness assemblies at roughly 70% of material share, and aerospace and defense applications, while a smaller slice of volume, demand the highest reliability, temperature resistance, and EMI shielding. That combination, high criticality and low automation, is the gap Senra is walking into.

The Bottleneck

Demand is going up and to the right across rockets, missiles, cars, and generators, meanwhile supply is shrinking. “Demand’s going up, but the supply chain is drastically decreasing because it’s a skilled assembly workforce that’s doing this for 20 to 50 years, all by hand, they have all the tribal knowledge, they are the machine at the end of the day.”

The workforce math is genuinely alarming. Jordan pegs the average wire harness technician at 48 to 55 years old, and the knowledge lives in their hands, not in any documentation. “If it’s Bob and Judy building in the corner and they leave, we don’t build more wire harnesses for this product line or this company,” Jordan said. Meanwhile the roughly 800 mom-and-pop shops scattered across the country are not growing. They are being absorbed by private equity or shutting down, and every closure is a single point of failure for some fighter jet or missile program.

As geopolitical competition intensifies and defense production requirements climb, the defense industrial base’s limited capacity threatens program schedules and readiness, which is the exact language the primes and the government now use. Most of Senra’s business today is defense, because that is what Jordan calls the hyper-growing part of the market, with commercial as the expansion path.

The demand pull is coming from the fast-scaling players. “We’re seeing a lot of the demand coming from these scaling companies like the Andurils of the world and the SpaceXs of the world that are moving very rapidly into their production phase.” These are companies that need supplier agility, not the multi-year lead times of a legacy shop, and that is the wedge.

From a SpaceX Carpet to Cypress

The origin story is quintessential El Segundo. Jordan managed the R&D team for electronics at SpaceX “Wire harnessing was the bane of our existence, and we outsourced 95% of it for Falcon and Dragon back in the day.” The process, in Jordan’s telling, had not fundamentally changed in decades. “Everyone hasn’t changed the process since the Cold War era.”

What tipped Jordan over was a trip to sort out what’s actually going on. He toured wire harness factories across Mexico, Puerto Rico, and Costa Rica, and found the same thing everywhere: wooden tables, Cold War era methods, and a fragmented market. Then Jordan tried to build a harness himself on a high-rate automotive line and couldn’t. “If someone like myself, who has studied this process and has really gotten into depth with it, can’t even go build a harness, then something’s fundamentally broken.” The problem wasn’t the people. It was the total absence of tools & scalable systems.

So Jordan quit, co-founded Senra with fellow SpaceX engineer Ben Shanahan, who had helped write the application software that scaled Starlink production, and started building harnesses on his apartment carpet. Two weeks later they had their first customer, the pizza automation company Stellar Pizza. Jordan’s early pitch to El Segundo startups was refreshingly unglamorous: whatever price, whatever lead time, just let me get my foot in the door.

The fundraising environment then was a different planet. Jordan flew around the country with a wire stripping machine the size of a printer and a 600-page book on how to build a harness, doing a live before-and-after.

• Old process: manual tools and mountains of documentation.

• New process: a configured machine and one button.

The Software Layer: Amp and Vertical Integration

The reason this is a software story and not a labor-arbitrage story is a platform called Amp. It is Senra’s proprietary operating system, and it runs the whole flow from quoting a harness to shipping it. Jordan draws the direct line to SpaceX: “Similar to the software system that SpaceX had, Warp Drive, that power the entire ecosystem of manufacturing, we essentially are doing the same thing.” Amp is what lets a bay reconfigure from one customer’s harness to a completely different one in hours instead of weeks.

The data play underneath it is the more ambitious bet. Senra is putting camera and vision systems across the factory to capture what a harness build actually looks like in the digital world, something Jordan says does not really exist today. The first live application is quality inspection. On the tour Jordan showed me a crimp inspection station where a 2D vision system flags a bad crimp instead of relying on a 20-year veteran’s eye. “Instead of visually inspecting every single time, I hit capture, and then it actually will tell me if it failed or not.” Every good and bad harness feeds a growing library that trains the model.

The Workforce Play

The industry’s standard path to a skilled technician is roughly 2 years of shadowing a veteran and hoping it sticks. Jordan’s analogy is sharp: “Imagine there’s no culinary school, and there’s no recipes, and you just have to look at the menu and go figure out how to make any of the things on there.” Senra compresses that to 4 weeks.

They can do it because of a credential worth flagging. According to Jordan, Senra runs the only US Department of Labor certified training program for wire harnessing. That is a specific and checkable claim, and if it holds, it is a real regulatory moat, not just a marketing line. Worth verifying against the DOL registered apprenticeship database before we put it in the newsletter as fact.

The hiring philosophy follows from the software. Because Amp delivers the recipe, Senra can hire on attitude and aptitude rather than decades of experience. “We hire based on attitude and aptitude of, can you use your hands? Can you read instructions?” The floor skews young on purpose. Most of the technicians had never seen a harness a few months before, which inverts the aging-workforce problem the rest of the industry is stuck with.

The training and the operating system are two halves of one system. Jordan is blunt that neither works alone. Hand someone a box of wires with no recipe and they flounder for weeks, months or years. Give them Amp with no training and the system has no operator. The combination is what lets Senra treat a scarce, tribal skill as something it can manufacture on demand, which is the whole ballgame for scaling a manual industry.

The Ken Venner Hire

The clearest signal of Senra’s ambition is who just joined. Ken Venner, the former SpaceX CIO, came on as Chief Technology and Product Officer, and his resume is a scaling resume. “In the 11 years I was there, grew to 10,000 employees and 8.6 billion in revenue, did 52 mergers and acquisitions,” Ken said of Broadcom, which he joined at 1,000 people and $400 million in revenue. Then at SpaceX Ken built Warp Drive, the custom platform that helped take the company from about one booster a year to, in his words, “40 boosters a year & 9,000 employees across multiple locations.”

Ken’s read on why he joined maps exactly onto the thesis. “Cable harnesses is such a large, fragmented, unautomated business, and yet it’s the backbone to everything that’s being built today.”

His playbook for scale? It’s deceptively simple “Get it defined, get it working consistently and reliably in such a way that you can now replicate it,” then replicate and improve across the organization. On Elon’s method, Ken distilled it to first principles and a sequencing rule, “remove before automate,” plus the responsible-engineer accountability model he says he learned there.

The most forward-looking thing Ken said is about leverage. He ran a team of about 175 building core platforms at SpaceX. At Senra he has 6. “I think I’ll be able to do exactly what I did at SpaceX with only 6 because AI drives most of what we’re doing.”

There is a nice throughline on scaling discipline that connects Ken to the factory strategy. Ken is candid that a model is not proven until it repeats. “Until you’ve built it three times, you don’t know if what you’ve built will replicate itself.” When I mentioned Tom Mueller telling us it takes three versions of an engine to get it right, Ken didn’t blink. Redondo was version two, Cypress is the test of true scale, and Factory 3 is where the copy-and-paste model either proves out or doesn’t.

Inside Factory 2, & What Comes Next

The Cypress facility is where the raise becomes concrete. It adds roughly 80,000 square feet, expands Senra’s production footprint by 5X, and is built to take the company from about 1,000 harnesses a month to 10,000 by next year. On the day I visited it was only about 25% built out, with active construction behind us, and Jordan’s plan is to hit close to full capacity by year end, staffing toward 150 technicians. Redondo is the prototype and new-product-introduction site. Cypress is prototype-to-production and scale.

Each bay is a different customer and a different harness, and Jordan’s point is that the people building look like any manual line until you notice what is missing. “Most people are building everything with their hands, but the biggest secret sauce is that they’re not looking at a drawing or trying to find a tooling. They’re incredibly efficient because the operating system is giving them that perfect recipe and that silver platter instructions every single time.” One bay was running 4,000-plus units a week for a single defense customer. Another table held a 120-foot harness for a defense system Senra was called in to build in two months because it was blocking a multi-billion dollar program.

As for what is next, the answer is more factories closer to customers.

Ken Venner: “We are building the infrastructure, systems and workforce needed to support the future of American manufacturing and the industrial base.”

Jordan, characteristically, put it plainest. “The world needs more wire harnesses, and I wanna build every one of them.”

→ Listen on X, Spotify, YouTube, Apple

Subscribe now

Share

The material presented on Molly O’Shea’s website are my opinions only and are provided for informational purposes and should not be construed as investment advice. It is not a recommendation of, or an offer to sell or solicitation of an offer to buy, any particular security, strategy, or investment product. Any analysis or discussion of investments, sectors or the market generally are based on current information, including from public sources, that I consider reliable, but I do not represent that any research or the information provided is accurate or complete, and it should not be relied on as such. My views and opinions expressed in any website content are current at the time of publication and are subject to change. Past performance is not indicative of future results.

Paid Endorsement. Brokerage services by Open to the Public Investing Inc, member FINRA & SIPC. Advisory services by Public Advisors LLC, SEC-registered adviser. Crypto trading provided by Zero Hash LLC, licensed by the NYSDFS. Generated Assets is an interactive analysis tool by Public Advisors. Output is for informational purposes only and is not an investment recommendation or advice. See disclosures at public.com/disclosures/ga. Matched funds must remain in your account for at least 5 years. Match rate and other terms are subject to change at any time.

AI's "Drug Pushers" & Sovereign AI

Andrew Feldman, Co-Founder & CEO of Cerebras Systems, joins Molly O'Shea at the RAISE Summit in Paris to talk all things chips, inference, AI deals, & IPO.

→ Listen on X, Spotify, YouTube, Apple

Recorded 2 months after Cerebras went public at a $56B valuation and popped to roughly $95B on its first day, this conversation covers the $20B+ OpenAI deal, why inference is the new battleground, the state of the AI data center build-out, and Feldman's take on the circular deals reshaping the industry.

We also get into what it means to create 1,000 millionaires, how co-design between hardware and software actually works, data centers in space, and why Feldman thinks AI could mean the next generation never knows anyone who dies of cancer.

Cerebras (Nasdaq: CBRS) builds the WSE-3, a single wafer-scale chip with 4 trillion transistors and 900,000 cores, and the CS-3 system it powers.

Special thank you to Brex, MongoDB, & AssemblyAI for helping make this RAISE AI Summit mini-series in Paris, France happen.

Subscribe now

Share

𝐓𝐈𝐌𝐄𝐒𝐓𝐀𝐌𝐏𝐒

(00:00) Andrew Feldman, Co-Founder & CEO at Cerebras Systems
(00:49) Why hardware suddenly became the coolest industry in tech
(01:51) What changed at Raise AI Summit
(03:01) Inside the $20 billion Cerebras - OpenAI deal
(05:50) What actually changes two months after an IPO
(06:32) Turning 1,000 employees into millionaires
(07:52) Staying sane during an AI gold rush
(10:44) Life after the IPO plateau
(11:45) The truth about the global data center shortage
(12:56) Why data centers are borrowing jet engines for power
(14:44) Are data centers really headed to space?
(15:37) The shift to designing chips and software together
(17:39) The biggest misconception about co-designing chips and software
(18:31) Inside SpaceX's multi-billion dollar AI deals
(20:06) NVIDIA's playbook for locking out competitors
(20:54) The hidden cost behind free tokens
(22:52) Andrew's response to Karp's sovereign AI thesis
(24:18) AI's biggest win might be curing cancer
(26:50) Peptides and biohacking
(28:00) How AI could finally fix the broken education problem
(29:46) The mentors who shaped Andrew Feldman's career

Brought to you by:

Brex—The intelligent finance platform: cards, expenses, travel, bill pay, banking—wrapped into a high-performance stack. Built for scale. Trusted by OpenAI, Anthropic, Vercel, Granola, Deepgram, & Sourcery.. teams that move fast AF. visit → brex.com/sourcery

MongoDB–Millions of developers and more than 65,200+ customers across industries – including ~75% of the Fortune 100 – rely on MongoDB for their most important applications. With integrated capabilities for operational data, search, real-time analytics, & AI-powered data retrieval, MongoDB helps organizations everywhere move faster, innovate more efficiently, & simplify complex architectures. visit → mongodb.com/ai

AssemblyAI–Millions of developers use AssemblyAI to power their voice ai apps & features. One API gives you access to best-in-class speech-to-text, voice agent, & speech understanding models for both pre-recorded and real-time audio. Granola, ClickUp & HeyGen are scaling with AssemblyAI - get $50 of free credits today → AssemblyAI.com/sourcery

Andrew Feldman on the $20B Bet, Creating 1,000 Millionaires, & What’s Actually Happening with AI Deals

Andrew Feldman, Co-Founder & CEO of Cerebras, joined Sourcery at the Raise Summit in Paris, 2 months after the company’s IPO. He covered the $20B OpenAI deal, why inference has become the center of gravity in AI compute, and why he believes the data center build-out is already behind.

He also weighed in on the circular deals reshaping the market, the difficulty of hardware-software co-design, and the case for AI in medicine and education.

→ Listen on X, Spotify, YouTube, Apple

Corporate Adulthood

Cerebras priced its IPO at $185 for a $56B valuation and opened at $350, one of the largest US tech IPOs in years. A first-day pop briefly carried its market cap to roughly $95B before it settled, and the stock now trades around a $60B valuation following a Q1 report that pressured the stock on a gross-margin forecast Feldman later called misunderstood on CNBC.

Feldman does not frame the IPO as a finish line. “Getting to an IPO is not the end of a journey. It’s sort of a plateau. It’s the arrival at corporate adulthood. It is the achieving one plateau so that you can climb others, & our opportunity’s gotten bigger.”

The main post-IPO change he cited was inbound volume, aka people wanting things from him. “If you get 80 emails a day of different people asking you for something, that’s not work. That’s just this range of people who want something of one form or another. To meet you, your time, for you to present, for you to donate to their cause.”

He also addressed the physical chip-on-the-shoulder image he posted after the IPO, and why he wanted to keep that tone as a public-company CEO. “People assume that as a CEO of any size or an entrepreneur that it’s always peaches & cream, & it’s just not the case. There’s enormous amount of hard work. There’s sacrifice that you make & that your family makes. They see you less, & it’s not for a little bit. It’s for years.”

The $20B Deal & Why Inference Won

The OpenAI agreement, announced in January, is the largest commercial driver behind Cerebras. “We announced in January a huge partnership, one of the biggest deals done in Silicon Valley history. We’ll be doing more than $20 billion of hardware for them over the next several years.” Public reporting puts it at 750MW of inference compute over 3 years, accounting for the majority of a backlog near $24.6B.

The deal reflects a broader shift in where value is moving, from training the models to running them. Feldman traced the logic from creation to consumption, and landed on speed as the differentiator. “We make AI with training, & we use AI with inference. And so as the models and as the AI we made becomes useful, everybody wants to use it, & inference is the mechanism through which we use it. And so now we have smart AI, & people wanna use it. When they wanna use it, they want it to be fast, & that’s where we come in, & where the fast is not by a little bit but by 20x.”

The underlying product is the WSE-3, a single wafer-scale chip with 4 trillion transistors and 900,000 cores, packaged into the CS-3 system, roughly the size of a mini-fridge. On public benchmarks the chip has reached about 2,500 tokens per second per user on Llama 4 Maverick, more than double Nvidia’s DGX B200 on the same model. His 20x figure is a broader comparison against typical GPU cloud throughput rather than the head-to-head B200 number.

We Are Behind

Against the common view that AI infrastructure is racing ahead of need, Feldman argued the opposite. Asked whether the industry is building for future demand or trying to catch up, he was direct. “We are behind.”

The reason is that physical infrastructure moves far more slowly than AI demand arrived. “Data centers had historically moved at the speed of real estate. Somebody would decide to build a building, and two years later, after permits, they poured concrete, & the building would go up. Two years ago, nobody cared about AI. The AI explosion’s happened so quickly. We’ve outstripped the demand for compute, for memory, and where do these things go? They go to these buildings, & we haven’t made them fast enough.” The binding constraint, in his telling, is capacity.

Cerebras is moving on both fronts. On July 9 the company announced a roughly 7x expansion of CS-3 production with Flex in Milpitas, alongside a plan for 200MW of AI compute across Europe by late 2027, with initial sites in France and the Nordics supporting OpenAI workloads. Feldman noted the scramble is reshaping the data center itself, with operators reaching for fuel cells and jet engines to power parts of the stack that had gone unchanged for about 20 years.

The same demand pressure is driving talk of moving compute off the planet entirely. Feldman sees a genuine fit for wafer-scale in orbit, since a single large chip sidesteps the problem of getting many small ones to communicate, but he was measured on the timeline. “I don’t think we’re in danger in the near term of having a data center in space. I think it’s more than five years away.”

AI “Drug Pushers”

Feldman was direct on the circular deals reshaping the market, framing his concern around their purpose. “I’m concerned that they are ways frequently for Nvidia to exercise market strength. They’re ways for Nvidia to use their balance sheet to limit competition. If they invest in a Neo cloud, the Neo cloud is less likely to use a non-Nvidia chip. If they invest in a model builder, there’s pressure not to use other people’s chips. That’s what I’m more worried about.”

He carried the same logic to the free compute credits being handed to startups, & did not soften the analogy. “I think these are drug pushers. Here, little girl, try a little bit. Just a little bit.” His solve for founders is to take the offer without becoming reliant on it, spreading demand across suppliers including AMD & Cerebras. “They should take it and then never be dependent. That never ends well.”

Even from inside the industry, he said, the true shape of many headline deals is hard to read, with some carrying real commitments and others little more than announcements. He offered the GPU-leasing market as an example of how the current dynamic emerged, tracing it back to X’s spare capacity. “X had available capacity because their Grok model wasn’t used very much. So they had these GPUs that were sitting around, & so they leased a whole block of them to Anthropic. And they looked up & said, ‘Whoa, that’s a pretty good idea.’”

That skepticism connects to the sovereign AI debate, recently pushed by Alex Karp on CNBC, over whether companies and nations should own the full stack. Feldman’s version is narrower, built around optionality rather than ownership. “You shouldn’t be dependent on Nvidia. You shouldn’t be dependent on one model maker. What you’d like to be as a nation or as a big company is you like to have choices.” The exception is proprietary data. “If you have unique data, be sure you don’t give that away, & you get the credit for your data, & it doesn’t help make somebody else’s model better.”

Co-Design & the Hard Part

If there is a durable edge in this cycle beyond raw speed, Feldman locates it in co-design, developing hardware and software together rather than separated by an operating system. He framed it as a break from how the industry historically worked.

"Historically you made chips & you ran software on them. AI has gotten so large & speed is so important that what they're doing is they're thinking about the design together. What changes could we make in software that would advantage the hardware, or as we're designing the hardware, what changes could we make that would make the software easier to run?"

He pointed to Cerebras's visibility into OpenAI's roadmap, and Google's development of the TPU alongside its Gemini and DeepMind teams, as the clearest examples.

The catch is that co-design is far harder than it sounds, and the assumption that it just takes goodwill misses the real friction. “The misconception is that it’s easy and all you need to do is get in a room. The software guys think one way. The hardware guys think a slightly different way. Anything you do to make it easier to write the software makes it harder to do the hardware. And these are really hard trade-offs.”

The product itself remains a strikingly physical object. Feldman described a chip that dwarfs anything else on the market yet still slots into standard infrastructure. “While we build this super big chip, the chip that’s like 58 times larger than any other chip, it goes in a metal enclosure that’s about the size of a fridge for a dorm room. And we put a couple of those in a standard rack.” Independent write-ups put the die closer to 50x a conventional GPU depending on the comparison, so 58x reflects his and Cerebras’s own framing.

