,

TomorrowX Summit Day 1: Gerstner, Druckenmiller and Solomon on the AI Trade

Day 1 of the TomorrowX Summit in Austin brought Brad Gerstner, Stan Druckenmiller, David Solomon, Guillermo Rauch and the event’s hosts to one stage to argue over the AI trade. The TomorrowX Summit, run by iConnections with Gavin Baker’s Atreides Management and Antonio Gracias’s Valor Equity Partners, took over the Moody Center on November 17, 2026, with retail investors, X and Substack writers, and the CEOs they cover. This post distills the five main stage sessions published so far: the hosts’ opening panel with CNBC’s Sara Eisen, Gerstner on why Anthropic’s IPO is oversubscribed, Druckenmiller on why he is “not playing,” Goldman Sachs CEO David Solomon on whether AI has peaked, and Vercel CEO Guillermo Rauch on the flood of software that coding agents are producing. It was written on the morning of November 18, before Anthropic’s IPO priced, and will be updated once it does.

TLDW

The hosts opened by arguing that retail investors are now the most powerful force in the market and that the AI industry owes ordinary Americans the truth about what AI can do for them. Gerstner made the bull case for Anthropic’s IPO on unit economics: a roughly $100 billion revenue run rate with about 5,000 people, compute costs falling about 90% a year, and only 26% of its 2026 compute spent serving customers. Druckenmiller said he has rarely had less clarity, owns almost none of his January portfolio, is roughly flat on equities, and thinks the next AI engine is use cases, not infrastructure, with the labs themselves “selling tasks instead of tokens.” Solomon called the selloff a recalibration in the early innings of a long cycle and put the biggest prize in the application layer, while Rauch predicted that coding agents will, within months, produce more software than in all prior history and said Vercel, now at $700 million in annualized revenue, wants to be a public company.

Thoughts

Four of Day 1’s speakers drew four different maps of where AI’s profits will land, and the Anthropic IPO is a bet on which map is right. Gerstner is a self-described “model maximalist”: he expects most of the value to go to the top one to three model makers, possibly a more natural monopoly than search or social, and his rule for everything else is to “own the shit that gets better as intelligence gets better.” His test is blunt: if a better model would make a holding worse off, the way it could a vertical application company, he does not want to own it. Solomon said the opposite, that the big value creation over the next 10 to 20 years sits in the application layer, the way the iPhone and Uber came a decade after the PC makers of 2000. Rauch wants Vercel to capture “every pixel and every token” at the layer where agent-built software gets deployed. Druckenmiller’s answer is the one that reconciles them, and it is the least comfortable for anyone holding software: the lab with the number one intelligence stops selling tokens and starts selling finished tasks, in cybersecurity, for example, and in healthcare. If he is right, Solomon’s application layer and Gerstner’s model layer become the same company, and that is the strongest argument for buying Anthropic and the strongest argument against owning most of what sits on top of it.

Gerstner’s best number is also the bear’s best number. Only 26% of Anthropic’s compute in the first nine months of 2026 went to serving the customers who produce its revenue, about 500 megawatts of two gigawatts, which he presented as evidence of a strong business model. Read the other way, roughly three quarters of the compute is a bet on the next model, and he said plainly that Anthropic wants much more of it because it thinks it is in “the foothills” of superintelligence. Whether that spending is a one-off or a treadmill decides whether the margins ever show up. Asked which AI bear thesis he respects most, Druckenmiller named “politics slash the financing,” and he did not tie it to any deal. Still, an IPO reported to raise more than $60 billion is about as direct an expression of that financing risk as the market offers. Baker’s version of the same worry was the most honest moment of the opening panel: he wants the build-out to slow, because a supply chain growing capacity at 200% would “for sure” overbuild, as it has in “the last literally 400 years of capitalism.” The bulls on stage were not denying the bear case. Baker, at least, was betting that wafer shortages, higher real rates and wider credit spreads would ration the build-out before it ran away.