The Case for AI

Asked about the broader upside of AI, Feldman pointed first to medicine. “We have a chance for our children or the next generation not only to not die from cancer, but to not know anybody who died from cancer. I think that is a real, achievable goal in 25 years. When we think about what AI can do, writing better code is cool, and there’s a huge market for that. But I think what it can do to better humanity is rid us of the number one killer of adults.”

He extended the point to self-driving, arguing machines already outperform human drivers, and to drug discovery, citing Brian Armstrong’s New Limit and the broader wave of longevity and GLP-1 work. On AI risk, he argued critics tend to weigh only one side. “The problem with the doomers is they’re only looking at one side of the ledger. To look at this with clear eyes, you gotta look at both sides of the ledger. You gotta say, ‘Look, we’re gonna use a lot of power.’ It’s true. And AI has some real risks. It’s true. And on the other side of the ledger, here’s some things it can do really differently.”

He used education as his central example. “We’ve known for 2,000 years the right way to educate children, and we never do it. Ever. We knew that the right way to educate Alexander the Great was to have a tutor, and the smartest tutor, Aristotle was, and that you teach each child differently. With AI, we can do that. We can get you a tutor that’s right for you, and what’s more, the tutor can be running in the background saying, ‘Look, 3% of students make this type of error, and the best way to teach them to overcome this weakness is with this approach.’”

Creating 1,000 Millionaires & Founder Lessons

Feldman has now taken 2 companies to major liquidity events, and he measures them partly by the wealth they created for the people who built them. His last company minted roughly 100 millionaires. Cerebras, at IPO, produced close to 1,000, a figure he clarified refers to current and former employees, not investors.

He was careful to separate that from chasing money as a goal. The people he wants to work with, he said, build hard things whether or not they are in fashion. “For the type of people I love working with, they like building stuff, & they like building hard stuff when it paid a little, when it was out of fashion. When hardware was uncool, they were still building hardware, because that’s what they like to build.”

What makes the wealth creation meaningful to him is the asymmetry of what employees put in versus what investors do. Investors are diversified across many bets; an early employee is not.

“When someone bets five or seven years of their career, & a career is 30 years, they’re betting a sixth of their professional career. And when you get to deliver for them, that’s a great feeling & one I’m proud of every day.”

Asked who helped him get there, Feldman credited venture capitalist Pierre Lamond and former Veritas CEO Mark Leslie, both for their standards. “There was a venture capitalist named Pierre Lamond, and he’s now in his 90s. He invested in us at Cerebras when he was the ripe age of 84. He’s forgotten more about making chips than I’ll ever know.” Leslie, who ran Veritas, was the other name, one of the “wise people” he said taught him to demand a great deal of himself and others.

Day to day, he pointed to his 5 co-founders, all carried over from his previous company, with a wry note that “five is, in almost every case, too many founders.” His summary of why it works came back to the same idea that ran through the rest of the conversation, that the work itself is the point.

“Chasing money is not the path to money. The path to happiness is working on projects you like, with colleagues that are interesting, for people with integrity. And if you do that, the money will come.”

→ Listen on X, Spotify, YouTube, Apple

Subscribe now

Share

The material presented on Molly O’Shea’s website are my opinions only and are provided for informational purposes and should not be construed as investment advice. It is not a recommendation of, or an offer to sell or solicitation of an offer to buy, any particular security, strategy, or investment product. Any analysis or discussion of investments, sectors or the market generally are based on current information, including from public sources, that I consider reliable, but I do not represent that any research or the information provided is accurate or complete, and it should not be relied on as such. My views and opinions expressed in any website content are current at the time of publication and are subject to change. Past performance is not indicative of future results.

Paid Endorsement. Brokerage services by Open to the Public Investing Inc, member FINRA & SIPC. Advisory services by Public Advisors LLC, SEC-registered adviser. Crypto trading provided by Zero Hash LLC, licensed by the NYSDFS. Generated Assets is an interactive analysis tool by Public Advisors. Output is for informational purposes only and is not an investment recommendation or advice. See disclosures at public.com/disclosures/ga. Matched funds must remain in your account for at least 5 years. Match rate and other terms are subject to change at any time.

Brex—The intelligent finance platform: cards, expenses, travel, bill pay, banking. Built for scale. Trusted by OpenAI, Anthropic, Vercel, Granola, Deepgram, & Sourcery.. teams that move fast AF. visit → brex.com/sourcery

The Biggest Ideas from Config 2026

More than 10,000 designers, engineers, founders, investors, and creators gathered in San Francisco for Figma’s annual Config conference. Beyond the keynote, I sat down with leaders from Figma, OpenAI, Boston Dynamics, MIT Media Lab, IVP, and more to talk about where design, software, robotics, and AI are headed.

Watch on YouTube

Below are some of the biggest ideas from those conversations

Guests:

• Dylan Field, CEO, Figma
• Ian Silber, Head of Product Design, OpenAI
• Brian Ringley, Distinguished Product Manager & Human-Robot Interaction Designer, Boston Dynamics
• Niko Klein, Product Manager, Figma
• Holly Herndon, AI Artist
• Zach Lieberman, MIT Media Lab
• Shreyas Garg, Partner, IVP

We discuss:

• Why AI shouldn't do your thinking
• The future of design in an AI-first world
• Why robotics is closer than most people realize
• Code as a creative material
• Open source vs. frontier AI models
• The next generation of AI products
• What every designer and engineer should learn now

Subscribe now

Share

Dylan Field, CEO, Figma

Dylan Field co-founded Figma in 2012 and has built it into the leading collaborative design platform. Minutes after stepping off the Config keynote stage, he shared why independent thinking may become the most valuable skill in the AI era.

“It is not enough to wear a thinking cap. You have to actually think. And there’s a world where a lot of people are getting into this space where they’re letting the AI think for them, and they’re tricking themselves into believing that they’re thinking themselves.”

“Especially when the AI is sycophantic and is telling you you’re so great and such a genius, but then kinda giving you what it wants to do. And you’re like, ‘Oh, yeah, it’s kinda good. I’ll follow that.’ That’s no longer your idea. That’s no longer your intention, and you get to an outcome, and you look at it later and you’re like, ‘What is this?’”

“What you need to do is use AI as a tool to learn, a tool to understand the world better, a tool for leverage to get things that are not the critical stuff done faster, but don’t give up thinking. Don’t give up your brain. Don’t join the hive mind.”

Brian Ringley, Distinguished Product Manager & Human Robot Interaction Designer, Boston Dynamics

Brian Ringley works on Atlas, Boston Dynamics’ humanoid robot, where he focuses on the intersection of robotics, product, and human interaction.

“I think people should be looking at what’s happening right now with your ability to have a conversation with software, and think that we’re not too many steps from you being able to do that with a machine. Think about what it would mean to have a machine in your physical environment.”

“It’s really not enough to have the world’s best hardware, which I believe we do, or to connect that to all the various agents and models and policies that actually give it the intelligence that’s required. It has to be reliable. It has to be serviceable. It has to be cheap and affordable. I think all of those things are what will actually make for a successful humanoid product in the long run.”

“We’re not trying to trick you into thinking it’s a human. We’re not trying to give it any kind of agency in that way. We just want you to look at it and be like, ‘Oh, that’s a useful thing that belongs in a factory,’ and you kind of just understand that by the visual language of the design of the robot and the way that it behaves.”

“There’s a lot of conversation about how the AI brain will commoditize hardware. But I actually think that the opposite is true. No matter what that intelligence is, you still need this physical body that’s capable of interacting with the world, that can do it over and over again, that won’t break down. I think hardware is more important than ever by virtue of this software explosion.”

Ian Silber, Head of Product Design, OpenAI

Ian Silber leads product design at OpenAI, helping translate frontier research into products used by hundreds of millions of people.

“It’s such a cool time to be designing because the way you work with computers is changing. I remember this moment when I typed a bunch of typos into ChatGPT and it just understood me. I was like, ‘That’s just not how computers worked up until a couple years ago.’”

“I actually think that the role of design hasn’t changed at all in many ways. If you think of a designer as someone who’s drawing rectangles, then maybe that’s dead. But great designers are really thinking about and understanding the psychology of how someone interacts with something, the way people are going to approach it, use your product, and who you’re building for.”

“You could now in one prompt make DoorDash. But that doesn’t mean you have a successful business. There’s still a lot of work that goes into understanding what problems you can go after and solve, and how you do that. I think products that have that true care and human touch will really stand out.”

“If I were just starting my career, I would obsess over the changing landscape and tools. Don’t be afraid of it, but embrace it. Technology’s always changed very fast, but it’s never quite changed like this.”

Niko Klein, Product Manager, Figma

Niko Klein spent seven years as a product designer at Figma before becoming a product manager. At Config, he introduced Code Layers and explained why code is becoming another creative material.

“The workflows around code are not fun right now. They’re built for production. They’re built so you don’t ship bugs into production. But it’s not about getting to one idea, it’s about getting to like 600 ideas. Figma’s in the business of getting the bad ideas out so you can find the great ones.”

“Everyone is gonna get to the average a lot faster. And to get to something that stands out, you need to be kind of obnoxiously pushing on your idea, seeing how far you can take it, push it further than you could before.”

“Your responsibility as a designer, as an engineer, as a creator is going to become to not be satisfied and add more things that make it you, that make it your unique take on something. You gotta do this by trying things out over and over and over again.”

“Experts will work in clusters of representations, not files.”

Zach Lieberman, MIT Media Lab

Zach Lieberman leads the Future Sketches Group at MIT Media Lab, where artists and engineers build new creative tools and explore the future of computational creativity.

“The best moment for our students is when they connect what they’re learning now, using new technologies, different forms of data visualization, using AI in interesting ways, but connecting it to who they are. To me, that’s the most interesting moment.”

“Students who don’t have code experience can make incredible work now. Things are blossoming in terms of people who want to express themselves computationally. But also, people can use AI to do their homework really quickly, and then they’re not learning anything.”

“Writing is thinking, drawing is thinking, coding is thinking. If we automate away all of the thinking, we’re getting to the end result faster, but we’re losing all of the learning.”

“Creativity actually comes from the friction.”

Holly Herndon, Artist & AI Researcher

Holly Herndon has spent years exploring generative AI in music and art, often building her own datasets and training custom models rather than relying on commercial systems.

“I was talking about making art with AI and trying to have a human centered approach, where human intent remains at the core.”

“My favorite AI tool? Choirs. I like making collectively trained models. I like recording various vocalists and then making my own datasets. Making datasets can actually be really fun, and it can be an art in and of itself.”

“I think that underestimates human ingenuity and creativity. I think it’s just gonna ask more of us. And I think that’s gonna make us maybe grow creatively.”

“I’m excited to further build out our studio using agents, and see really how far we can take that.”

Shreyas Garg, Partner, IVP

Shreyas Garg is a partner at IVP and an investor in companies including Anthropic and Suno. We discussed investing, frontier models, robotics, and what comes after foundation models.

“Design is gonna become more and more important in the future. You’re not gonna have some agent be able to design beautiful things. It’ll have to be the human, and humans need the granularity to shape the small things that make a product beautiful and really fun to use.”

“I think a lot of people are looking for local maximums because they’ve missed the big labs, looking for companies that could be like that instead of spending time thinking, ‘What’s actually gonna be the next thing?’ I think a lot of people are spending time in the local maxima and not really thinking bigger picture.”

“I don’t view it as nearly so binary. I think the answer is just AI is gonna be big, and both will win. You’ll see companies benefit from both trends.”

“There’ll be tasks that you don’t need to use an Anthropic for. But there will be a lot of tasks. I’m invested in a protein model helping pharma companies develop better drugs. I want frontier intelligence on that. I want that thing trying to solve cancer. You’re not gonna use the open source cheap model for that. You’re gonna use the expensive stuff because you actually want to solve a very, very important problem. I think the value of frontier intelligence will be pretty durable.”

Subscribe now

Share

Want more Figma? Check out our interview with CEO Dylan Field

→ Listen on X, Spotify, YouTube, Apple

Brought to you by:

Turing—Turing partners with frontier AI labs to improve model capabilities in coding, reasoning, tool use, & multimodality, as well as with Fortune 500 enterprises to build & deploy end-to-end agentic AI systems in mission-critical workflows Visit: turing.com/sourcery

VCX—VCX is the public ticker for private tech, allowing investors of all sizes to invest in venture capital. View The Portfolio at GetVCX.com

Deel—Deel is the global people platform that helps startups hire, manage, pay, and equip anyone, anywhere. Trusted by more than 35,000 fast-growing companies, Deel is the people platform that just works, so teams can scale without the chaos. Visit: deel.com/sourcery

Public-–Investing platform Public just launched Generated Assets, which lets you turn any idea into an investable index with AI. With Generated Assets, you can build, backtest, refine, and invest in any thesis with AI. Gone are the days of one-size-fits-all ETFs. Try it today: public.com/sourcery

The material presented on Molly O’Shea’s website are my opinions only and are provided for informational purposes and should not be construed as investment advice. It is not a recommendation of, or an offer to sell or solicitation of an offer to buy, any particular security, strategy, or investment product. Any analysis or discussion of investments, sectors or the market generally are based on current information, including from public sources, that I consider reliable, but I do not represent that any research or the information provided is accurate or complete, and it should not be relied on as such. My views and opinions expressed in any website content are current at the time of publication and are subject to change. Past performance is not indicative of future results.

Paid Endorsement. Brokerage services by Open to the Public Investing Inc, member FINRA & SIPC. Advisory services by Public Advisors LLC, SEC-registered adviser. Crypto trading provided by Zero Hash LLC, licensed by the NYSDFS. Generated Assets is an interactive analysis tool by Public Advisors. Output is for informational purposes only and is not an investment recommendation or advice. See disclosures at public.com/disclosures/ga. Matched funds must remain in your account for at least 5 years. Match rate and other terms are subject to change at any time.

Cyan F*cking Banister

Cyan Banister put her entire IronPort exit into SpaceX when the rockets were still blowing up on the launchpad. She never sold. It became the best investment of her life.

→ Listen on X, Spotify, YouTube, Apple

She is Co-Founder and General Partner at Long Journey Ventures, spent 4 years as a partner at Founders Fund, & with her husband Scott Banister has built one of the most successful angel portfolios in the world across 20+ years of investing together:

SpaceX. Anduril. Uber. PayPal. Affirm. Flexport. Postmates. Niantic. Opendoor. Carta. Together AI. Diamond Foundry. Crusoe. Flock Safety. Substrate. Brave. Depop. Calm. Fivetran. Zappos.. & many more..

Cyan joins Sourcery at Long Journey's HQ to break down how she actually sources deals, why she refuses to chase hot rounds, and what she is buying next in nanotech and biotech. She found Niantic in a customer support inbox at Hint Water. She found Uber by asking taxi drivers what their day was actually like. Her point is that the alpha is in paying attention, and that the best investments arrive as a feeling before they arrive as a thesis.

This was such a pleasure to record. Cyan is the GOAT & one of the best storytellers of all time. May we all try to live life the way Cyan does..

Subscribe now

Share

𝐓𝐈𝐌𝐄𝐒𝐓𝐀𝐌𝐏𝐒

(00:00) Cyan Banister, Long Journey Ventures
(01:13) Inside Long Journey HQ
(02:17) The wild story behind the office mural
(05:32) What Cyan's actual house looks like
(07:18) The candle ritual
(08:08) What the Long Journey logo actually means
(10:41) How Cyan became one of Mafia's breakout stars
(13:37) The Mafia game they should've won
(14:45) Cyan's actual strategy for winning Mafia
(15:42) The first time she's ever had to lie to win
(16:30) What really happens off-camera at Mafia
(18:58) The blooper reel nobody's seen yet
(20:58) The massive production behind one card game
(22:33) The one thing Mafia is missing
(23:54) Brian Singerman's x $40-a-month board game club
(27:05) Board games & startups
(29:10) Confessions of an accidental poker legend
(31:36) Cyan's Serendipity
(33:37) What relentless optimism reveals about people
(35:42) Why so many people are scared to be weird
(37:32) The daily ritual she prescribes to everyone
(43:56) The "why" behind her biggest investments
(50:49) Why sci-fi movies keep predicting the future
(53:37) Should you even go to college right now?
(59:58) Unconventional investing philosophy
(1:01:57) Full list of Cyan's investments
(1:04:28) All-in on SpaceX
(1:08:57) Why she never sold a single SpaceX share
(1:12:53) Top Bets: Becoming Bio + Substrate
(1:15:52) Why America's reindustrialization moment is now
(1:19:16) The mentors who shaped who she is today
(1:22:54) Manifesting Rick Rubin
(1:24:00) Real take on the Peter Thiel events

Brought to you by:

Brex—The intelligent finance platform: cards, expenses, travel, bill pay, banking—wrapped into a high-performance stack. Built for scale. Trusted by OpenAI, Anthropic, Vercel, Granola, Deepgram, & Sourcery.. teams that move fast AF. visit → brex.com/sourcery

Turing—Turing partners with frontier AI labs to improve model capabilities in coding, reasoning, tool use, & multimodality, as well as with Fortune 500 enterprises to build & deploy end-to-end agentic AI systems in mission-critical workflows Visit: turing.com/sourcery

VCX—VCX is the public ticker for private tech, allowing investors of all sizes to invest in venture capital. View The Portfolio at GetVCX.com

Deel—Deel is the global people platform that helps startups hire, manage, pay, and equip anyone, anywhere. Trusted by more than 35,000 fast-growing companies, Deel is the people platform that just works, so teams can scale without the chaos. Visit: deel.com/sourcery

Public-–Investing platform Public just launched Generated Assets, which lets you turn any idea into an investable index with AI. With Generated Assets, you can build, backtest, refine, and invest in any thesis with AI. Gone are the days of one-size-fits-all ETFs. Try it today: public.com/sourcery

Cyan Banister’s Legendary Track Record

Cyan Banister’s first investment ever was SpaceX. She put her entire 2007 IronPort exit into it. She never sold. Savage.

In this episode of Sourcery, I sat down with the Co-Founder & GP of Long Journey Ventures at the firm’s magically weird HQ to get inside the brain of one of Silicon Valley’s greatest investors of all time.

→ Listen on X, Spotify, YouTube, Apple

The SpaceX Bet

IronPort, the email security company Scott Banister co-founded, sold to Cisco in 2007 for $830 million. Cyan had traded salary for equity on every promotion, so it was her first real liquidity event, and she was unsure what to do with it. Then Luke Nosek came over to the house, got down on the floor, and made his pitch. “Banister’s, I need you to dig through your couches. Whatever money that you have that’s spare, anything liquid you have, I need it. SpaceX needs it.”

Rockets were still failing on the launchpad. The consensus was that private citizens had no business in space and that the failures proved it. Luke and Scott made the opposite argument: if you’re able-bodied, employable, young, and have a safety net, this is what the safety net is for. She said yes, then immediately panicked. “So I did it, and then afterwards, I was like, ‘What have I done?’” Friends told her she’d lit her money on fire.

She never sold.

“I just literally put all my chips on a color, and just let it ride for 20-some odd years now. And it ended up being the best investment I’ll probably ever make in my life.”

SpaceX went public on the Nasdaq on June 12, 2026, in the largest IPO in history, pricing 555.56 million shares at $135 and raising $75 billion at an initial valuation of $1.77 trillion. By its third day of trading the stock closed at $201.68, roughly 50% above the IPO price, making SpaceX the sixth-largest publicly traded company in the US market.. not long after it hit $2.99 trillion market cap.

Since that initial post-IPO peak, the valuation has cooled and currently hovers between $1.9 trillion and $2.1 trillion.. for context, SpaceX was valued around $300B just a year ago.

Like many investors, Cyan is still in lockup post-IPO.

The #1 Angels in the World

Cyan and her husband Scott Banister have invested together for over 20 years, and by her account, combined they’re the top-ranked angel investors in the world. “I think the two of us combined are definitely #1 in the world.” The nuance she insists on: they operate as one unit. “Everything that’s mine is his, and everything that’s his is mine, and we’ve been partners since we started investing together.”

Even separated out, they’d each land top 10 on the Stanford list; combined, #1. When TechCrunch gave them Angel of the Year, she refused to accept it solo. “I don’t wanna go on stage and accept an award that my husband also deserves.” The Banisters won that award for early bets on SpaceX, Uber, & DeepMind.

Others include: SpaceX. Anduril. Uber. Zappos. PayPal. Affirm. Flexport. Checkr. Density. Flock Safety. Brave. Zenefits. CTRL-labs. Depop. Substrate. Carta. Together AI. Turing. Postmates. Niantic. Cargomatic. Diamond Foundry. Upstart. Fivetran. Forge. Opendoor. Calm. TrueMed. Crusoe.

Several have compounded into landmarks. Anduril raised a $5 billion Series H at a $61 billion valuation in May 2026, more than double its valuation a year earlier.

Ask her to pick a favorite and she won’t. “It’s like children. You can’t pick your favorite.” What she’ll admit to is the joy of an investment escaping the garage and showing up in the physical world. “I always take pictures of billboards because I call it seeing my investments in the wild.” The through line, she says, is companies that create jobs and offer flexibility, from rocket ships to sandwich delivery.

How Cyan Actually Sources Deals

The SpaceX story gets told as a conviction story. It’s closer to a sourcing story, and the sourcing story is the one worth learning from. With Niantic, she didn’t notice the game. She noticed the behavior. Players of a fringe app called Ingress were renting helicopters, chartering boats, ditching friends at DEF CON to drive into the desert and claim objects that existed only inside an app. “That is remarkable. What could get people to do this?”

So she played it, then asked the question most people skip: why is Google doing this? The answer was that Ingress players were producing categorized, geotagged, photographed map data for free. She wanted in & couldn’t, because it was Google.

When Alphabet spun Niantic out, she had no way to reach anyone there, but she had an unrelated memory: while helping set up a support ticketing system at Hint Water, another of her investments, she’d kept seeing “Ingress code” in the subject lines. Hint was putting game codes in its bottle caps through a Niantic partnership. “If I wasn’t paying attention to what was coming to the inbox of that ticketing system, and how did I even end up there in the first place, that’s the another miracle.”

That thread led to Kara Goldin, then to John Hanke, whose first response was that Niantic had Nintendo and Google and didn’t need her. “Please, just hear me out. Just take an hour of my time, and I guarantee you, you will not regret it.” She showed up with a best friend who was an Ingress player; Hanke hired him nearly on the spot. Then Pokemon Go happened.

Niantic sold its games division to Scopely in March 2025 for $3.5 billion, with an additional $350 million in cash bringing total value to equity holders to $3.85 billion. The remaining geospatial business spun out as Niantic Spatial with $250 million in capital, still led by Hanke.

The Uber Method, Run in Reverse

She didn’t find Uber. She’d already done the thinking before Uber existed, because she’d spent years asking taxi drivers what their day was actually like. What they told her was a product spec. “Well, I wake up, I go to the taxi yard, I give them $200. I’m in the hole $200 from the moment I walk in. I drive my car out of the gate. I immediately have to make that $200 back.”

A driver who starts every shift $200 underwater will drive fast, arrive cranky, and lose your bag. The ride was bad because the job was bad. Uber’s flexible employment wasn’t a feature bolted on; it was the answer to a question she’d been asking for years, and when Uber arrived she recognized it instantly.

This is the engine underneath the whole portfolio. “When you ask why about everything in the world, it’s just gonna make you a better investor. It’s gonna make you a better entrepreneur, a better everything.” Her phrase for it: “Human beings are why machines.” She doesn’t start her day until she has a why to carry through it.

Personality Is Not Fixed

It really isn’t until the last 15% of the conversation do we actually start talking about investing, up until then we talk human behavior, curiosity, how the world works.. and as a life/performance/career coach, an incredibly high EQ, Cyan has a real talent with helping lead people to more personal clarity & agency.

“We have this misconception that our personality is fixed, and we say, ‘This is just how I am.’” Her claim is that this is false, that most people are running excuses rather than decisions, and that the excuses are load-bearing. I’m a Scorpio. I’m Irish, I have a temper. She thinks personality is editable, and that the edit is expensive because it requires looking directly at things you’ve arranged not to see.

“You can actually surgically go in and alter your personality, where you can actually change these things, but it takes effort and time and a lot of facing the ugly truth about yourself.”

She runs a thought experiment every week, and the current one is: “How wide is the space between your values and your actions?”

She sits down with paper. If she claims to be honest, where did she lie this week, including the small ones, the meeting she skipped, the email she promised.

Her explanation for why this happens is unusually precise and mostly generous. “That person in the moment intends to send that email, but the person two hours later is not the same person.” She notes, with obvious relish, that Marc Andreessen advises against introspection, and that she’d like a word with him about it.

What Worries Her

Peter Thiel used to ask her what worries you, and she thinks it’s the better question. What excites you is interesting; what worries you is diagnostic. What worries her is a surveillance state operated by AI & enforced by robotics, and she frames it not as an abstraction but as a logistics problem that’s already half-solved. Skynet.

“Right now, we’re moving towards a future where you won’t be able to get into a car & drive yourself somewhere, & if you think about it, that freedom of movement is a human right.”

She takes it a bit further.. “What if someone says, ‘You know what? You’re just not going anywhere. I’m just gonna shut your Waymo down, and not only that, but I’m gonna lock you inside.’” That’d be a huge bummer.. lol.

For context, Waymo now runs roughly 500,000 paid rides a week across 11 US metros with about 3,000 vehicles, at a $126 billion valuation, targeting 1 million weekly rides by the end of this year. Let’s hope that doesn’t happen! (it won’t)

Her framing is that the technology isn’t the variable. “A gun can be a paperweight or a weapon. You know, AI is the same.” Which is why she backs open source and decentralized control. She got into the industry in 1999, when the assumption was the whole ecosystem would stay open and distributed, and she watched it centralize instead. Her worry is one ideology at the controls, “you don’t want one company with one ideology ruling it all. That’s dangerous.”

AI as a New Paintbrush

The excitement runs on the same logic as the worry. “I’m really excited about AI being a new paintbrush and a new tool to unlock all the creativity in this world that has been dormant.” Her bet isn’t that AI produces great work; it’s that AI removes the capital requirement from ambition. The person with a million-dollar idea who can’t afford a manufacturing run, a film crew, an engineering team.

Vibe coding is here. Vibe manufacturing, she thinks, is next, producing a wave of newly minted millionaires running fully agenticized businesses, most not venture-scalable & not caring.

On the slop objection, she’s direct. “Right now, we’re in a period where it’s the worst it’ll ever be.” She’s also unromantic about where the models came from and where the fight goes, predicting China trains on Western IP regardless, remixes it, and sells it back to us, and that we buy it. She’s writing a book with AI assistance now and says the tells are obvious to anyone who’s read enough of it, which is exactly the job: the model does the heavy lifting, the human massages it back into a human shape.

The consumer wave, she thinks, lands in two to three years, and there’s one company she’s still hunting. “I am still looking for the Alibaba of the Americas.” She’s turning over stones. Nobody has pitched it yet.

Why Now vs. What’s Coming

Ask her about the deep tech boom and she’ll explain why now with real precision. Manufacturing and robotics costs have collapsed. SpaceX proved the category could exist at all, which is what let a Varda be imagined. There’s a technology war underway with countries that don’t share our interests. Wars will be fought with robotics. And then she tells you she doesn’t invest on any of that.

“I tend to not play in the why now space. I tend to play in the what’s coming space.” Six years ago that meant the infrastructure layer under crypto and AI, before either was consensus. Today it means nanotech, biotech, and the bottlenecks that fall away once AI can read data sets nobody’s been able to read. Her most exciting current positions are Substrate, Becoming Bio, & Diamond Foundry, which she frames as the substrate of technology, the substrate of biology, and the material the first two run on.

Diamond Foundry is the cleanest illustration. When she invested, the visible business was fancy diamonds and everyone stopped there. The company now produces single-crystal diamond wafers for high-power chips in AI, cloud, EVs, and 5G, and was valued at $1.8 billion after a $200 million Fidelity round. It opened a commercial plant in Trujillo, Spain, in 2025. The jewelry funded the wafers.

“If I’m chasing everything in the hyper-competitive deals of today, I’m gonna lose money.”

She says this should be every seed fund’s secret, and that it mostly isn’t. Long Journey, co-founded with Lee Jacobs & Arielle Zuckerberg, backs what the firm calls magically weird founders; Fund IV closed at approximately $181.8 million in March 2025. The logo is a sheep’s eye named Bellwether, looking sideways, because the sheep that leads the flock is the one watching for what’s coming from the side.

Free Speech Is the Number One Cause

The last thing she says is the thing she cares about most. “A lot of people wanna save the oceans, wanna bring water to Africa, wanna cure cancer. I wanna protect free speech.” It’s not abstract for her. It’s why she defends going to Peter Thiel’s events, and why she found the recent wave of distancing statements exhausting. Attendee records for Dialog, the invitation-only network Thiel co-founded with Auren Hoffman, were exposed in June 2026, and Ezra Klein acknowledged on X that he’d attended conferences in 2018 and 2022 while stating he wasn’t a member.

Her position is that the shame is the problem. “If you get invited to a Peter Thiel event, go. Like, do not shy away from it.” The argument underneath isn’t a defense of Thiel; it’s a defense of category integrity. “Racism is real. It’s a problem. And so if we call everyone a racist, then we can’t find the real ones. If we call everyone a pedophile, how are we gonna find the real pedophiles?”

Then the line that made the rounds on X, delivered with a straight face and explained with a laugh: “I eat Mars babies. I was like, yeah, I eat babies, but only from Mars.” The joke is a proof. There are no babies on Mars. She extends it into a real thesis: “You can tell the health of a nation by its ability to tolerate comedy and rap music. Those two things have to exist for freedom, and if either of those start to disappear, you’re in trouble.”

On Peter, she offers the line she’s clearly used before and still means: “I always like to joke that I didn’t go to university, but I went to the Peter Thiel University.” And then the part that’s actually useful and applies to anyone: “He could be your mentor without ever meeting him if you’re open-minded and you ask yourself why. Why would he do this? And don’t make assumptions.”

The Two Who Shaped Her

Asked who inspires her, Cyan names Peter Thiel first. He was her partner for four years at Founders Fund, and her framing of the education is the line she keeps returning to: “I always like to joke that I didn’t go to university, but I went to the Peter Thiel University.”

Her defense of him is that he’s misread. “I’ve never met anyone more tolerant, more open-minded. He gets such a bad rep for a lot of things, but he’s just probably one of the most poorly understood people.” The reason, she says, is distance. “He’s operating 1,000 feet above everybody else. His mind is not like other minds. And so when you encounter something like that, you don’t know what to do with it, so it’s just easier to hate it.”

Marc Andreessen was the other, and the older one. He was her hero as a teenager, on magazine covers, the Netscape founder whose work she credits with changing her world. “I told myself when I was a teenager, ‘I’m gonna meet that guy someday. I’m gonna sit in a room with him, and he’s gonna ask me a question, and he’s gonna value me.’ And I didn’t meet him until that was true.”

However, she made a point that she didn’t actually have to know either to learn from them.. “You can learn a lot from people without ever knowing them, by just watching them and observing them.”

She rounds it out with a 4th name most people wouldn’t guess. Scott Cook, the Intuit co-founder behind QuickBooks, TurboTax, and Mint. Her reason is executive function, not fame. “I’ve never met anybody in my life with more executive function, who has a complete understanding of the lay of the land of his company, his employees, his culture, his board, and the emerging technologies of the world, who is so down to earth.” She calls him the gold standard for a fine human being in an executive position.

Manifesting Rick Rubin

Cyan closes by putting a wish on the record. “It is my dream, I’m gonna put it out in the universe, that I get to meet Rick Rubin someday.” She feels a rare kinship with him.. “I do think that he’s the first person, besides Bill Murray, that I’ve ever seen in my life who I think my mind is similar to.”

What she means is that they both work “based off of this openness to, for lack of a better term, the universe presenting to you solutions.”

The only problem is access. “I don’t know how to reach him. I don’t know how to reach his team. I just keep putting it out in the universe, and I figure it’ll eventually happen.”

The irony is that his podcast, Tetragrammaton, has already hosted Peter Thiel, her former partner of four years at Founders Fund.. and “He brought on Bill Gurley, and Bill, you know, I’m sure they talked about Uber, but we brought Uber to Bill Gurley. Maybe that’ll get me on.”

PSA: If you can help connect the two, that’s a collab the whole world needs to see

→ Listen on X, Spotify, YouTube, Apple

Subscribe now

Share

The material presented on Molly O’Shea’s website are my opinions only and are provided for informational purposes and should not be construed as investment advice. It is not a recommendation of, or an offer to sell or solicitation of an offer to buy, any particular security, strategy, or investment product. Any analysis or discussion of investments, sectors or the market generally are based on current information, including from public sources, that I consider reliable, but I do not represent that any research or the information provided is accurate or complete, and it should not be relied on as such. My views and opinions expressed in any website content are current at the time of publication and are subject to change. Past performance is not indicative of future results.

Paid Endorsement. Brokerage services by Open to the Public Investing Inc, member FINRA & SIPC. Advisory services by Public Advisors LLC, SEC-registered adviser. Crypto trading provided by Zero Hash LLC, licensed by the NYSDFS. Generated Assets is an interactive analysis tool by Public Advisors. Output is for informational purposes only and is not an investment recommendation or advice. See disclosures at public.com/disclosures/ga. Matched funds must remain in your account for at least 5 years. Match rate and other terms are subject to change at any time.

The ‘Quiet’ Silicon Valley Legend

Accel Late-Stage Partners Arun Mathew, Miles Clements, & Matt Weigand join Sourcery to go deep into the lore of the quiet, yet legendary, Silicon Valley firm.

→ Listen on X, Spotify, YouTube, Apple

We go back through Accel's 40 years of history, starting with the 10% Facebook stake and the secondary they modeled at 5X, which under-shot the outcome by an order of magnitude, and how the firm now runs a global AI portfolio spanning chips, neoclouds, labs, and applications, with exposure across Cursor, Anthropic, and Nebius.

The three have invested together for 15 years, and they break down how a generalist team wins competitive deals against flashier names, why late-stage returns are starting to match early-stage top quartile, and how check sizes moved from a $480M first growth fund to single investments north of $500M and, at times, over $1B.

Topics include Matt's $150M Nebius PIPE, now up 10-13X, and its $26B Meta deal; the "agentic influence" thesis behind the Lovable investment; Supabase growing 350% on nearly all inbound at 9M developers; Cyera's path from Series A to the highest-valued private security company; tokenmaxxing; and 3x trillion dollar IPOs.

We also get into the hot topics moving tech, AI, and investing right now: the agent economy and real enterprise adoption, the inference buildout and whether infrastructure is the rate limiter, AI optimization replacing SEO as the new distribution channel, the concentration of capital into a handful of late-stage rounds, security as an AI tailwind rather than a casualty after Mythos, the race toward $10 trillion companies, and what it means for retail when SpaceX, and others, finally hit the public markets.

Subscribe now

Share

𝐓𝐈𝐌𝐄𝐒𝐓𝐀𝐌𝐏𝐒

(00:00) Arun Mathew, Miles Clements, Matt Weigand, Partners at Accel
(00:49) Meet the Accel growth team
(01:48) The Facebook deal almost no one believed in
(06:19) The Thesis behind every Accel bet
(09:06) The silent strategy behind Accel's biggest wins
(09:51) The real reason Cursor picked Accel
(14:02) $200B into three companies in one quarter
(16:05) Early-stage vs. Late-stage returns
(17:18) Why trillion dollar companies are coming next
(20:13) The $26B Nebius deal, explained
(23:15) Building the AI infrastructure everyone will need
(26:22) The $150M bet that's up 13X today
(27:44) How agents are quietly rewriting enterprise software
(35:16) The three categories getting the biggest tailwinds
(36:35) From data security to AI security: the Cyera story
(39:48) How AI agents are creating infinite breach risk
(41:51) Is "token maxing" actually a real problem?
(44:22) Betting big on AI's exploding token economy
(47:53) Bracing for three Trillion Dollar IPOs
(51:05) What Accel is betting on next
(52:45) The unexpected place AI is already changing lives

Brought to you by:

Brex—The intelligent finance platform: cards, expenses, travel, bill pay, banking—wrapped into a high-performance stack. Built for scale. Trusted by OpenAI, Anthropic, Vercel, Granola, Deepgram, & Sourcery.. teams that move fast AF. visit → brex.com/sourcery

Turing—Turing partners with frontier AI labs to improve model capabilities in coding, reasoning, tool use, & multimodality, as well as with Fortune 500 enterprises to build & deploy end-to-end agentic AI systems in mission-critical workflows Visit: turing.com/sourcery

VCX—VCX is the public ticker for private tech, allowing investors of all sizes to invest in venture capital. View The Portfolio at GetVCX.com

Deel—Deel is the global people platform that helps startups hire, manage, pay, and equip anyone, anywhere. Trusted by more than 35,000 fast-growing companies, Deel is the people platform that just works, so teams can scale without the chaos. Visit: deel.com/sourcery

Public-–Investing platform Public just launched Generated Assets, which lets you turn any idea into an investable index with AI. With Generated Assets, you can build, backtest, refine, and invest in any thesis with AI. Gone are the days of one-size-fits-all ETFs. Try it today: public.com/sourcery

Inside Accel’s Growth Machine:

40 Years, 10% Facebook Stake, AI Supercycle

From a 10% Facebook Stake to Three Tech Cycles

Accel late-stage Partners, Arun Mathew, Miles Clements, & Matt Weigand, have worked together for 15 years at one of the most quietly dominant firms in Silicon Valley. The firm’s identity traces back to stakes that now read as inevitable but were anything but at the time. Accel led Facebook’s Series A in 2005 out of its early-stage fund and later held a stake reported at around 10%.

The story the partners keep returning to is the secondary. Accel created its growth vehicle in 2008, and one of its first opportunities was buying Facebook secondary shares shortly after Microsoft invested at a $15 billion valuation. Accel had the chance to buy in a little south of $20 billion. As Arun recalls the partner debate “We actually think this could be $100 billion business or more, that we could generate north of a 5X.” Both numbers proved far too conservative.

That miss is now cultural shorthand at the firm. “Lo & behold, that was way under-shooting what the opportunity and the potential of Facebook was,” Arun said. The lesson was not just about scale but about structural humility. Matt describes early partner Ryan Sweeney, who helped build the growth practice and drilled a specific discipline into the team, he “would remind us every day to be humble and to hustle.”

The through-line to today is the platform pattern. Miles points to Facebook as the first platform company the team watched founders build on top of, a dynamic he sees repeating now with the labs. The partners frame the current AI cycle as the same distribution shift they lived through once already, which shapes how aggressively they move on it.

→ Listen on X, Spotify, YouTube, Apple

Meet Arun Mathew, Miles Clements, & Matt Weigand

The growth team’s edge is overlapping coverage, not just tenure. All three partners joined Accel within a four-year window and have compounded relationships and pattern recognition across enterprise, infrastructure, & application software through multiple cycles.