The day’s most useful contrast was between Baker’s “not a bubble,” Solomon’s “recalibration,” and the most experienced macro investor in the building, who said nothing about bubbles and simply stepped aside. Baker said “we’re certainly not in a bubble right now” and read 10 to 20 year low valuation multiples as the market pricing in peak spending. Solomon called the drawdown a recalibration and said multiples look better than they did six months ago. Druckenmiller, who rode the AI infrastructure trade this year after crediting Baker with calling the whole boom, described a macro picture that has flipped since January: a Fed that is hiking after what he called a regime change, market breadth that peaked on August 14 while the averages kept rising into October, widening credit spreads, defensives leading cyclicals, and a risk that the second half of 2027 disappoints. He would not short the innovation theme and called parts of it washed out, but his conclusion was capital preservation. For a room full of individual investors, that combination is the actual lesson: you can believe AI is the biggest thing in the history of American business and still decide you do not know where prices will be in 12 months.

TomorrowX sells democratized access, and it deserves credit for how much of it is real, but the price list tells the rest of the story. The summit’s own site describes it as “the first investment conference of its kind to democratize access for all investors,” and Baker opened by calling retail “the most powerful force in the market,” recalling a stock that rose 7% on a bank’s strong sell because one line from Citrini moved it. The passes run from $500 for upper-deck seating to $10,000 for “Smart Money,” which requires organizer approval and is the only tier with private one-on-one meetings with the companies. Solomon, meanwhile, described Goldman’s ultra high net worth business, roughly 45,000 clients with about $25 million or more each, as a place where clients get early access to names like SpaceX and Anthropic. Access is still tiered by wealth, at the summit and on Wall Street. The most radical democratization idea of the day had nothing to do with conference passes: Gerstner’s Invest America accounts put $1,000 in the S&P 500 for every American child at birth. That, not a better seat in an arena, is what gives a kid from a town of 3,200 in rural Indiana, which is where Gerstner grew up, a stake in the companies on stage.

Nobody on Day 1 answered the jobs question directly, but three speakers answered it sideways, and their answers do not fit together. Solomon said Goldman will “probably” have fewer engineers in five years, but more “super super powerful, very senior engineers,” and said the real opportunity is reinvesting AI efficiency in people and growth rather than margins. That leaves open where tomorrow’s senior engineers come from if the junior rungs disappear. Rauch called all hand-written software “almost a liability,” then closed by telling people not to become “a useful idiot” and to keep reading the code, which sits uneasily with his first line. Gerstner was the only one to frame a policy response, comparing Invest America to Social Security, born when unemployment was 22%, and noting that unemployment today is 4.1%. His point was that this time the safety net can be built before the shock, as ownership rather than dependency. Whether ownership of a $1,000 index stake is enough for workers whose wages, as Gracias noted on the opening panel, have barely moved since before Covid is the question the summit’s American dream theme raised and did not settle.