Arun Mathew joined Accel in 2009 after starting as an associate at Insight Venture Partners, and holds degrees from the University of Pennsylvania (Wharton) and Stanford’s Graduate School of Business. He leads growth investments across enterprise, security, and infrastructure, and runs the firm’s Tech Council. His portfolio spans 1Password, Tenable, PagerDuty, Webflow, G2, Squarespace, and Qualtrics (acquired by SAP), alongside a deep India book that includes Flipkart, Freshworks, BookMyShow, and Ola. He also sourced Accel’s early investment in Groupon.

Miles Clements joined in 2009 as Accel’s first associate, coming from a product role at lynda.com (later acquired by LinkedIn), with degrees from the University of Virginia and Harvard. He helps lead the growth fund across AI, cloud, and enterprise software, and oversees Accel’s Ignite program for revenue leaders. He has led the firm’s investments in Linear (an $82 million round in March 2026 valued the company at $1.25 billion), Cursor ($60B SpaceX acquisition), Podium, and Laravel (a $57 million Series A), and worked on landmark deals in Atlassian and UiPath earlier in his career.

Matt Weigand joined in 2013 from William Blair’s technology investment banking group, & earlier Deloitte, with a degree from Miami University. He focuses on growth-stage enterprise software, security, fintech, and infrastructure. He led Accel’s $150 million Nebius PIPE and sits on the boards of Snyk, Cognite, Veriff, & DriveWealth, with active involvement across Scale AI, Axonius, G2, Xero, GoFundMe.

The Generalist Model & Why Late-Stage Now Rivals Early-Stage Returns

Accel’s growth team runs a generalist book with individual areas of depth. Arun concentrates on infrastructure and cybersecurity, Matt on infrastructure, and Miles on application software and developer tooling. Miles frames the requirement bluntly, in a market where categories appear & scale in months, an investor needs “some mental plasticity to wrap your head around these new categories as they evolve.”

The firm’s competitive edge, as the partners describe it, is reputation rather than marketing. Miles recounts asking Cursor CEO Michael Truell why he picked Accel in a crowded round. The answer “I asked around, & people said really nice things about Accel.” Matt adds that for most of the firm’s history, being first mattered above all “it was almost a religion here to be the first institutional investor into a company.”

The structural shift the team is underwriting is that late-stage returns are converging with early-stage. Matt notes that when he joined, Accel had roughly 90% coverage of every seed and Series A, a level no longer possible given how many companies now start. But a small cohort of outliers runs into the late stage, and Matt argues top-quartile late-stage funds will roughly match top-quartile early-stage this cycle. Brian Singerman, formerly of Founders Fund, made a similar point in a prior Sourcery interview.

The check sizes reflect the change. Accel’s first growth fund was $480 million. Single investments of $500 million, and at times over $1 billion, are now routine. Arun frames it as rational given the prize “if we believe that these companies can be trillion, multi-trillion dollar companies within a very short hold period, we should reflect that in check size.”

Six Investors: How Accel’s Partnership Is Built

For a firm operating across the full stack & three geographies, the late-stage engine is small. Matt notes about 6 investors deploy most of that capital, working in concert with the wider team. The continuity is the point “the three of us have worked together for 15 years, and a lot of it has grown from within.”

The firm’s investing philosophy rewards moving outside a lane. Arun cites Andrew Braccia, whom he calls “one of the best enterprise software investors in the last two decades,” despite a decade at Yahoo as a consumer internet operator. Braccia “led our seed investment into Slack early on.” Sameer Gandhi, a prolific consumer investor, became one of the firm’s leading cybersecurity investors. The instruction Arun draws from this is to hold focus and domain depth while staying open “it’s also really important to repot yourself and to look out.”

That model depends on a bench maturing into lead roles. Asked what he is most looking forward to, Miles pointed not to a company but to his younger colleagues, naming Ben Quazzo, Christine Esserman, Gonzalo Mocorrea, Josh Fang, & Rohan Kamat. Several have led investments that are not yet public knowledge, and Miles expects some to become known names in the coming months “I think we have some rising stars that are gonna really, really hit their stride.”

The takeaway for founders and LPs is that Accel’s edge is built on generational continuity rather than star individuals. A tight core, a wide bench, & a culture that pushes investors to keep repotting is how the firm sustains coverage as the market fragments.

Nebius, Neocloud Bet, & Inference Buildout

One of the most recent killer deals: $150 million PIPE into Nebius, led by Matt, roughly 16 months before the company became a consensus name. Miles, bragging on Matt’s behalf because “he’s not gonna brag on himself,” pegged the position as up 13X (as of June 2, 2026). The context around Nebius has since caught up to the thesis in the public data.

Nebius, the AI-infrastructure company built from the team that followed founder Arkady Volozh out of the former Yandex, reported Q1 2026 revenue of $399 million, up roughly 684% year over year, with AI cloud making up the vast majority of it. The company has guided to exit-2026 annualized run-rate revenue of $7 billion to $9 billion. It has signed roughly $46 billion in multi-year capacity contracts, headlined by a five-year deal with Meta reported near $27 billion and a Microsoft agreement around $17 billion, and Nvidia has committed to invest $2 billion. The stock has climbed well over 150% in 2026, putting the market cap in the high-$50-billion to high-$60-billion range.

Matt’s framing on Arkady captures why Accel structured a public-market investment at all “he is the most quietly humble killer I’ve ever met.” Nebius’ vision, in Matt’s telling, was a vertically integrated hyperscaler owning everything from the data center and server racks through the software layer, built for a world where inference demand compounds as agents proliferate.

That inference wave is the core of the bull case. Estimates vary widely across research shops, but multiple 2025 baselines put the AI inference market near $106 billion, growing to roughly $255 billion by 2030 at high-teens annual rates. Matt’s own read is that the buildout has barely started “I still view it as we’re inning one, and we’re just getting going.” Accel is pushing deeper into the stack alongside Nebius, including inference software from the ex-xAI team at RadixArk.

Agent Economy, Supabase, & “Agentic Influence”

The team’s clearest signal that agent adoption has moved from demo to production comes from its own portfolio telemetry. Arun points to Supabase, the Postgres-based backend for a large share of agent workflows “That’s growing 350% at hundreds of million dollars of scale, & they have virtually no salespeople. It’s all inbound.” Supabase has crossed 9 million developers, up from under 1 million when Accel first invested, with most of the growth in the last three or four months.

The mechanism, Arun argues, is a distribution shift from search engine optimization to what he calls AI optimization: coding tools like Claude and Codex now recommend backend products, and they favor the ones with the best developer experience. He notes roughly 60% of YC companies now choose Supabase, and that a framework easy for humans to use turns out to be easy for AI to use.

The pattern crystallized into a thesis Miles and Arun call “agentic influence.” After seeing simultaneous spikes across a Supabase board meeting and a Linear board meeting, the team recognized that “all of a sudden these agents are making decisions about downstream workflows and downstream tool creation.” The conclusion was to invest in every company sitting in that flow, especially the choke points metering the decisions.

That gave the team, working with partners Ben Fletcher & Zhenya Loginov, the conviction to back Lovable, and it reinforced existing positions in Vercel & Cursor. Arun’s broader point is that enterprise usage, not just vibe-coding hobbyists, is driving the curve, which he says gives him confidence the adoption is still in early innings.

Cursor & the $60 Billion SpaceX Acquisition

Cursor recurs throughout the interview as a portfolio company and as a canonical “choke point” in the agent economy. Miles describes a competitive early process and a second, conviction-led follow-on. Importantly, the episode was recorded before the deal below was finalized, so the partners discuss Cursor purely as a private holding and do not address an acquisition. The following is context added after taping.

Since recording, SpaceX agreed to acquire Cursor parent Anysphere in an all-stock deal valuing it at $60 billion, announced June 16, 2026 and expected to close in Q3 2026 pending regulatory approval. The transaction traces to an option SpaceX secured in April, which it exercised days after its Nasdaq IPO. Reporting describes it as one of the largest venture-backed startup acquisitions on record.

The financing mechanics underscore the “IPO as currency” theme the partners raise elsewhere. SpaceX debuted on Nasdaq on June 12, 2026, raising in the range of $75 billion to $86 billion and reaching a valuation above $2 trillion, then used its new public equity to move on Cursor within days. Cursor had reached roughly $4 billion in annualized revenue in under four years.

Bloomberg has solid coverage, complete with an estimated stake cap table.

Cyera & Why Security Became an AI Tailwind

Cybersecurity is one of three categories Arun flags as seeing dramatic tailwinds, alongside developer-first companies and AI infrastructure. The counterintuitive part is timing. After the release of the Mythos model, the market initially expected AI to kill incumbent security names. The opposite happened “The last six weeks for our security companies have been tremendous,” Arun said, pointing to CrowdStrike’s move over the same window.

Accel’s flagship position in the category is Cyera. Arun traces the firm’s data-security lineage back to Varonis in Israel, and credits London partner Philippe Botteri, who intersected Cyera CEO Yotam Segev on a biannual Israel trip, with sourcing the deal. Accel co-led the Series A with Sequoia, then leaned in through a difficult 2022 quarter to lead the Series B, and later led a subsequent round.

The public data confirms Arun’s claim that Cyera is among the most valuable private security companies. In June 2026 Cyera raised $600 million at a $12 billion valuation, led by Evolution Equity Partners with Accel participating, bringing total funding above $2 billion. That marked a roughly fourfold increase in about 18 months, following a $9 billion valuation in January 2026 and $6 billion in mid-2025. The company reports annual recurring revenue tripling for three consecutive years and roughly one-fifth of the Fortune 500 as customers.

Arun’s structural view is that value will concentrate, not fragment, he expects a few trusted platforms, Cyera, CrowdStrike, & Palo Alto Networks among them, to accrue most of the value as the attack surface expands. The driver is scale of creation: with hundreds of millions of people now building on AI, the surface area needing protection is growing at an exponential pace.

What Happens After Tokenmaxxing

“Tokenmaxxing,” the episodic overconsumption of model tokens, has grabbed headlines, but the team’s read is that the dominant trend runs the other way. Miles cites an Accel blind developer survey asking whether CFOs were pushing teams to spend less or more on consumption. The result “seven times more companies are being told to let it rip and spend more.”

Arun ties consumption directly to the pace of capability gains. Because what is possible on AI expands week to week, he argues the priority for most companies is deeper adoption rather than edge optimization, and that cost-cutting behavior will arrive only when the capability frontier plateaus. Matt frames the same point around forecasting history “I think you’re gonna see revisions up on tokens, for sure,” comparing it to every underestimated infrastructure forecast.

Matt does acknowledge a governor. Consumption scales only as far as the value returned, and if models asymptote in capability, the indigestion around token spend becomes real. For scale context, the broader AI market was estimated near $390.9 billion in 2025 and around $539.5 billion in 2026, a base against which even large token-spend figures represent early penetration.

The practical takeaway for operators is that current overconsumption examples are outliers that will be curtailed, while aggregate consumption keeps climbing. The team’s confidence rests on the absence of evidence that capability improvement is slowing.

3x Trillion-Dollar IPOs & the Race to $10 Trillion

The partners are constructive on the wave of mega-cap IPOs approaching the public markets, against a backdrop Miles describes as “an unmistakable concentration of capital around a handful of late-stage private companies,” with nearly $200 billion flowing into just two or three names in Q1 alone. Matt’s priority is durability over near-term volatility, the question that matters to him is “do I think those three companies have generational runs ahead of them, and I think they do.” He expects the debuts could be rocky early but favors being a long-term holder of the basket.

Arun frames the IPOs as a positive for market access, not just returns. He highlights the retail dimension of the SpaceX offering: “Elon reserving 30% of the SpaceX IPO for retail investors I think is a really great thing,” tying it to a broader cultural argument that ordinary investors have been shut out of this part of the cycle until now. He also connects that exclusion to the backlash against AI data centers.

The scale of the prize keeps expanding. Miles notes that 10 years ago no publicly traded company was worth $1 trillion, 5 years ago there were 5, and today there are 14, with several private pre-IPO candidates. His forward view “it’s fair to expect there will be $10 trillion companies and beyond over the next cycle.”

The closing note is applied rather than financial. Arun describes his wife, an ER doctor implementing an AI triage tool where “the early results from implementing this is that it made them 100% more effective.”

Matt’s most-anticipated development over the next 12 months is the same shift, AI value dispersing well beyond the small frontier of power users and outside Silicon Valley entirely.

→ Listen on X, Spotify, YouTube, Apple

The views expressed here are those of the individual personnel quoted and are not the views of Accel Management Co. L.L.C. or its affiliates (collectively, “Accel”). This content may include third-party advertisements; Accel has not reviewed such advertisements and does not endorse any advertising content contained therein. This content is provided for informational purposes only, and should not be relied upon as legal, business, investment, or tax advice. You should consult your own advisers as to those matters. References to any securities or digital assets are for illustrative purposes only, and do not constitute an investment recommendation or offer to provide investment advisory services. Furthermore, this content is not directed at nor intended for use by any investors or prospective investors, and may not under any circumstances be relied upon when making a decision to invest in any fund managed by Accel. An offering to invest in an Accel fund will be made only by the relevant offering documentation of any such fund, which should be read in its entirety. Any investments or portfolio companies mentioned, referred to, or described are not representative of all investments in vehicles managed by Accel, and there can be no assurance that the investments will be profitable or that other investments made in the future will have similar characteristics or results. A list of investments made by funds managed by Accel is available at https://www.accel.com/relationships.

Subscribe now

Share

The material presented on Molly O’Shea’s website are my opinions only and are provided for informational purposes and should not be construed as investment advice. It is not a recommendation of, or an offer to sell or solicitation of an offer to buy, any particular security, strategy, or investment product. Any analysis or discussion of investments, sectors or the market generally are based on current information, including from public sources, that I consider reliable, but I do not represent that any research or the information provided is accurate or complete, and it should not be relied on as such. My views and opinions expressed in any website content are current at the time of publication and are subject to change. Past performance is not indicative of future results.

Paid Endorsement. Brokerage services by Open to the Public Investing Inc, member FINRA & SIPC. Advisory services by Public Advisors LLC, SEC-registered adviser. Crypto trading provided by Zero Hash LLC, licensed by the NYSDFS. Generated Assets is an interactive analysis tool by Public Advisors. Output is for informational purposes only and is not an investment recommendation or advice. See disclosures at public.com/disclosures/ga. Matched funds must remain in your account for at least 5 years. Match rate and other terms are subject to change at any time.

Founders Fund Mafia, Thiel Fellowship, Prompts vs Design, Vibe Mathing & AI's D̶a̶d̶d̶y̶ Trust Issue

Dylan Field Co-Founder & CEO of Figma (NYSE: FIG) joins Sourcery behind-the-scenes at Day 0 of Config 2026, aka the “Coachella for design,” just before 10,000 attendees arrived.

→ Listen on X, Spotify, YouTube, Apple

The headline going around is that “design is dead” & AI has turned it into a commodity. Field’s argument is that the market has this backwards. Anyone can prompt a model into the average, & in a templated world the average does not stand out, so taste + a real point of view are what separate the products people actually use from everything else.

He also points to a shift in who is on the platform. Engineers now come to Figma to code their hearts out.. to create something beautiful. The surface area for design, in his telling, is expanding, not shrinking.

We Cover:

› Vibe Mathing
› Elon Musk: Products vs Logos
› AI's trust issue
› Agents vs Employees
› Engineers coding design
› American Enterprise >
› IQ ≠ Judgment or taste
› What actually is a jailbreak?

Subscribe now

Share

𝐓𝐈𝐌𝐄𝐒𝐓𝐀𝐌𝐏𝐒

(00:00) Dylan Field, Co-Founder & CEO at Figma
(00:57) Day Zero begins: backstage at Config
(01:44) The “Coachella for Design” is real
(02:35) Why Dylan started Config back in 2020
(03:47) Why “design is dead” is completely wrong
(06:11) How Figma turned engineers into designers
(06:54) Inside Config before the crowd arrives
(07:55) The Config speakers Dylan can’t wait to watch
(11:34) Inside the exclusive Pantone Lounge
(12:00) The one question every designer gets asked
(13:16) Getting killed in every single game of Mafia
(15:33) How to escape the permanent underclass of zero taste
(19:51) Why execution is cheap but taste is everything
(22:25) What actually builds a trusted brand
(25:51) How AI agents are rewiring company structure
(28:54) The AI safety debate happening right now
(31:57) Why deep curiosity is Dylan’s secret weapon
(34:53) The real lesson from being a Thiel fellow
(36:52) Rapid fire: Elad Gil’s wildest question
(38:04) What people are missing about AI and enterprise software
(40:16) Where Figma’s platform goes next
(41:21) Why Dylan is bullish on SpaceX
(42:43) Elon Musk’s logo theory
(43:26) The mentors who shaped Dylan Field
(45:20) Alan Kay, VR, and the future of human-computer interaction

Brought to you by:

Brex—The intelligent finance platform: cards, expenses, travel, bill pay, banking—wrapped into a high-performance stack. Built for scale. Trusted by OpenAI, Anthropic, Vercel, Granola, Deepgram, & Sourcery.. teams that move fast AF. visit → brex.com/sourcery

Turing—Turing partners with frontier AI labs to improve model capabilities in coding, reasoning, tool use, & multimodality, as well as with Fortune 500 enterprises to build & deploy end-to-end agentic AI systems in mission-critical workflows Visit: turing.com/sourcery

VCX—VCX is the public ticker for private tech, allowing investors of all sizes to invest in venture capital. View The Portfolio at GetVCX.com

Deel—Deel is the global people platform that helps startups hire, manage, pay, and equip anyone, anywhere. Trusted by more than 35,000 fast-growing companies, Deel is the people platform that just works, so teams can scale without the chaos. Visit: deel.com/sourcery

Public-–Investing platform Public just launched Generated Assets, which lets you turn any idea into an investable index with AI. With Generated Assets, you can build, backtest, refine, and invest in any thesis with AI. Gone are the days of one-size-fits-all ETFs. Try it today: public.com/sourcery

Figma CEO Dylan Field on Why the ‘Market’ Has Design Backwards

I just sat down with Dylan Field, Figma’s co-founder & CEO, at Config 2026, Figma’s annual conference in San Francisco. While the larger debate circling market narratives is whether AI turns design into a commodity.. its safe to say Figma does not agree (and neither do the 10,000 IRL attendees).

→ Listen on X, Spotify, YouTube, Apple

How Config Started

Field said Config began in 2020, right before the pandemic, and that this was technically the 10th one counting the international editions. It grew out of Figma user groups popping up around the world, and the team wanted a way to bring everyone together. The first plan was for a few hundred people, then they went bigger. “Let’s go big, like 800 people. And that sold out instantly,” he said. The timing turned out to be lucky. “We got so lucky the first time because we were able to, three weeks before the pandemic started, had the chance to bring everyone together,” he said, and added, “I can’t imagine what it would be if it was three weeks earlier.”

The contrast between that first Config and this one is worth sitting with. In 2020 the company announced 7 features, and one of the ones the crowd loved most was simply being able to add hyperlinks on the canvas. This year the keynote was a stack of AI launches at once, including Code Layers, Motion, shaders, generative plugins, & an upgraded design agent, aimed at a much broader group of users than the designers who filled that first room.

The Case Against “Design Is Dead”

Field pushed back on the idea that design is dead. He thinks people are getting carried away with AI and talking down every role but their own. “Everyone is kind of putting down all the other roles,” he said, with engineers saying they do not need PMs or designers, and designers saying the same in reverse. He called it a temporary phase where people are excited about AI but “not yet realizing all the limitations that they have.” He also thinks working solo is a trap. “It’s very lonely, the idea that you’re just gonna work alone on everything forever.”

He also said design is more than how something looks. “Design is so much more than visuals and aesthetic.. That’s just the tip of the iceberg.” For him it includes understanding the user and the experience, not just the visuals. “If you wanna build scalable real systems, explore broadly, build the right thing, not just something fast, then I think it’s really important to lean into design.”

The Coders Moving Into Figma

One thing Field kept coming back to is that engineers are starting to use Figma. “We’re seeing so many engineers now just get interested in Figma, and as that happens, they’re starting to really pick it up because you go build something really fast, that’s awesome, but then you’re like, ‘Okay, can I actually now make it great?’” His point is that AI makes it easy to build something fast, and then people want to make it good.

I said it sounded like engineers are coding design, and he agreed. He said the launches coming the next day at Config would go “even more in that direction.” And it goes both ways. “Designers are giving us the feedback that we used to get from developers now.” In the keynote the next morning, he put the idea plainly, that code is not the opposite of design but a material for it.

Why AI Prompts Give You the Average

His main point on AI and design is that if you let a model do everything, you get something generic. “If you just use an LM for everything and have it literally tell you how it wants to go build a thing, how to go design a thing, you’re gonna get something that it wants to give you, and it’s kind of diverged at that point from whatever’s in your head to however it processed it.” The result, in his words, is “you’re gonna get the average, and it’s not going to stand out.” He separates two things people tend to blur. “There’s a taste and there’s sort of the distribution, and they’re very separate,.”

I called that the “permanent underclass of zero taste.” His answer to it is having a point of view. “It takes not only human steering and human intent, but also it takes a point of view, and I think that is something that people fall down on all the time because it’s kinda scary to have a point of view,” he said, adding that it means being willing “to be bold and to take a risk.” He mentioned Figma Weave as one way to steer AI output through a workflow instead of taking whatever the model hands back.

What Risks to Take

I asked what risks people should be taking and what he thinks is overlooked. His answer was broad. “Everything from what problem are you even tackling to how do you brand, what positions do you take as a company, how do you connect with people, and what emotions are you trying to bring through your brand and marketing,” and the same for the software itself.

His warning was that it is easy to copy. “It’s so much easier to apply patterns that you’ve already seen,” he said, and the harder path is figuring out the core thing people actually need and building all the way to it. He was clear that this is a real bet. “There’s just this huge unexplored space out there of stuff that, it takes a long time to get right, and that is a risk because you might not actually hit,” he said. He also said teams have to make an idea stick internally, which is why “you kind of have to meme-ify it inside your company sometimes.”

What Getting Smarter Does Not Fix

Field drew a line between what AI is good at and what it is not. He said smarter models do great on things you can verify, like math, but that other things do not follow automatically. “You can get to amazing outcomes with enough IQ points, and yet judgment is somehow not purely correlated with just adding IQ points to the model… Design certainly does not seem to be correlated with adding IQ points to the model.” He said the same is true of strategy in people. “The folks that are really good at strategy, I have not even found with humans the common characteristic of how to go find them.”

He used empathy as an example. “How do you teach a model true empathy for a user?” describing the harder version as getting a model to understand the relationships a user has and the inflections in their voice when they describe a problem.. “I think it’s really hard.”

This is also why he does math with models for fun, which he jokingly called “a form of LLM psychosis.” He said it is “a bit of an antidote to the non-verifiable domains that I’m usually in. So for Figma, it’s all vibes.” His example of an easy, verifiable miss was blunt. “You didn’t center the div, like learn to center divs AI.”

The Jailbreak Question

Field said the biggest open question in AI right now is what a jailbreak even is. He mentioned a live situation at the time between the US government & Anthropic over the Mythos / Fable jailbreak, and said the term is understudied.

“What is a jailbreak? It’s not like you have to do some fancy JSON thing.. Just talking with models long enough leads to states that you could call a jailbreak for basically any model.” He also said the conversation cannot really happen without a shared definition. “I don’t think we can have a conversation like this without even defining what a jailbreak is.”

Field thinks there should be more incentives for people to test models, similar to how Hacker One pays researchers to find security flaws. He said he now spends more time on interpretability research with open weight models when he can get the compute.

Agents vs Employees

I brought up the shift in how people talk about AI, from job loss to productivity, and the idea of having more agents than employees. Field was cautious about the framing. “It’s not even clear how do you define the unit of an agent.”

His view is that humans should stay in the loop. “Humans being in the loop will be the way that you steer towards the best outcomes.” He noted that people at Figma who understand their infrastructure are getting strong results by having agents break up tasks. “They’re already 10X, & they can 10X themselves again, by having agents really break up tasks properly,” but the steering still comes from people.

Nevertheless, Figma is on a Tear

137 Venture’s Christian Garrett shared a few great questions for Dylan, starting with the obvious tension: Figma is on a tear but the markets are punishing it, what are people missing about AI in enterprise software?

Field’s answer was about what makes software durable. Network effects, including marketplace liquidity, which he counts as its own kind. The mix of customers a company has. And software people might call boring. “Boring software is not to be undercounted.”

On top of that, he kept coming back to company/product/user data. “The data liquidity that you can establish & how that can create context, how that context can create capability in a system, that is its own flywheel.” The more a team uses a tool, the more useful it gets for them.

He certainly does not buy that software is dead. “A lot of software will do just fine or actually even amazing.”

The numbers back the point. In its most recent quarter, reported in May, Figma grew revenue 46% year over year to $333 million, its second straight quarter of accelerating growth, and beat expectations. Net dollar retention hit 139%, its highest in more than 2 years, which means existing customers spent about 39% more than a year earlier. Paid customers grew 54% year over year to roughly 690,000, accounts spending more than $100,000 a year rose 48%, and entry-level Pro signups more than doubled. Weekly active use of the Figma feature that lets AI coding agents work directly with Figma files grew 5x in a single quarter. Figma raised its full-year revenue guidance to about $1.42 billion.

That is the case against design being a commodity. The canvas where designers & engineers actually work is not something you can pull out of their hands. It lines up with what he told me at Config, that execution is cheap now and design & creativity are what separate the wheat from the chaff.

The same point showed up elsewhere the same week. On CNBC, Palantir CEO Alex Karp went on a tear arguing that American enterprises are smart, durable buyers who are not being fooled by AI pricing that charges for tokens while delivering little in return. He said companies are livid about paying for usage that creates no value and handing over their data in the process. "This is the voice of American business that is being channeled through me," Karp. His read is that serious buyers can tell the difference between a tool that compounds value and one that just runs up a bill, which is the same reason a product people actually build in keeps expanding inside its accounts.

Still Stuck in the Prompting Age

Field said he is surprised we still mostly use AI by typing prompts. “I’m still shocked that we’re in a prompting age. It’s been, what, 5 years since GPT-3 now,” he said. He expects the way we interact with models to change a lot, and noted “there are other ways to explore latent space.”

He thinks software design has been stuck too. He pointed out that the last interaction pattern people can name is pull to refresh, and asked “if you look at an iPad, why is it a big iPhone?” He said he wants to get back to the kind of thinking from the 1960s through 1980s and mentioned Alan Kay, one of the computer scientists behind early ideas about personal computing. The main place he sees new ideas right now is VR, and he said “the future of design and interactivity is really gonna be fun.”

On where Figma goes, he wants to bring its different tools together so people do not have to think about where to start. “There’s a huge opportunity in doing that in one place and not having to think about where do I start.” He wants to keep expanding what it does for designers and to help people build their own tools on the platform.

AI’s Trust Issues

I told him AI has knocked down a lot of trust in tech, worse than ICE by some measures, and asked what builds a trusted brand. Field said feelings about AI are split. On one end are people who feel real affection for a model. “In the case of a 4o or Claude, in love with the model.” On the other are people worried it will take their job. On building trust, he said it comes down to “consistency over time” and showing up in a way people can expect.

He brought up a line from his first internship at O’Reilly Media, from Tim O’Reilly, “create more value than you capture,” and tied it to what a platform is. “The definition of a platform is that you’re creating even more value outside of the platform than you are from the platform itself.” He also said the mood needs to change. “The doomerism has to stop as well, and it has to stop with good rationale,” and it should be replaced with a plan to “paint an optimistic future vision for humanity.”

On Elon, Logos, & Brand

I asked about Elon Musk’s line that a “special enough product needs no logo.” Field mostly agreed, then added that a product that special usually has enough of a brand that it works like a logo anyway.

“If the product is that special, then I think it’s probably got enough brand characteristics that it kind of functions as a logo.”

His example was Tesla. Take the badge off and “you probably still know it’s a Tesla.” His broader point was that “a logo’s just a part of it.”

Why He’s Bullish on SpaceX

Field said SpaceX is a good example of a company with missionary talent that keeps taking on hard problems. “Amazing talent who are really missionary,” describing a team that “quests and conquers after challenge, after challenge, after challenge.”

His main point was about imagination. “People lack imagination when it comes to what they can do on Earth, and also what they can do in the stars.” He thinks it could be one of the most important companies of this century or the next.. “I hope Figma can follow too.”

The Speakers He’s Excited About

Field named two speakers he was especially looking forward to. The first was Holly Herndon, the experimental musician who built Holly+, an AI clone of her own voice that she released around 2021 for other people to sing with. He called her “a total visionary artist” and said she sees things early. “The timescale that she’s operating on is different than the rest of us,” pointing to how she was working with AI voices and NFT-style ideas before most people were.

The second was Grant Sanderson, who makes the 3Blue1Brown math videos on YouTube known for their animations. Field called him a friend and praised the care he puts into explaining math with graphics. Sanderson “really uses graphics and design to be really craft forward in his explanations.”

Murdered on Founders Fund’s Mafia

Field recently appeared on Founders Fund’s Mafia show, the game show created by the firm’s CMO Mike Solana where founders and investors play the social deduction game Mafia. It filmed at Tosca Cafe in San Francisco, the North Beach spot from the original PayPal Mafia photo, and the wider cast included Sam Altman, Palmer Luckey, Bryan Johnson, Moxie Marlinspike, Trae Stephens, & Cyan Banister.

Field did not do well, and he was upfront about it. “Game one where I was killed by the mafia the first night,” and in game three he “got killed by the village first night.”

He said Palmer Luckey was the standout because he gave nothing away. “He was just the same energy throughout.” He also said he mostly prefers to moderate now, because winning and losing both feel bad. “I figured out that it was a lose-lose.” As for what the game teaches, he was blunt. “There’s not much to learn about human psychology from Mafia,” but beyond that “people’s attention spans are short.”

What the Thiel Fellowship Gave Him

Field was part of the Thiel Fellowship, the program Peter Thiel started that pays young people to skip or leave college and build something, with the early days run by Danielle Strachman, now of 1517 Fund. Field joined in 2012. When I asked what he got out of it, he said the money was not the main thing. “100K over two years, amazing, but the community was just even more special,” and called the learning from that group “off the charts.”

The reason was simple. “It’s just so lonely to start a company when you’re younger.” He talked about interning at Flipboard where his coworkers were great but a decade older.. having a group of peers to see each week mattered more than he expected.

The class itself is a good argument for that point. Field’s 2012 cohort, around 20 people, included Chris Olah, who went on to co-found Anthropic, the AI company most recently valued at $965 billion in its May 2026 funding round. It also included Noor Siddiqui, who founded the genetic-testing startup Orchid, which has raised about $16.5 million to date. Field built Figma out of the same group.

Who Taught Him, & Why He Stays Curious

Field said he has learned from people all over, including mentors from before Figma at O’Reilly, LinkedIn, and Microsoft Research, his board, and his own employees. Some of them do not even know it. “There are people that have mentored me & have no idea.. They’re voices in my head.”

He singled out Figma’s first manager hire, Sho, as director of engineering. “He really taught me management. I didn’t know anything. I was an intern before Figma.” But also surprisingly Figma’s customers.. “We are very lucky as a design company to have designers as users and customers because they are trained to give good feedback.”

When I asked why he stays so curious about everything, he kept it simple.

“I just think the world is amazing.”

→ Listen on X, Spotify, YouTube, Apple

Subscribe now

Share

The material presented on Molly O’Shea’s website are my opinions only and are provided for informational purposes and should not be construed as investment advice. It is not a recommendation of, or an offer to sell or solicitation of an offer to buy, any particular security, strategy, or investment product. Any analysis or discussion of investments, sectors or the market generally are based on current information, including from public sources, that I consider reliable, but I do not represent that any research or the information provided is accurate or complete, and it should not be relied on as such. My views and opinions expressed in any website content are current at the time of publication and are subject to change. Past performance is not indicative of future results.

Paid Endorsement. Brokerage services by Open to the Public Investing Inc, member FINRA & SIPC. Advisory services by Public Advisors LLC, SEC-registered adviser. Crypto trading provided by Zero Hash LLC, licensed by the NYSDFS. Generated Assets is an interactive analysis tool by Public Advisors. Output is for informational purposes only and is not an investment recommendation or advice. See disclosures at public.com/disclosures/ga. Matched funds must remain in your account for at least 5 years. Match rate and other terms are subject to change at any time.

The Golden Rules Are Gone

Everett “Ev” Randle, General Partner at Benchmark, joins Sourcery to break down how AI has rewritten the rules of venture and growth investing.

→ Listen on X, Spotify, YouTube, Apple

Ev explains why the old inverse relationship between scale and risk has collapsed, why the golden rules of SaaS now run in reverse, and how he underwrites AI companies when the spreadsheet no longer does the work.

We cover the new AI taxonomy through a P x Q x M lens, the economics of inference and agents, Claude Code at $36,000 per developer, the AI mom test for frontier versus open source demand, and the trillion-dollar question of whether frontier labs hold pricing power. Ev also walks through the coming liquidity shock from the $380B Anthropic round, why late-stage can now beat a Series C, and how venture firms became alternative asset managers.

Benchmark's AI portfolio spans Cerebras, Sierra, Fireworks AI, Legora, Mercor, HeyGen, Decart, Manus, Gumloop, Exa, StarCloud, Reducto, Eigen, 11x, and LangChain

Subscribe now

Share

𝐓𝐈𝐌𝐄𝐒𝐓𝐀𝐌𝐏𝐒

(00:00) Everett Randle, General Partner at Benchmark
(00:58) Coming off Benchmark's AGM
(04:05) The Golden rules of Investing are all gone
(08:58) Who actually has a handle on AI Economics?
(12:57) Why Benchmark bets on Founders, not Categories
(15:31) Brad Gerstner's "Age of Inference" Thesis
(19:07) The most important shift since the Cloud
(23:55) Inside Gumloop's AI automation canvas
(26:40) The Token Maxing problem nobody's solving
(27:25) Ev's "Mom Test" for frontier AI
(31:33) What happens when frontier models get too cheap
(34:44) Inside the new funding playbook
(40:31) Venture Capital became a product, not a firm
(42:35) The Biggest IPO wave Wall Street's ever seen
(49:08) The secret behind Benchmark's wildly diverse bets
(52:54) The mentors who shaped him

Brought to you by:

Brex—The intelligent finance platform: cards, expenses, travel, bill pay, banking—wrapped into a high-performance stack. Built for scale. Trusted by OpenAI, Anthropic, Vercel, Granola, Deepgram, & Sourcery.. teams that move fast AF. visit → brex.com/sourcery

Turing—Turing partners with frontier AI labs to improve model capabilities in coding, reasoning, tool use, & multimodality, as well as with Fortune 500 enterprises to build & deploy end-to-end agentic AI systems in mission-critical workflows Visit: turing.com/sourcery

VCX—VCX is the public ticker for private tech, allowing investors of all sizes to invest in venture capital. View The Portfolio at GetVCX.com

Deel—Deel is the global people platform that helps startups hire, manage, pay, and equip anyone, anywhere. Trusted by more than 35,000 fast-growing companies, Deel is the people platform that just works, so teams can scale without the chaos. Visit: deel.com/sourcery

Public-–Investing platform Public just launched Generated Assets, which lets you turn any idea into an investable index with AI. With Generated Assets, you can build, backtest, refine, and invest in any thesis with AI. Gone are the days of one-size-fits-all ETFs. Try it today: public.com/sourcery

Merge—The leading provider of customer-facing integrations and agentic tools for frontier LLMs, Fortune 500 organizations, and B2B SaaS companies. Visit: https://merge.dev

The Death of Spreadsheet Investing: Inside Benchmark’s New AI Playbook

Ev Randle, General Partner at Benchmark, joined Sourcery to lay out how AI has broken the underwriting logic that governed software investing for two decades. The conversation moved from why the market feels disorienting, to the new taxonomy he uses to evaluate AI companies, to the liquidity shock coming from a pipeline of trillion-dollar IPOs.

Benchmark’s AI portfolio referenced across the conversation includes Mercor, HeyGen, Decart, Manus, Gumloop, Exa, StarCloud, Reducto, Eigen, Cerebras, 11x, Sierra, Fireworks AI, Legora, and LangChain.

→ Listen on X, Spotify, YouTube, Apple

The Disorientation: Scale No Longer De-Risks a Company

Randle’s starting point is a single inverted relationship. In the prior software paradigm, scale and the risk of mega-impairment moved in opposite directions. As a startup grew, it sequentially de-risked product market fit, then unit economics, then total addressable market, then market leadership. If a company failed to clear those gates, it stopped growing. Bigger therefore meant safer.

AI broke that relationship. “You can have businesses that are well over a billion dollars in revenue that haven’t proven out their unit economics,” Randle said. “The risk of impairment over time is actually sort of flat, or even maybe there’s a weird positive correlated relationship with scale and risk.”

The practical consequence is that investors can no longer read scale as a proxy for safety. A company past $1B in revenue may still carry unresolved questions on margins, durability, and differentiation that scale used to answer on its own.

What Happened to the Golden Rules?

For roughly 20 years, the most attractive software companies shared a predictable profile to measure ‘greatness’ off of: gross margins of 70% to 90%, pure software with minimal services load, high gross retention, low capital intensity, and operating leverage on R&D.

A company with 100 customers and 90%-plus gross retention would lose fewer than 10 of them a year. The output was a capital-light business that generated high free cash flow margins growing at a multiple of GDP for a long time.

That profile was legible and, in Randle’s word, “spreadsheetable.” It produced shorthand metrics like rule of 40, where growth rate + free cash flow margin could compress company quality into a single score. “Sometimes you could literally abstract the quality of a software company into a single metric, and some investors would invest just on a rule-of-40 score,” he said.

AI Inverted Nearly Every Rule

The most popular AI companies & categories now look almost like the exact inverse of their expected technology investment profile.

Distribution shifted from product-led growth to forward-deployed engineers, what Randle called “the Palantirification of everything.” That reintroduces services and implementation load that pure-software orthodoxy treated as a defect.

Gross margins flipped sign. “Now gross margins, if your gross margins are high, that’s actually a bad thing, because AI inference costs a lot of money, & if you have an AI product with high gross margins, that means that no one’s using your AI features.”

Capital intensity returned. To stay defensible against the foundation model labs, companies are increasingly expected to train or post-train their own models, which requires GPU spend that the old software model never carried. “All of the golden rules of the past that defined the spreadsheet investing era are all gone,” Randle said. “The most popular companies and categories are almost the inverse of all these golden rules.”

His point is not that the golden rules were wrong. In a vacuum they remain the first-principles building blocks of a high-quality company, since durable free cash flow over a long horizon is still what earns a high valuation. The difficulty is that the most popular AI companies look less attractive on those first principles than classic SaaS, which forces a rethink of how to underwrite them.

The New Taxonomy: “P” x “Q” x “M”

Randle’s organizing framework is P x Q x M, where P is price, Q is quantity, and M is margin. In SaaS, P was annual contract value, Q was the customer base or addressable market, and M was a gross margin of 70% to 90%.