Key Takeaways

  • Gavin Baker traced TomorrowX to one day when a bulge bracket bank cut a semiconductor equipment stock to strong sell and it rose 7% anyway, because one line from the Citrini Research account mattered more. His conclusion: retail is now the most powerful force in the market.
  • The hosts said the event is not meant to pump AI stocks. Its stated goal is to bridge the X, Substack and Reddit voices that move markets with the management teams they write about, and to get the AI industry to tell ordinary Americans the truth about what AI does for them.
  • Baker framed the AI build-out as telecom or railroads. If token demand keeps growing as AI gets more useful, it looks like the railroads, which he said lasted about 60 years at a share of GDP roughly comparable to today.
  • Baker wants the build-out slower and smoother for longer, and he called higher rates, wider spreads and a wafer shortage good news because they prevent an overbuild.
  • Gerstner said Anthropic reached a $100 billion revenue run rate with about 5,000 people, against Google’s 105,000 at the same milestone, and that compute costs are falling about 90% a year while what Anthropic can charge has upside as models improve.
  • Only 26% of Anthropic’s compute in the first nine months of 2026 went to serving customers, by Gerstner’s slides. The rest went mostly to training, which he presented as the reason fears about data centers are overdone.
  • Gerstner called the Anthropic offering roughly 10x oversubscribed and likely Altimeter’s biggest order in fund history. Bloomberg’s reported figure in the days before was about 6x.
  • Gerstner said he may hedge Anthropic in his hedge fund if it overshoots after the IPO. Ron Biscardi noted that this year’s ten biggest IPOs are down about 20% from their issue prices on average.
  • Druckenmiller owns “basically almost none” of his January portfolio. He is roughly flat on equities, short copper, and in capital preservation mode for “two days or six months.”
  • Druckenmiller thinks AI infrastructure is a tactical trade now, not the engine it was for four years, and that the most interesting AI use case is a frontier lab selling finished tasks in fields like cybersecurity and healthcare.
  • Druckenmiller’s ten-year calls: a 20% chance that fusion arrives or is clearly coming, at least 70% odds of data centers in space, and his belief that cancer becomes a managed disease within 10 years.
  • Solomon sees AI in the early innings of a long cycle and the current drawdown as a recalibration. He expects the biggest value over 10 to 20 years in the application layer, not in infrastructure or the first models.
  • Goldman has 13,000 engineers in a 45,000-person firm. Solomon expects fewer engineers but more very senior ones in five years, and he wants AI efficiency reinvested in growth rather than margins.
  • Vercel’s annualized revenue went from over $600 million in early October to $700 million five weeks later, growing 180% a year, according to Rauch. Vercel now handles about 1.25 billion deployments every four months, after taking 10 years to reach its first billion.
  • Gerstner and Solomon both urged a slow, deliberate process at Nvidia after Jensen Huang’s resignation as CEO, and Gerstner added that “Jensen Huang is not above the law.”

Session by Session

Retail, AI and the American Dream: Gavin Baker, Antonio Gracias and Ron Biscardi

The summit opened with its three hosts in conversation with CNBC’s Sara Eisen, who noted that the idea “started on X” and ended up in a basketball arena. Baker told the origin story: a prestigious bank put a strong sell and a big price target cut on a semiconductor capital equipment stock, which should have fallen 5% to 10%. Instead it rose 7%, because Citrini, a Substack and X writer he counts as a friend, had put it in a basket and written one line about it. To Baker that crystallized six or seven years of investing becoming democratized, and he called that a good thing. He also passed on a point he credited to Gracias’s team: in China people are overwhelmingly positive about AI, while Americans are scared of it. He said the event is not meant to pump AI stocks but to get the industry to tell the truth about how AI can make ordinary Americans’ lives better.

Biscardi explained why iConnections, a capital introduction platform built for sovereign wealth funds, endowments and foundations, took on its first retail event, and why it went to an arena instead of a hotel ballroom. Gracias framed the mission as reviving the American dream. He pointed to record low poverty alongside wages that have barely moved since before Covid, and argued that past golden ages from Athens to the Renaissance “were all built on human misery” while this one could be built on well-being: more time with patients for doctors, help for caregivers, faster science. His shorthand was moving Americans from “P doom” to “P boom,” and the panel’s tagline was that AI will be Star Trek rather than Terminator if we make it so.

The back half was Baker’s market view. He said equities broadly trade at 10, 15 and in some cases 20 year low multiples, which tells him the market believes this is peak spending. The test he applies to management teams is whether the elasticity of token demand stays positive as AI gets more useful. If it does, the build-out looks like the railroads, redistributed across what he calls the five-layer cake of land, power, shells, chips and compute. He said two or three big cards would turn over that week, naming Nvidia’s earnings that night and Anthropic’s IPO. The hosts closed by noting more than 100 private companies at the event, almost all represented by their CEOs, and more than a thousand scholarships for X, Substack and Reddit contributors.

Brad Gerstner: Why Anthropic’s IPO Is Oversubscribed

The Altimeter founder’s session, moderated by Biscardi, is the video at the top of this post. Gerstner said the setup could hardly have been worse for a giant IPO: a semiconductor selloff, an election that had people “scared” about data centers, and Jensen Huang, in his words, “pushed out” as Nvidia’s CEO the week before (Nvidia announced on November 12 that Huang had resigned and that CFO Colette Kress would serve as interim CEO). He would have guessed the deal might be pulled. Instead he described it as roughly 10x oversubscribed and likely the largest order in Altimeter’s history, because Anthropic addresses the largest market in tech history and is diffusing faster than any technology before it. He added that reaching $1 trillion in revenue would take only about 4% of global knowledge work.