In AI, the 3 variables move in different directions. Q, the quantity of customers, is roughly the same population that would buy SaaS. M, the margin, is almost always lower. P, the price, can be far higher, because inference platforms now sign nine-figure contracts with startups, a contract scale that was rare in SaaS even with the largest enterprises. In Randle’s words:

“In SaaS land, you had price was your ACV, what is the annual contract value of your contracts that you sell to customers. Your Q is how many customers either do you have or that are in your TAM, and then your M was gross margin, which was 70 to 90%.

Well, now in AI, if we take an AI app company, the Q is probably the same, you’re selling to the same people that would buy a SaaS. The M is almost definitively lower, for I think 99% of AI app companies it’s lower than 70%. But the P can be immensely high.

You have these inference platforms that have nine-figure contracts with startups. There’s very rare SaaS companies that have nine-figure contracts with anyone, much less a startup.”

He also stressed that AI companies are far less alike than software companies were. Robert Smith of Vista Equity Partners, where Randle started his career, used to tell the team that “software tastes like chicken, and that’s why it’s beautiful,” because every software P&L converged on the same line items at maturity. That convergence no longer holds.

Randle contrasted two Benchmark-adjacent inference names to make the point: Fireworks leases inference capacity & GPUs and monetizes the software layer that reduces cost and latency on top, while Crusoe builds data centers, acquires power, land, and permits, and operates a different business model with a different margin profile and capital intensity entirely. From the outside they look like 2 inference companies, however, they are more different than alike.

Inference Is the Demand Engine

Randle’s framing for the current market is to “walk under the waterfall.” Working with a portfolio company that had strong developer usage but no settled business model, he told them to stop sitting on the riverbank trying to fill a bucket and instead get under the waterfall, because the waterfall is inference. The volume of revenue and demand flowing out of inference, he argued, cannot be ignored.

He traced the new growth curves, companies going from 1 to 20 to 100 rather than the old 1 to 3 to 9 to 20, back to inference-enabled business models. Instead of charging a dollar amount per seat, companies now charge a margin on the inference they resell. Some models are thin, closer to brokering inference. Others abstract it heavily, such as Sierra’s outcome-based pricing on completed customer support deflections. In both cases the monetized unit is inference, which removes the rate limiter on revenue growth.

The clearest data point came from coding agents. Randle said developers were spending $3,000 per month each on Claude Code, or $36,000 per developer per year. In SaaS, a $50,000 ACV was a solid overall contract. Now that figure can recur per developer and keep growing. He framed the shift from a $200K line item for the average company toward a potential $20M line item, with some customers far higher.

He named the agent economy as the most important product and business-model shift in technology since the start of SaaS and the cloud, while noting the term itself has been over-marketed to the point of being “cooked.”

Frontier vs Open Source: The AI “Mom” Test

To gauge where frontier demand holds, Randle applies what he calls his AI mom test. The question is how many tasks a non-technical user actually needs from a frontier or near-frontier model rather than a cost-effective open-source one. Two years ago the answer was unclear. Today, he said, essentially none of his mother’s queries require the frontier (damn). Layering that logic across user types reveals a growing share of economy tasks that do not need frontier intelligence.

He cited Cognition’s published work, which post-trained an open-source model on low-complexity tasks that were popular inside the product but did not require frontier capability. Moving those tasks off the frontier produced roughly 95% savings on that action.

The conclusion is not that the frontier loses. Demand for frontier intelligence is also growing, and Randle attributed the parabolic growth in Claude Code and Anthropic revenue to a genuine breakthrough in coding model quality around Opus 4.5.

Citing his partner Eric Vishria, he framed the market as non-zero-sum: on-device inference, open-source inference, and proprietary models all show rising demand at once. The open question is pricing power. If capabilities keep climbing toward recursive self-improvement, frontier labs retain the ability to charge a premium. If capabilities plateau and distillation lets open source reach 95% of the ceiling, premium margin compresses, even if the labs survive on product strength, since most of ChatGPT’s 900 million weekly active users could not identify which model they are using.

*DISCLOSURE: This could also be your dad, your aunt/uncle, your dog/cat/fish, or your grandparents, roommate or baby. We are well aware of some hard core tokenmaxxing moms out there. Damn Ev.

Liquidity Shock: Reframing the Returns Maths

Randle’s most concrete data exercise concerned the coming IPO wave. He charted 4 of the best pre-IPO rounds of the past decade, Slack, DoorDash, Snowflake, & Nubank. Those rounds were typically $500M to $2B in size and returned 2x to 5x over a 4-year period. The Snowflake pre-IPO round turned roughly $500M into about $2.5B, an outcome that left everyone involved satisfied.

He then applied the same math to Anthropic’s $380B round. If Anthropic reaches liquidity at $1T to $1.5T, a valuation he characterized as not particularly aggressive given the round priced near $1T, the $30B raised in that round would return roughly 35 times the Snowflake pre-IPO round. “It’s 35 Snowflake pre-IPO rounds in a single round,” he said.

The scale is hard to internalize. 5 years ago, before COVID inflated fund sizes, a normal growth fund was about $1B. Randle noted he knows individuals with $3B to $4B invested in Anthropic who could return 5x in under 5 years.

He flagged the second-order effects:

• new company formation

• reinvestment decisions

• the San Francisco housing market, where he said homes are going for 2x asking price in cash or in lab equity (!!)

The unknown caution is that the ecosystem is not priced for the liquidity this pipeline will release.. and no one knows what will result because of it.

However, this also tied the wave to two structural shifts:

1. Companies stay private far longer, supported by deep venture-growth capital, so businesses that would have been public for years by 2005 standards remain private

2. AI has introduced large day-one costs, where a research direction might require $2B of compute before product market fit is even testable, a different funding equation from a $500K seed.

The combination produces what he called company rebirths, where a late-stage company can carry higher upside than a Series C. His own first investment at Kleiner Perkins was SpaceX at a $100B-plus valuation, where the Starlink broadband business, now the majority of revenue in the S-1, became a second growth engine that justified entry at a triple-digit-billion price.

Founders Over Themes

Against that backdrop, Randle described Benchmark’s strategy as entrepreneur-out rather than theme-in. The firm partners early, often at inception around a $50M post valuation, which sidesteps many late-stage questions about exit multiples and capital efficiency. “Great founders are always in style,” he said, “whereas these business models can go in and out of style.”

That discipline shows in a portfolio that looks thematically diverse but was not assembled thematically:

• Semiconductors: Cerebras

• Vertical AI: Legora

• Horizontal: Sierra

• Developer: LangChain

• Prosumer: HeyGen

• Orbital data centers: StarCloud

But, he said that coherence was retrospective. The investments were founder-driven, several were pivots, and some landed in categories the firm had not set out to back. Chetan Puttagunta closed the StarCloud investment weeks before Elon Musk began publicly championing orbital data centers, on the strength of a team that already had a GPU operating in space, not on a category call.

Gumloop, Randle’s first Benchmark deal, fits the same pattern. It is a collaborative AI agent and automation canvas for enterprises, letting every employee, not only developers, build and run agents across functions. The thesis is that what happened in code, where Opus 4.5-class agents let developers offload a bulk of their work, extends to most white-collar and eventually blue-collar functions.

Randle also argued for the value of an independent third-party vendor in a world where models are jagged and good at different things, so customers need a router watching spend rather than checking the weather with a frontier model.

Venture Capital as a “Product,” Not a “Firm”

Randle closed on a distinction the industry has not kept up with linguistically. Many firms still described as venture capital have become alternative asset managers. General Catalyst, Thrive Capital, and Andreessen Horowitz run venture, growth, debt, and additional products, and Randle put a name to the shift. “Venture in many ways is still the same, but it’s a product now for many of these firms. It’s not the firms themselves.”

The PitchBook data shows how concentrated the capital behind that shift has become. In Q1, 73.1% of LP commitments went to 5 firms, and 6 megafund managers absorbed 76.2% of the quarter’s capital. When a small group of multi-product managers takes the bulk of new commitments, the economics of the business tilt toward gathering and deploying assets at scale rather than the small, early, concentrated bets that defined classic venture.

The deal side mirrors the same crowding. PitchBook put Q1 deal value at $267.2B, but stripping out the top 5 deals cuts that figure by 73.2%. AI took 88.8% of the dollars on 42.5% of deals. When 5 companies effectively are the market, the asset-gathering model and the venture model start to look like different businesses wearing the same label.

Benchmark sits outside that structure by design. It remains a small equal partnership rather than a multi-product manager, and it does not appear among the megafunds absorbing the quarter’s commitments. PitchBook has described the current environment as an era of consensus deals, with dry powder piling into the same perceived winners. Randle’s argument is that staying small & early is a feature, not a gap. The firm does not need to rebrand around whatever category is in style, because the constant it underwrites is the founder, & a concentrated early-stage partnership is built to make non-consensus bets precisely when capital is crowding the same names.

→ Listen on X, Spotify, YouTube, Apple

Subscribe now

Share

The material presented on Molly O’Shea’s website are my opinions only and are provided for informational purposes and should not be construed as investment advice. It is not a recommendation of, or an offer to sell or solicitation of an offer to buy, any particular security, strategy, or investment product. Any analysis or discussion of investments, sectors or the market generally are based on current information, including from public sources, that I consider reliable, but I do not represent that any research or the information provided is accurate or complete, and it should not be relied on as such. My views and opinions expressed in any website content are current at the time of publication and are subject to change. Past performance is not indicative of future results.

Paid Endorsement. Brokerage services by Open to the Public Investing Inc, member FINRA & SIPC. Advisory services by Public Advisors LLC, SEC-registered adviser. Crypto trading provided by Zero Hash LLC, licensed by the NYSDFS. Generated Assets is an interactive analysis tool by Public Advisors. Output is for informational purposes only and is not an investment recommendation or advice. See disclosures at public.com/disclosures/ga. Matched funds must remain in your account for at least 5 years. Match rate and other terms are subject to change at any time.

Brex—The intelligent finance platform: cards, expenses, travel, bill pay, banking. Built for scale. Trusted by OpenAI, Anthropic, Vercel, Granola, Deepgram, & Sourcery.. teams that move fast AF. visit → brex.com/sourcery

ElevenLabs Homecoming: Inside the Warsaw Summit & Poland’s $1T Rise

Pictured: Charlotte + Molly (Warsaw Nov 2025)

Live from Warsaw, on June 1st, the ElevenLabs “Homecoming” Summit doubled as a showcase for Polish Dynamism, as Poland crossed into the world’s top 20 economies, locked in historic defence funding, and kept building toward one of Europe’s most interesting tech hubs. Poland intends to back its founders, weaponise its geography, and sit at the frontier of AI.

ICYMI: ElevenLabs Polska “Homecoming”

Hi, it’s Charlotte from Sourcery. 👋 I’m Molly’s head of Special Ops, and a few weeks ago I embarked on a very Special Operation in Warsaw, Poland.

ElevenLabs hosted their biggest summit yet, their “Homecoming,” at Teatr Wielki, Poland’s National Opera, in the city where the company started. I was on the ground to make sure you don’t miss anything. Around 2,500 founders, leaders, institutions, researchers, investors, artists, and more came from across Poland, Europe, and beyond, and ElevenLabs got all of them in the room.

President Karol Nawrocki opened the event. The Deputy PM and Defence Minister, Władysław Kosiniak-Kamysz, spoke, as did former president and Nobel laureate Lech Wałęsa, Bolt’s Markus Villig, and filmmaker Darren Aronofsky.

ElevenLabs’ own board members flew in too: Andrew Reed of Sequoia, Jennifer Li of a16z, and Seth Pierrepont of ICONIQ, alongside executives from LOT, Polsat, InPost, and PKO Bank, several of Poland’s legacy companies.

It was really 2 stories at once. ElevenLabs, founded in Warsaw in 2022, is now valued at $11B and is one of the most consequential AI businesses in the world. And Poland, the country it came from, crossed $1 trillion in nominal GDP last September, sits on the cusp of the G20, and is growing faster than almost anyone in the EU. The summit put both side by side.

Subscribe now

Share

Back Where It All Began

On June 1st the summit opened at Teatr Wielki, one of Poland’s most significant cultural institutions. Built 150 years ago, it remains one of the largest opera stages in Europe, and was nearly destroyed during WWII before being rebuilt.

The parallel to the company was hard to miss: a venue defined by its technical core. The auditorium has a 1,150 m² stage, a flying system with more than 100 bars for scenery, and 6 two-level hydraulic traps. Engineering in service of performance, not unlike the Met. A 150-year-old institution where art and machinery meet, now hosting a company doing the same with the human voice.

Warsaw holds a special significance for ElevenLabs, not only because both Piotr + Mati grew up there, but because it is where the early days of the company took place.

“My co-founder Piotr built our first ever model right here in Warsaw. It was the first human-like model that was contextually aware, emotional, and actually crossed the uncanny valley of speech.”

The Polish Engineer

There’s a running joke that Poland produces the world’s best engineers, a reputation built on serious technical education and a healthy dose of grit and ambition.

Here are some of the biggest Polish tech founders:

• OpenAI: Wojciech Zaremba, a Polish-born mathematician and computer scientist, was one of OpenAI’s co-founders, while Jakub Pachocki has become one of the company’s leading researchers.

• Snowflake: Marcin Żukowski, a Polish co-founder, helped build technology that became foundational to Snowflake’s data infrastructure.

• ICEYE: Rafał Modrzewski, educated in Warsaw and at Aalto, leads one of the world’s leading SAR satellite companies, which raised a €200 million Series E in December 2025 at a €2.4 billion valuation.

• Verkada: Polish-born co-founder and CEO Filip Kaliszan built Verkada into one of the most valuable companies in physical security, with its December 2025 round valuing the company at around $5.8 billion.

• Fireworks AI: Paweł Garbacki, educated at the University of Warsaw, was one of the PyTorch engineers who co-founded Fireworks AI, which raised a $250 million Series C at a $4 billion valuation in October 2025.

• Klarna: Sebastian Siemiatkowski remains one of the most visible founders with Polish roots in European tech.

• ElevenLabs: Mati Staniszewski and Piotr Dąbkowski, and key members of their team including CFO Maciej Mylik, came up through Warsaw, making the company’s return feel symbolically resonant.

All the People in the Room

President Karol Nawrocki, one of the company’s early users, vowed to continue his support for emerging Polish technological growth, backing Polish founders and innovation as he opened the event.

The Deputy PM and Defence Minister put AI at the centre of Polish security rather than at its edge. He didn’t soften the backdrop. These are scary times, he told the room, with the war next door and a volatile neighbourhood, and Europe can’t afford to treat AI as optional. He pointed to Poland’s role in DIANA, NATO’s Defence Innovation Accelerator for the North Atlantic, the Alliance’s deep-tech engine for dual-use startups working on AI, autonomy, and quantum. The message: AI integration across Europe is now a matter of security, not just productivity, and Poland intends to be at the front of it.

Executives from Poland’s legacy institutions, Polsat Plus Group, LOT Polish Airlines, InPost, and PKO Bank, took the stage on how they’re deploying AI. ElevenLabs announced a partnership with LOT Polish Airlines to overhaul customer experience with voice agents. Mati spoke with InPost’s Rafał Brzoska (on Day 2) about reinventing parcel delivery across Europe, and with Piotr Żak about positioning Polsat Plus Group for the decades ahead.

$3.3B, $6.6B, $11B..

Back in February, Molly asked Mati about the significance of the number 11. The answer is that it’s everywhere in ElevenLabs’ identity: Mati’s birthday, Polish Independence Day (November 11), a play on the pause sign, mathematical properties, even Apollo 11. It shows up in the valuations too.

The company’s marks have come in multiples of eleven:

• $3.3B Series C in January 2025 (co-led by a16z and ICONIQ Growth)

• $6.6B employee tender offer that September (led by Sequoia and ICONIQ, a secondary sale rather than a fundraise)

• $11B Series D in February 2026 (led by Sequoia, with Andrew Reed joining the board).

If the pattern holds… $22B coming soon..

ElevenLabs was founded in 2022 and hit roughly $330M ARR by the end of 2025. By the end of April 2026, 4 months later, it had crossed $500M ARR.

The Series D first closed in February 2026 at the $11B valuation. A later close in May 2026 added BlackRock, Wellington Management, Salesforce, & Santander, joining Nvidia, which had invested the previous September. That is 4 years from the first model built by hand in Piotr’s Warsaw apartment to a Series D at $11B, and it makes ElevenLabs one of Europe’s most valuable AI companies.

Poland Takes a Stake in ElevenLabs

On June 17th, a few weeks after the summit, Vinci, the investment arm of state development bank BGK, took a minority stake in ElevenLabs. The investment was about 42 million zloty, or $11M, putting the Polish state on the cap table alongside a16z, Sequoia, and ICONIQ. The size of the stake was not disclosed.

The investment was paired with AI Lab Poland, a national initiative to give early-stage Polish AI startups access to capital, expert networks, and connections through ElevenLabs. The stated goal is to build the domestic AI ecosystem, using ElevenLabs as the reference case.

ElevenLabs is also expanding locally. The company plans to grow its Warsaw team from around 60 to more than 200 over the next 18 months to serve customers across Central and Eastern Europe, and Staniszewski said the investment will go toward regional expansion. Finance Minister Andrzej Domański framed the move as part of a push for Poland to create technology rather than only use it.

Polish Economic History in a Nutshell

A bit of context for readers less familiar with the story (or who did not take Polish history):

In 1989 to 1991, Poland’s economy went through “shock therapy” under the Balcerowicz Plan, switching rapidly from a planned economy to capitalism. Those transition years brought hyperinflation, mass privatisation, and a collapse in output. But the same period set the foundation for everything that followed: a high-energy, gritty, deeply ambitious push to rebuild Poland after years of occupation and nearly half a century as a Soviet satellite state.

Polish economic growth in key points:

• Around 1991, GDP growth was still negative as Poland absorbed the shock of transition.

• By 2000, growth had stabilised in the mid-single digits, and from 2000 to 2019 Poland expanded almost every year, including through the 2008 to 2009 crisis, when it was the only EU economy to avoid a recession.

• In 2020, during COVID, GDP fell by about 2%, then rebounded by nearly 7% in 2021, with continued positive growth in 2022 to 2025 (around 3% to 5%).

• 2025 closed with growth of 3.6%, among the fastest in the EU, and the IMF forecasts roughly 3.3% for 2026.

Why Now?

Poland’s geography was historically one of its biggest weaknesses, the cause of repeated partitions and occupations, most recently during WWII, when its territory was divided between the Third Reich and the USSR. Today, the same location is its biggest advantage. It sits at the hinge between Western Europe and Ukraine, Scandinavia and the Balkans, EU supply chains and NATO’s eastern front.

What makes it work, combined with geography:

• EU-scale capital.

• A large, educated workforce feeding off strong technical universities, the source of the “Polish engineer” talent pipeline.

• A new wave of ambition and innovation, a founder generation born into a capitalist Poland.

Defence is the other engine. Under the EU’s SAFE programme, Poland secured €43.7 billion in low-interest defence loans, the single largest national allocation under the €150 billion facility. The funds are earmarked for priority defence investments including air defence, anti-drone systems, artillery, and border infrastructure, and the first payment of €6.6 billion (15% of the total) landed in May 2026.

Poland is also pushing at physical infrastructure. Varso Tower, the Foster + Partners skyscraper near Warsaw Central Station, topped out at 310 metres in 2021 and was completed in 2022, making it the tallest building in the EU. Even bigger is Port Polska (formerly the Central Communication Port, or CPK), the Foster + Partners greenfield mega-airport and high-speed-rail hub rising between Warsaw and Łódź, the largest infrastructure project in Central and Eastern Europe and one of the most ambitious in Europe, with a roughly €30 billion price tag. Opening in 2032, it’s built to handle 34 to 44 million passengers a year (scalable toward 100 million), anchored by a 480km high-speed rail line linking Poland’s largest cities.

A Foster + Partners architect, who worked on both buildings, spoke during the summit.

The one obvious shortcoming is energy: Poland still has a significant transition to make to reduce its dependence on legacy fuels. But even that weakness is now forcing investment into new infrastructure and, increasingly, into tech-enabled efficiency and grid management.

Bolt’s Markus Villig x Sequoia’s Andrew Reed

Markus Villig, the Estonian founder & CEO of Bolt, sat on a panel with Sequoia’s Andrew Reed. Reed is Sequoia’s lead on both companies. He led the firm’s investment in Bolt and sits on its board, having backed the €600M Series E in 2021 and the $709M round Sequoia co-led in January 2022 at an $8.4B valuation. He also led Sequoia’s $11B Series D into ElevenLabs in February 2026 and joined that board.

Bolt has a direct Poland connection. Villig founded the company in Tallinn in 2013, originally as Taxify, and it now has more than 100 million customers across 45+ countries and 400+ cities in Europe and Africa. Warsaw is Bolt’s second-largest office and one of its main engineering hubs, running Delivery, Platform, and Data Engineering teams. Villig noted that the Polish engineers Bolt hires in Warsaw are central to how the company builds.

The panel also covered the mechanics of scaling a mobility business: barriers to entry, product differentiation, and the question of a future IPO.

Darren Aronofsky on Creating with AI

Filmmaker Darren Aronofsky sat with ElevenLabs’ Alex Haskell on creating with AI. In May 2025, Aronofsky launched Primordial Soup, a creative studio built to use generative AI in filmmaking, announced at Google I/O alongside a partnership with Google DeepMind. The studio’s filmmakers get early access to DeepMind’s video models, including Veo, and feed product feedback back to the research team. Its first project, Eliza McNitt’s “Ancestra,” premiered at the Tribeca Festival in June 2025 and combined live-action performances from SAG-AFTRA actors with AI-generated visuals. Its animated series “On this day… 1776” runs on TIME’s YouTube channel through 2026, timed to the 250th anniversary of the Declaration of Independence.

Aronofsky’s framing is that AI is the next entry in a lineage that runs through sound, color, and VFX, and that the goal is to keep artists in control rather than replace them. He has a long track record of pushing film technology, including the 18K “Big Sky” camera system built for his Sphere production in Las Vegas and the earlier VR project “Spheres.” The use of generative AI in film also draws significant pushback from parts of the industry over authorship and labor, which is part of why the keep-artists-in-the-driver’s-seat positioning matters to how these tools get adopted.

The connection to ElevenLabs is the shared thesis. Both companies are betting that AI becomes a tool inside the storytelling stack: Primordial Soup works the visual and narrative layer, while ElevenLabs supplies the voice and audio layer through text-to-speech, dubbing, and making content work in any language and voice. Holding the conversation in an opera house, a venue built around the human voice, put that division of labor on display. Dubbing in particular is core to ElevenLabs’ product and to Aronofsky’s stated interest in reaching wider audiences without losing the original performance.

Jennifer Li Fireside

Jennifer Li is a general partner at a16z, where she leads the firm’s AI infrastructure investing. a16z co-led the ElevenLabs Series A in 2023, the deal Bryan Kim won and signed (Kim is a board observer). Li took the board seat and has held it through the Series D at an $11B valuation.

Her read on the company ties to its pitch: she tracked synthetic voice for years and waited for a model to cross the uncanny valley, the point at which a generated voice carries real intonation and emotion rather than sounding robotic. ElevenLabs was the one that did, the same milestone Mati described on stage when he called the first Warsaw model the first to cross the uncanny valley of speech.

Her wider thesis frames why a company like ElevenLabs fits the summit’s argument. Li holds that infrastructure is becoming the most important layer of the AI stack and that distribution is the key differentiator, with AI’s impact comparable to the internet’s. Before a16z she led product at Solvvy, a conversational AI company acquired by Zoom, and earlier worked at AppDynamics, acquired by Cisco for $3.7B.

Check out her cameo in the ElevenLabs doc.

Two New Models

Mati showcased 2 new models:

1. Dubbing V2, the first end-to-end dubbing model, conditioning on the original audio to solve flat sound and editable after the fact, shown via a Ramp Super Bowl ad.

1. Eleven V4 (out in weeks), a text-to-speech model with line-by-line control over a full range of emotions. The live demo of an AI booking experience closed on the iconic voice of Polish actor Piotr Fronczewski, a deliberate local flourish.

Even More to Build

It was fitting that the man who opened Polish Dynamism Day (June 2nd) was Lech Wałęsa, Poland’s first post-communist president, founder of Solidarność, Nobel Peace laureate, and one of the most consequential figures in the country’s history. The day before, he’d been shown ElevenLabs’ original machine, the one Piotr assembled by hand in his Warsaw apartment at the very beginning.

Pictured: Charlotte + Lech Wałęsa

His message to a room of founders and leaders who mostly keep their distance from politics: don’t. He argued that collaboration between tech leaders like ElevenLabs and the government is what sustains a capitalist system, and that strong democracy and free markets need founders to stay engaged.

The day before, Mati opened the summit on a very similar note, the company and the country both coming home:

“Poland has been through a remarkable transformation… We are only in the very early days of discovering what AI makes possible. The ambition is there. The technology is advancing. There is so much more to build, and we are proud to be pulling this future into existence with all of you today.”

Subscribe now

Share

Brought to you by:

Turing—Turing partners with frontier AI labs to improve model capabilities in coding, reasoning, tool use, & multimodality, as well as with Fortune 500 enterprises to build & deploy end-to-end agentic AI systems in mission-critical workflows Visit: turing.com/sourcery

VCX—VCX is the public ticker for private tech, allowing investors of all sizes to invest in venture capital. View The Portfolio at GetVCX.com

Deel—Deel is the global people platform that helps startups hire, manage, pay, and equip anyone, anywhere. Trusted by more than 35,000 fast-growing companies, Deel is the people platform that just works, so teams can scale without the chaos. Visit: deel.com/sourcery

Public-–Investing platform Public just launched Generated Assets, which lets you turn any idea into an investable index with AI. With Generated Assets, you can build, backtest, refine, and invest in any thesis with AI. Gone are the days of one-size-fits-all ETFs. Try it today: public.com/sourcery

Merge—The leading provider of customer-facing integrations and agentic tools for frontier LLMs, Fortune 500 organizations, and B2B SaaS companies. Visit: https://merge.dev

The material presented on Molly O’Shea’s website are my opinions only and are provided for informational purposes and should not be construed as investment advice. It is not a recommendation of, or an offer to sell or solicitation of an offer to buy, any particular security, strategy, or investment product. Any analysis or discussion of investments, sectors or the market generally are based on current information, including from public sources, that I consider reliable, but I do not represent that any research or the information provided is accurate or complete, and it should not be relied on as such. My views and opinions expressed in any website content are current at the time of publication and are subject to change. Past performance is not indicative of future results.

Paid Endorsement. Brokerage services by Open to the Public Investing Inc, member FINRA & SIPC. Advisory services by Public Advisors LLC, SEC-registered adviser. Crypto trading provided by Zero Hash LLC, licensed by the NYSDFS. Generated Assets is an interactive analysis tool by Public Advisors. Output is for informational purposes only and is not an investment recommendation or advice. See disclosures at public.com/disclosures/ga. Matched funds must remain in your account for at least 5 years. Match rate and other terms are subject to change at any time.

EVERYTHING EXCHANGE

Brian Armstrong, Co-Founder & CEO of Coinbase, takes us BTS of the Coinbase System Update.. going deeper into how Coinbase is rebuilding around intelligence & making investing more accessible for the entire world.

→ Listen on X, Spotify, YouTube, Apple

We cover tokenized equities, an SEC-registered AI agent advisor, and the unified global liquidity approvals that bring Coinbase's offshore and US order books together. Brian explains why 4 billion people remain unbrokered, how the Everything Exchange works, and where pre-IPO perps fit as nearly $4 trillion in new IPOs come to market (SpaceX, OpenAI, Anthropic).

On AI: 1,200 full-time agents now work at Coinbase, team sizes are shrinking to as few as 1 person, and intelligent routing to open source models is flattening the company's AI cost curve.

Brian also gets into economic freedom, his time in Argentina, accredited investor reform, NYT hit piece, & the mentors he leans on.

BIG thank you to the New York Stock Exchange for video production.

Subscribe now

Share

𝐓𝐈𝐌𝐄𝐒𝐓𝐀𝐌𝐏𝐒

(00:00) Brian Armstrong, Co-Founder & CEO of Coinbase
(00:45) Backstage before the Announcements drop
(01:47) Why Tokenized Stocks never worked before until now
(03:58) Inside Coinbase Advisor: The SEC-Registered AI Investment Agent
(05:36) The real mission behind every Coinbase product
(06:55) Does the average American feel economically free?
(08:42) What Argentina taught Brian about Capitalism
(11:16) Why younger generations are drifting toward Socialism
(13:01) What Roblox reveals about the next generation
(13:57) Why no one in Argentina plans for the future
(18:17) How Crypto is rewriting financial infrastructure
(20:20) The Strategy behind Coinbase's "Everything Exchange"
(21:28) Building the most trusted brand in the Crypto Industry
(23:49) Prediction Markets, Dry Powder & Reading Market Cycles
(26:07) The Pre-IPO loophole Coinbase built for everyone else
(28:18) Why Coinbase wins the super app war
(29:40) The tweet that got 24 million views
(32:22) Will AI actually kill jobs?
(36:03) How Brian stays on top of everything
(38:19) Founder Energy vs. Operator Energy
(39:33) AI has a branding problem
(46:41) Will Anthropic and OpenAI just build every company?
(48:12) Who is Coinbase's real competitor?
(51:25) The Announcement Brian is most excited about
(53:03) Getting $4 trillion of new IPO wealth onto Coinbase
(54:08) The mentors who shaped Brian Armstrong

Brought to you by:

Brex—The intelligent finance platform: cards, expenses, travel, bill pay, banking—wrapped into a high-performance stack. Built for scale. Trusted by teams that move fast. visit → brex.com/sourcery

Turing—Turing partners with frontier AI labs to improve model capabilities in coding, reasoning, tool use, & multimodality, as well as with Fortune 500 enterprises to build & deploy end-to-end agentic AI systems in mission-critical workflows Visit: turing.com/sourcery

VCX—VCX is the public ticker for private tech, allowing investors of all sizes to invest in venture capital. View The Portfolio at GetVCX.com

Deel—Deel is the global people platform that helps startups hire, manage, pay, and equip anyone, anywhere. Trusted by more than 35,000 fast-growing companies, Deel is the people platform that just works, so teams can scale without the chaos. Visit: deel.com/sourcery

Public-–Investing platform Public just launched Generated Assets, which lets you turn any idea into an investable index with AI. With Generated Assets, you can build, backtest, refine, and invest in any thesis with AI. Gone are the days of one-size-fits-all ETFs. Try it today: public.com/sourcery

Merge—The leading provider of customer-facing integrations and agentic tools for frontier LLMs, Fortune 500 organizations, and B2B SaaS companies. Visit: https://merge.dev

Brian Armstrong: Tokenized Stocks, AI Agents, & Rebuilding Coinbase Around *Intelligence*

Brian Armstrong, co-founder and CEO of Coinbase, sat down with Sourcery backstage ahead of the Coinbase System Update, a twice-yearly event where the company details everything it built in the prior half. The conversation moved across tokenized equities, an SEC-registered AI advisor, the structure of private markets, and how a regulated financial company is reorganizing itself around AI.

→ Listen on X, Spotify, YouTube, Apple

Putting Stocks on Crypto Rails

Coinbase is moving stocks onto crypto rails backed one-to-one by the underlying share, rather than as the synthetics and derivatives that earlier products relied on. Armstrong drew the parallel to stablecoins, where a trusted, fully backed token unlocked broad adoption. The same dynamic, in his framing, is now within reach for equities as market structure legislation advances.

The reach of the product is the point. “There’s actually about 4 billion people in the world today who are unbrokered. They don’t have access to any high-quality US investments.” Tokenized stocks are designed to trade 24/7, settle on chain, and move as easily as any other token, including peer-to-peer transfers.

Armstrong positioned Coinbase as the natural issuer of that one-to-one token. “Coinbase is the most trusted brand in crypto.” The company holds the underlying stock, mints a token on top of it, & opens access to holders worldwide who previously had no route into US companies.

An SEC-Registered AI Advisor

Coinbase Advisor is built directly into the Coinbase app and can research, conduct trades, run tax loss harvesting strategies, and handle financial education in plain English. The distinguishing move is regulatory. Rather than disclaim its output, Coinbase registered the agent itself. “We’re the most trusted brand in crypto. Why don’t we build this the right way & actually make it real investment advice?”

That required a conversation with the SEC about registering an AI agent as the investment advisor, not a human behind it. Armstrong said the regulator agreed. “They let us register an AI agent as an SEC-registered investment advisor,” with the result that the guidance the product gives can be described as investment advice.

The human remains in the loop on recommendations today, with the agent researching and proposing while the user approves. Armstrong framed the data this generates as a long-term asset. The company is collecting it to build what he called a frontier model for investing, on the view that scaling laws apply to that data set the way they apply to any other.

Economic Freedom as the Operating Thesis

The product roadmap sits inside a broader argument about economic freedom. Armstrong tied Coinbase’s mission to research correlating high-freedom economies with outcomes beyond GDP per capita, including happiness, environmental treatment, conditions for the poorest 10%, and reduced corruption, war, and infant mortality. “We believe that economic freedom is a foundational necessity for all civilizational progress.”

His year in Argentina is the reference case. He described how hyperinflation and weak property rights reshape behavior, pushing people toward short horizons and away from the 10 to 20 year timelines that wealth generation and company building requires. “It really changes the entire fabric of society to have high inflation and low economic freedom in a negative way.” High inflation, in his account, harms the poorest most, since they hold wealth in cash while the wealthy access inflation-resistant assets.

Armstrong was direct about his economic priors. “We believe capitalism is a good thing, and we want everybody to have a piece of it.” On the generational drift toward socialism, he argued the case has to be remade each cycle. “Every generation has to kinda learn this from first principles if they haven’t been exposed to it.”

The Everything Exchange & the Case Against Accredited Investor Rules

The Everything Exchange is Coinbase’s attempt to make every asset class tradable in a single account. “How do we get every asset class in the world tradable in one account?” The customer benefit Armstrong cited is capital efficiency, since assets in one place support better margin and credit, alongside 24/7 trading for anything running on crypto rails.

That framing led into private markets, where Armstrong was sharpest. With nearly $4 trillion in new IPOs coming to market, he argued AI companies have absorbed much of the world’s risk capital. “People are just trying to get as much of their risk capital into SpaceX, OpenAI, and Anthropic right now.” Coinbase’s answer is pre-IPO perps, which launched alongside SpaceX before it went public and, per Armstrong, saw significant volume.

His critique of the accredited investor regime was the core of the segment. The rules, designed to protect retail, in his view produce the opposite. “It makes it so only rich people can get richer. It’s like the most regressive tax.” Retail can only access many companies once they are already valued in the trillions, after the upside has accrued. His proposed fix swaps the net worth threshold for a knowledge test. “The better way to do it would be a financial literacy test.”

Rebuilding the Company Around Intelligence

Internally, Coinbase is reorganizing around AI in ways that show up in the engineering metrics. “The amount of code being shipped per developer is up about 2X year over year,” with top engineers shipping on the order of 100 pull requests a week. Armstrong said the company is taking those outlier builders and using them to train the rest of the organization.

Team structure is changing with it. The traditional pod of 10 is compressing to two to four people, and in some cases one. “In some cases, we’re seeing a one-person team,” because the human works alongside roughly 10 AI agents that create pull requests and designs and participate in Slack. By Armstrong’s estimate, “there’s about 1,200 full-time agents working at Coinbase now,” measured by agent working hours against a typical 40 to 60 hour week.

Crucially, he tied the output gains to quality rather than against it. “We’re actually seeing the rate of bugs and incidents go down per line of code shipped.” The next reach is recursive self-improvement, where agents aggregate customer feedback, plan and draft the code, and a human reviews a batch of changes each day rather than editing each one. Armstrong’s emphasis was on updating the context that generates the work, not the work itself.

On unit economics, Coinbase built intelligent routing on top of an open source project and modified it, sending simpler queries to cheaper models. Open source models run roughly 3 to 6 months behind the frontier but about 99% cheaper for inference. The result is that token usage keeps growing while cost growth flattens. “My guess is that within 12 to 18 months, 80% of our workloads will be going toward models that are 99% cheaper,” with the remaining 20% reserved for frontier-level tasks aka “IQ-maxxxxxxing.”

Building the Financial Rails for the Agentic Economy

Armstrong’s view of where this leads is an economy run by orchestrated agents. “You’re increasingly gonna talk to one agent. That agent is gonna orchestrate hundreds of thousands of other agents.” Those agents pay for goods and services and hire other agents into teams, which requires payment infrastructure built for very small, very frequent transactions.

The constraint is identity. Agents cannot hold a piece of government-issued paper or fill out a CAPTCHA, so they need their own financial accounts and self-custodial wallets. Coinbase is powering this through its base MCP API, giving agents a way to transact natively. Armstrong called this the infrastructure layer for the agentic economy and the announcement he was most excited about.

On the labor question that sits underneath all of it, Armstrong landed in the optimistic camp.

“AI is gonna eliminate tasks & toil. It’s not gonna eliminate jobs, in my view.”

His expectation is that smaller average team sizes let companies do more things, employment stays roughly constant, and work becomes optional over the long term while most people still choose to work, much as podcasting or running a company would not have counted as work a century ago.

He was equally clear about who builds what. Armstrong does not expect the large model labs to build every company, arguing they are focused on broadly applicable foundational models and are unlikely to enter regulated financial services. That leaves room for vertical frontier models, including the investing model Coinbase intends to build on its own customer data.

Distribution, Media, & Standing for a Position

The final throughline was posture. Armstrong spends roughly 80% of his media time on new media, including podcasts, Substack, + X, and about 20% on traditional outlets, mostly to reach policymakers and readers over 50 or 60. His argument is that companies should own their distribution and publish directly.

He recounted a negative New York Times piece that followed Coinbase’s Mission First blog post and described how little it ultimately moved the business. “It took getting a negative article or two, which was scary at first, to realize it just didn’t matter.” That realization, in his telling, freed him to be more creative and more confident rather than optimizing for optics.

The advice he drew from it was a stance, not a tactic. “I think you should just have an opinion, stand for it, and don’t apologize.” He extended the same logic to AI’s reputation, arguing the labs would be better served by making the case for the technology directly than by fearmongering or inviting regulation.

He closed on the people he learns from, citing the idea that “you’re the average of your five closest friends,” naming Marc Andreessen on policy + comms, and Tobi Lutke on building an AI-native organization, and describing a structured mastermind cadence of monthly meetings & an annual trip that keeps those relationships current.

This was so much fun, really enjoyed spending more time with the Coinbase team & having a sneak peek into their latest updates. Check out more by following Coinbase & follow Brian on X.. you won’t be disappointed.

→ Listen on X, Spotify, YouTube, Apple

Subscribe now

Share

The material presented on Molly O’Shea’s website are my opinions only and are provided for informational purposes and should not be construed as investment advice. It is not a recommendation of, or an offer to sell or solicitation of an offer to buy, any particular security, strategy, or investment product. Any analysis or discussion of investments, sectors or the market generally are based on current information, including from public sources, that I consider reliable, but I do not represent that any research or the information provided is accurate or complete, and it should not be relied on as such. My views and opinions expressed in any website content are current at the time of publication and are subject to change. Past performance is not indicative of future results.