His main answer to the skeptics was unit economics. Bill Gurley, his podcast partner, spent three years telling him Anthropic would never be profitable because of its compute spend, and Gerstner admitted he “didn’t know the answer.” Now, he said, the company is near a $100 billion run rate with 5,000 people, compute costs fall about 90% a year while what Anthropic can charge has upside as its models get more capable, and only 26% of 2026 compute served customers. With consensus calling for another $100 billion of revenue next year, he argued Anthropic would need only about 500 megawatts to serve it. Biscardi said the deal would price the following night and trade Thursday. Gerstner warned the room that almost every IPO falls sharply from peak to trough in the year after its peak, recalled watching his biggest Google position drop 10% to 15% in a day in 2005, and told people who cannot stomach that volatility not to buy.

He then walked through his venture framework. In 1999, while helping start General Catalyst, his rule was search and then commerce, with everything else a quick no, an idea he credited to Essentialism. In 2012, when most investors thought the cloud was insecure and expensive, Altimeter took the other side and, he said, turned that fund into a more than 50x return with Snowflake, MongoDB, Twilio and Okta. On Nvidia, he and Baker had written to the board asking for a slow, deliberate process. He said he does not know the facts, that Huang has acknowledged personal transgressions, and that political pressure was producing a “massive rush to judgment,” while stressing that no founder is above the law. He finished on Invest America: $1,000 in the S&P 500 for every child at birth, $10 a week growing to about $1 million by 55, a SpaceX share per account that he said Elon Musk is adding, and a comparison he clearly believes, that the Invest America Act will be to the age of AI what Social Security was to the industrial age.

Stan Druckenmiller on AI, Markets and the Next Ten Years

Interviewed by Morgan Stanley’s Iliana Bouzali, Druckenmiller opened by giving Baker credit. On a Zoom call of macro investors about a month after ChatGPT launched, a man he had never met told the group they were arguing over nickels and dimes while missing the biggest thing in the history of American business. In January, Druckenmiller said, he had unusual clarity: loose policy, commodities and market breadth at highs, tight credit spreads, and excitement about Claude Code among people close to AI, so he played big and let AI infrastructure clean out the rest of his portfolio. Now he owns “basically almost none” of that book. The Fed is hiking after what he called a regime change, the advance/decline line peaked on August 14, credit spreads are widening, and Duquesne is short copper.

His method is the most transferable part of the talk. Duquesne does not use VaR or stress tests. He reads his own P&L as the signal, believes in hot and cold streaks the way a baseball hitter does, and trades bonds, currencies and commodities as well as stocks so he is never forced into equities. He said his record of no down calendar years was “the luck of the calendar,” hiding big drawdowns within years. On AI he wants use cases, not infrastructure, and his favorite use case is a lab: whoever has the best intelligence can sell tasks instead of tokens and build its own businesses, beginning with cybersecurity. Asked what is happening quietly, he named the decline of K-12 education, an immigration policy that could cut off the talent that runs most big tech companies, and a fiscal path he expects to end in a market-driven reform within ten years.

On investing temperament, he repeated a line he says he took from Mark Twain: put all your eggs in one basket and then watch the basket very carefully. Great investors, he said, are passionate, curious, disciplined and “competitive to the point of a sickness.” His own process is “invest, then investigate”: he takes a position “big enough to really hurt if I’m wrong” with about 15% of the information a traditional investor would want, then gives research two or three weeks to tell him whether to add or get out. His costliest mistake was breaking that discipline in 2000, selling the Nasdaq in January and buying it back on the day of the top out of FOMO. He closed with ten-year predictions, including his view, after 30 years on the board of Memorial Sloan Kettering, that cancer becomes a managed disease within ten years.

Goldman Sachs CEO David Solomon: Have We Seen the Peak in AI?