Paid Endorsement. Brokerage services by Open to the Public Investing Inc, member FINRA & SIPC. Advisory services by Public Advisors LLC, SEC-registered adviser. Crypto trading provided by Zero Hash LLC, licensed by the NYSDFS. Generated Assets is an interactive analysis tool by Public Advisors. Output is for informational purposes only and is not an investment recommendation or advice. See disclosures at public.com/disclosures/ga. Matched funds must remain in your account for at least 5 years. Match rate and other terms are subject to change at any time.

“Your # 1 job as a founder is to be right”

Mark Pincus, founder of Zynga and author of the newly released Life at the Speed of Play aka "Product Maker Bible" (HarperCollins, foreword by Reid Hoffman, joins Sourcery to break down the framework he’s used over the past three decades to build hit products. Pincus took games like FarmVille & Words With Friends to more than 1 billion users in 4 years and a $12.7B exit, and was an early investor in Facebook, Twitter, & Polymarket. This conversation is a tactical playbook for builders and the investors backing them.

→ Listen on X, Spotify, YouTube, Apple

Pincus covers separating winning instincts from losing ideas, his Proven Better New framework to increase odds of success, why consumer products win on “day 365 retention” instead of virality, how he represents the ‘dumb fucks’ and uses his iPhone home screen as market research, the tech assistant model & other unscalable hiring tactics he learned from Jeff Bezos, his new enterprise AI company Hivemind, how to spot a fake CEO, the 2006 think weekend he hosted with Peter Thiel, Mark Zuckerberg, + Sean Parker, & why he is an AI maximalist who still thinks consumer AI is un-investable.

Get the book now on Amazon or at lifeatthespeedofplay.com, + follow Mark on X @markpinc

Read more, watch more, listen to more, and subscribe to our newsletter for more conversations with the operators & investors building the next cycle.

Subscribe now

Share

𝐓𝐈𝐌𝐄𝐒𝐓𝐀𝐌𝐏𝐒

(00:00) Mark Pincus, Chairman & Founder at Zynga
(01:07) Why it took so long to write the book
(01:52) Why the book starts with Elon
(07:29) How frustration becomes market research
(08:16) The gap between Airbnb and hotels
(09:14) Why people don't want flight attendants on private jets
(13:13) The goal behind Life at the Speed of Play
(15:07) The instincts vs. ideas framework
(21:01) Do instincts improve with experience?
(22:49) A framework for building winning products
(29:38) Becoming a student of yourself
(31:16) Finding great ideas in overlooked products
(34:39) Why AI makes building easier and winning harder
(39:17) Cracking Zynga's record-setting retention
(44:59) What Mark looks for in startups
(50:42) Jeff Bezos' boldest decision
(53:03) What happens when founders realize they're wrong
(59:37) How Mark found talent others overlooked
(1:06:58) Don't be a fake CEO
(1:12:15) Inside Silicon Valley's early days
(1:14:58) Why he funded Friendster and Napster as "experiments"
(1:17:00) The think weekend with Thiel, Zuckerberg, Reid Hoffman & Sean Parker
(1:18:58) Launching Zynga when nobody believed him
(1:21:42) Why Mark is an AI maximalist

Brought to you by:

Brex—The intelligent finance platform: cards, expenses, travel, bill pay, banking—wrapped into a high-performance stack. Built for scale. Trusted by teams that move fast. visit → brex.com/sourcery

Turing—Turing partners with frontier AI labs to improve model capabilities in coding, reasoning, tool use, & multimodality, as well as with Fortune 500 enterprises to build & deploy end-to-end agentic AI systems in mission-critical workflows Visit: turing.com/sourcery

VCX—VCX is the public ticker for private tech, allowing investors of all sizes to invest in venture capital. View The Portfolio at GetVCX.com

Deel—Deel is the global people platform that helps startups hire, manage, pay, and equip anyone, anywhere. Trusted by more than 35,000 fast-growing companies, Deel is the people platform that just works, so teams can scale without the chaos. Visit: deel.com/sourcery

Public-–Investing platform Public just launched Generated Assets, which lets you turn any idea into an investable index with AI. With Generated Assets, you can build, backtest, refine, and invest in any thesis with AI. Gone are the days of one-size-fits-all ETFs. Try it today: public.com/sourcery

Merge—The leading provider of customer-facing integrations and agentic tools for frontier LLMs, Fortune 500 organizations, and B2B SaaS companies. Visit: https://merge.dev

Mark Pincus

The Winter That Built Silicon Valley Giants, & How to Create Billion-Dollar Products

Mark Pincus founded Zynga, took FarmVille and Words With Friends past 1 billion users in 4 years, and exited for $12.7B. His new book, Life at the Speed of Play, is out today, June 23, 2026 from HarperCollins with a foreword by Reid Hoffman. To go back to the beginning, his story starts in the years everyone else wrote off, the years that turned out to seed the next decade of giants.

Take-Two Interactive acquired Zynga on May 23, 2022, in a cash-and-stock transaction valued at $12.7 billion, ranking among the largest mergers in video game history. This acquisition combines Take-Two’s console and PC franchises with Zynga’s mobile portfolio to significantly expand its mobile gaming business and pursue annual cost synergies.

→ Listen on X, Spotify, YouTube, Apple

The quiet years that seeded the giants

Mark dates the stretch from 2002 to 2007 as the nuclear winter for the internet, and he says it with a smile, because the people who stayed are the ones who built the next cycle. San Francisco had emptied out. “It was like they’d burned the boats, and there was like... It was a ghost town,” he told me. The MBAs and the VCs had cleared out, and many of his friends left the industry entirely for real estate and other businesses.

A small group stayed and kept the flame. Mark’s read at the time was that the behavior had not changed, only the sentiment, and that the noise leaving was a gift. “It’s not over, folks. The dogs are still eating the dog food. They’re just not eating pet.com dog food.”

The proof came from the numbers. In Q4 2002, Mark and Reid Hoffman looked at the figures Amazon had just printed. “Holy fuck, they just grew everything 25% year over year,” he remembered thinking. “This shit is not over, it’s just starting to happen.” That instinct, watch what consumers actually do and not what Wall Street prices in, is the same one he says he still runs today. He pointed to the early mobile phone industry, where valuations spiked and fell while usage kept climbing the entire time.

He turned out to be right. The companies started in that supposedly ‘dead’ window, LinkedIn in 2003, Facebook in 2004, Twitter and Zynga in 2006 and 2007, became the $10 billion to $1 trillion businesses today that now define the industry. Facebook, worth $100 million when it was breaking out in 2006, became Meta. The winter was not the end of the boom. It was the ground the next one grew from.

The 2006 outsiders who turned out to be right

In March 2006, Mark hosted a *think weekend* at his Colorado ranch, which he is careful to describe as a ranchita, 3 acres, beautiful and broken down. The guest list was Peter Thiel, Reid Hoffman, Mark Zuckerberg, and Sean Parker, with about 35 people total. Thiel presented on the housing bubble he had bet his hedge fund Clarium on too early. That same fund held Palantir, which he handed out as a stub to everyone at the end.

What stands out is how far outside the establishment the group felt at the time. “In a way we were all outsiders. I know now it seems like the insiders, but none of us were part of a branded venture fund.” None of them sat inside a Sequoia or Kleiner or showed up at the Allen conference. They shared one conviction nobody else wanted, that consumer internet still mattered, and they were early on every part of it.

That conviction produced a working theory of where the value would go. Reid coined the term Web 2.0 out of these brainstorms, and the underlying rule was simple. “If it can be free, it will be free,” especially data. “The data wants to be free.” Mark and Reid put up the original money for Friendster “as a science experiment. We didn’t think it would work.” Mark put up all the original money for Napster, since Sean Parker had worked for him as an intern at 16. The bets were small, the conviction was large, and the thesis held.

Everyone said no, & he was right anyway

Mark launched Zynga in the spring of 2007 as a project, while living in New York. He was 41, had a few exits behind him, and had a clear hunger to get back to building consumer products. The people who had backed him for years would not fund it.

He went to Fred Wilson, who has backed everything he has done, and asked for a million dollars at 25% of the company. Fred’s answer: “Mark, you know I love you and I’d back anything, but I have a tough time believing that you’re really gonna go back to work.” First Round had a standing offer of a $250K bridge note at a $10 million cap for any second-time founder, sight unseen. Mark met Josh Kopelman at Penn Station and said, “I’m here for my money. Can I have my 250K?” Kopelman told him, “Mark, I just have trouble believing that this will be your company.”

The doubt was about the whole category, not just him. The advice he kept hearing was to go be a VC, because “you’re making dumb little apps on Facebook.” As he put it, “It was me and Max Levchin, and a bunch of college kids and people in China.” He sees the same energy in the best opportunities today. “Consumer, I think, always seems un-investable. It was like that then. It feels like that again now.” Even the games bet drew skepticism. Dave Morin at Facebook sat him down before the platform launched and said, “Mark, anything but games. They’re just not viral.” Games were not viral. They had something more durable, retention, and that is what Mark built the entire company around.

Retention won, and the household names followed

Zynga built every game against day 365 retention, the metric Mark cared about most, using social loops that reminded players and their friends to come back. The engine worked. By 2009, as he writes in the book’s OKR review, the company had “9 of the top 10 social games” and was “as big as our next 11 competitors combined. Our games had become household names.” FarmVille and Words With Friends carried that engine past 1 billion users in 4 years, and Zynga Poker and Words With Friends still rank among the company’s biggest games nearly 20 years later.

The doubt that met him at Penn Station in 2007 ended in one of the largest deals in gaming history. Take-Two announced its acquisition of Zynga on January 10, 2022, in a deal valued at $12.7 billion in cash and stock, and closed it in May 2022. Zynga shareholders received $3.50 in cash and 0.0406 of a Take-Two share for each Zynga share, bringing a library of games downloaded more than 4 billion times under Take-Two. The ‘little apps on Facebook that nobody wanted to fund’ became a $12.7 billion company.

Why the book starts with Elon

Life at the Speed of Play opens on Elon, because Mark wanted readers to feel the ceiling of what is possible. “I’d like people to picture the kinds of crazy things that Elon can go do that none of us can do.” Fart mode in a car. Deciding to start a tunnel boring company while stuck in LA traffic, then raising $1 billion within a month of the tweet. “He’s just having fun. You can tell by watching him.”

Keeping Elon in the opening took some ownership. His co-author, publisher, and chief of staff pushed back. “Mark, don’t go there. Elon’s controversial, and you shouldn’t just call him Elon, you should say Elon Musk.” Mark kept him, and used it to make a point about voice. “Why can’t we just be fucking normal? Why do we have to be something else the minute that we have a camera on or we’re publishing?”

The everyday version of that freedom, he argues, is posting. His most viral tweet ever came from a late check-in at the Peninsula in Chicago, when he asked, “Why is it that the nicer the hotel is, the longer it takes them to check us in?” It cleared roughly 13,000 likes. He treats that charge as market research, a live signal of something people instinctively know could be better. And he frames the upside of getting product direction right in hard numbers. A correct roadmap can put 80% of your engineering days into shots on goal, against maybe 20% for your best competitors, which is 400% of their efficiency.

Trust the instinct, free yourself to find the A

The most freeing idea in the book, and the core of the Stanford course Mark has taught for 10 years, is one distinction. “The most foundational thing we need to get good at as founders, product founders, is separating our winning instincts from our losing ideas.” The instinct is almost always right. The first idea you attach to it usually is not, and realizing that is what opens the path forward.

His own examples are the proof. He registered smstaxi.com and wrote a full business plan years before ridesharing like Uber & Lyft took off, then set it aside. “My instinct was right, obviously, that eventually we would order cars through our phones, but that one idea I put on it was wrong.” Tribe.net went further. He started one of the first social networks, a month or two after Reid began LinkedIn and before Facebook and MySpace, and everything worked, virality included, and it still missed because the idea was wrong. “Tribe became three different industries, but one failed company.” The media tear sheet puts it in numbers: the product instinct is right about 95% of the time, and the specific idea is wrong at least 75% of the time.

Letting go of the merely good idea is what unlocks the great one. “A B+ is the enemy of an A. A B+ often, if we stick with the B+, is taking up space. It’s distracting us from unlocking the A.” On knowing when you have the real thing, he compares it to finding a partner. “If you’re asking, ‘Is this it? Could it be it?’ It’s definitely not it.” When it is right, the work stops feeling like a fight. “Life is too short to struggle.”

Proven, Better, New

The framework for stacking the odds in your favor is Proven Better New, which the book was nearly titled after. Mark’s instruction is to get out of your own way. “We should be like scientists with white lab coats.”

“You need to feel no attachment or connection to your ideas, have a white lab coat, and just do science experiments.”

Proven means deconstructing what already works for your exact audience and use case. Better means the small things that 10 out of 10 of that audience agree are improvements, often something as plain as price or removing a download step. New is the novel bet, “and we have to accept that the new will probably fail.” Doing proven and better well is what keeps a failed new idea from reading as a false negative.

His example is Nikita Beer and tbh, who found a working honesty-box concept buried inside a narrow Arabic-language product, copied it, and put it front and center. Mark runs the same test on himself. Any app that earns a spot on his iPhone home screen gets studied, because he has calibrated that he represents the early majority about 18 months out. He used to put the stick value of that signal at a billion dollars and now says maybe 5 billion. He cold emailed Raya, then invested and put up most of the money.

The hard part is distribution, and Zynga is the playbook. “We download an average of zero new apps a month.” Last year about 40,000 games launched in the App Store and 0% sustained a top 25 position. So Zynga built every game against day 365 retention, using social loops to pull players and their friends back. Zynga Poker and Words With Friends are still among the company’s biggest games nearly 20 years later.

The deepest lever was reward. “The highest reward for most people is social status.” Zynga even counted the marriages that came out of strangers meeting in its games and treated them as the ultimate proof of engagement.

Be right, then have the nerve to act on it

“Your number one job as a founder is to be right.” The harder, more inspiring part is what you do once you see the truth, especially when your own team disagrees. Mark points to Jeff Bezos launching Amazon Prime against his CFO and most of the company, on an instinct built from watching how people behaved at Costco. The discipline is refusing to confuse conviction with evidence. “Hope is not a strategy.” The job is to stay honest and keep moving toward it. “We have to be a heat-seeking missile, and we have to be guided by this pursuit of the truth and this intellectual honesty.”

Get the book now on Amazon or at lifeatthespeedofplay.com, + follow Mark on X @markpinc

That same clarity shapes how a founder should spend a day. “Don’t be a fake CEO.” His definition of the real job is narrow and freeing. “My job is to work with great product teams on great products that our users love. That’s my job. Nothing else is my job.” He had his assistant track what share of his calendar went to the real work versus conferences and press, and aimed to keep it above 50%. His method for scaling judgment is the tech assistant, a lineage he traces from Andy Grove to Bill Gates to Jeff Bezos, whose entire C-staff started in the role. “What you’re really doing is transferring your vampire blood.” Ian Cinnamon, an early MIT grad whom Zynga’s poker team had written off as too entrepreneurial, became Mark’s tech assistant and now runs Apex.

On AI, Mark is an optimist with his eyes open. “I’m an AI maximalist.” He separates this cycle from 2000 on the fundamentals. “It’s not dark fiber. These GPUs are burning hot, and people are getting real value out of it.” He is invested in NVIDIA and the memory names and likes the setup. “I love the trade because they’re at a PEG ratio below one.”

He thinks today’s $2 to $5 trillion companies become $10 to $20 trillion companies. He still calls consumer AI un-investable for now, naming only Suno and Midjourney as cases that make sense to him, and he does not buy that LLMs eat every business, pointing to Cursor and Harvey as proof the best wrappers found real footing and keep going deeper. The throughline from the winter of 2002 to the AI cycle of 2026 is the same one that has carried his whole career: watch what people actually do, trust the instinct, and have the nerve to build it.

→ Listen on X, Spotify, YouTube, Apple

Get the book now on Amazon or at lifeatthespeedofplay.com, + follow Mark on X @markpinc

Subscribe now

Share

The material presented on Molly O’Shea’s website are my opinions only and are provided for informational purposes and should not be construed as investment advice. It is not a recommendation of, or an offer to sell or solicitation of an offer to buy, any particular security, strategy, or investment product. Any analysis or discussion of investments, sectors or the market generally are based on current information, including from public sources, that I consider reliable, but I do not represent that any research or the information provided is accurate or complete, and it should not be relied on as such. My views and opinions expressed in any website content are current at the time of publication and are subject to change. Past performance is not indicative of future results.

Paid Endorsement. Brokerage services by Open to the Public Investing Inc, member FINRA & SIPC. Advisory services by Public Advisors LLC, SEC-registered adviser. Crypto trading provided by Zero Hash LLC, licensed by the NYSDFS. Generated Assets is an interactive analysis tool by Public Advisors. Output is for informational purposes only and is not an investment recommendation or advice. See disclosures at public.com/disclosures/ga. Matched funds must remain in your account for at least 5 years. Match rate and other terms are subject to change at any time.

SpaceX, OpenAI, Anthropic, Cerebras, Cursor..

We're in the biggest wealth event of all time, with $4T+ in combined value on just the big names, with more heading to public markets & trading in secondaries, more millionaires + billionaires are being minted than ever before.

→ Listen on X, Spotify, YouTube, Apple

The most sought-after next step is flying private.. and it pulls them straight toward $75 million jets, often skipping every step the industry is used to.

Israel ‘Izzy’ Slodowitz is the Founder of Craft, a Part 135 charter operator that also flies supplemental lift for NetJets, Flexjet, and Wheels Up. He turned that flying fleet into something most operators can’t offer: an exchange fund built for the founders & early employees whose wealth is locked in a single stock. It takes in that concentrated position tax-deferred and backs it with real aircraft, not the self-storage and real estate that prop up legacy funds. Members diversify out of one position, defer the tax, and keep access to the planes.

I joined him on one of their Challenger 350’s from Van Nuys to San Francisco to break down what this wave means for private aviation. Who is buying, what it really costs, why jets are now appreciating, & why a generation whose net worth is locked in stock needs a new structure the legacy ownership models were never built to offer.

We cover:

• Who flies more: Elon, Sergey Brin, Jensen, Alex Karp, Taylor Swift?

• PJ shortage?!

• Charter vs jet card vs fractional vs own

• A G650 that sold in one hour

• What a jet really costs per year

• Why capital gains wealth can’t write off a jet

• The $10M deduction w/ $2M down (+recapture trap)

• Hidden fees, safety gaps, & first-timer mistakes that cost the most

• Does flying private actually cure jet lag?

• Starlink

You can reach Izzy & Craft at flycraft.com or useradar.com

Subscribe now

Share

𝐓𝐈𝐌𝐄𝐒𝐓𝐀𝐌𝐏𝐒

(00:00) Israel Slodowitz, Founder & CEO of Craft
(00:58) The state of Private Aviation in 2026
(04:09) Why first time buyers are buying the biggest planes
(05:14) A coming shortage of private jets
(06:00) The Private Jet size chart nobody explains
(08:26) When does it make sense to buy vs. charter?
(11:24) The real cost of a $75M jet
(13:30) Breaking down prices by jet category
(15:07) Charter vs. Jet card vs. Fractional ownership
(19:31) Inside Craft's exchange fund model
(25:33) The most economical way to fly private
(26:24) Alex Karp's $17M flight bill
(28:01) How Aircraft financing actually works
(29:25) The Biggest mistakes people make buying a plane
(35:48) How Starlink changed the business