Baker interviewed Solomon, thanking Goldman for being one of Atreides’s two prime brokers at launch. Solomon described a 157-year-old firm with about $115 billion of capital supporting a $2.3 trillion balance sheet, more than $6 billion a year of technology spending, and revenue that has roughly doubled in eight years. He singled out the ultra high net worth business, roughly 45,000 clients with $25 million or more each, as “a people first business enhanced by technology.” On markets he said he has no idea about the next three to six months, is very optimistic about the next ten, and is not worried by a 5% ten-year Treasury because it sat above 5% for most of his life. He and Baker also agreed that US politics keeps swinging because the middle keeps moving, with control of the House, Senate or presidency changing hands in 11 of the last 12 two-year elections.

On Nvidia, Solomon said he agreed with “Brad’s letter”: move slowly and thoughtfully, because he knows very few facts and Huang is going to testify. Baker praised Kress as an exceptional CFO and a capable interim CEO. On the title question, Solomon said we are still in the early innings of a long cycle. Estimates of $6 trillion to $10 trillion of US AI infrastructure investment get recalibrated as information arrives, the current moment is one of those recalibrations, and multiples are more attractive than they were six months ago. The biggest value, he said, will come in the application layer over the next 10 to 20 years, and adoption will take time because technology moves faster than enterprises and people.

Inside Goldman, the newest partner class has the highest share of engineers ever, and the firm is rebuilding eight core processes, including client onboarding, know-your-customer checks and anti-money laundering, from the top down. Solomon expects fewer but more senior engineers in five years and said the gains should fund growth, not margin expansion. He closed with three examples: a Mayo Clinic AI model that flagged 73% of pancreatic cancers in scans taken up to three years before diagnosis, a $500 million next phase of Goldman’s 10,000 Small Businesses program that includes an AI institute with Google, Anthropic, OpenAI and Meta as partners, and his view that AI is the internet’s democratization of opportunity “on steroids.”

Vercel CEO Guillermo Rauch: More Software Is Being Built Than Ever Before

Rauch, interviewed by Glenn Solomon of Notable Capital, an early Vercel investor, described Vercel as the infrastructure behind “almost every agentic interaction”: when a coding agent is asked to ship an app, more often than not it deploys there. He told his origin story, a self-taught developer from the suburbs of Buenos Aires who came to the US at 19, and said the hardest part for him, learning to program, is exactly what coding agents now do for everyone. Vercel took ten years to reach a billion deployments and now does about 1.25 billion every four months. He said many companies adopting Vercel have “basically stopped buying bespoke software” because they can generate it, and called hand-written software “almost a liability,” a line he admitted was controversial.

On security, he said the opposite of “problem solved.” Agents are currently finding latent vulnerabilities in kernels and operating systems, and the answer is to “combat bad tokens with good tokens.” Vercel’s security review agent scans code on every push for problems such as poisoned open-source dependencies, and he said it found 1,000 vulnerabilities in vibe-coded software within a day. His advice to a young founder today is to build the agent before buying the domain name. Glenn Solomon put numbers on the business: about $6 million of annualized revenue when he first invested roughly six years ago, more than $600 million as reported by The Information on October 1, and, Rauch added, $700 million five weeks later, growing 180% a year with nearly 600,000 paying customers.

Rauch’s ambition is to capture “every pixel and every token,” and he believes today’s “internet of tokens” is perhaps 0.1% of what it will become. He said the only way to build a generational company is to go public, a contrarian view in San Francisco, and compared Vercel’s role for developers to Nvidia’s, though he gave no timeline. He closed on optimism with a twist. Every industry will be disrupted, AI “first came for software engineers,” and there is always a bear case for a new technology, but he urged people to use AI to get smarter, to keep reading the code, and not to become “a useful idiot.”

Notable Quotes

“I think retail is now the most powerful force in the market.”

Gavin Baker, on why TomorrowX exists

“Because the one thing no one wants is a bubble. That would be terrible. And we’re certainly not in a bubble right now.”

Gavin Baker, on why a slower AI build-out is good news

“Own the shit that gets better as intelligence gets better.”

Brad Gerstner, on his one rule for investing in AI

“I can’t remember a period where I had less clarity on where we’re going to be in 12 months, unlike last January.”

Stan Druckenmiller, on why he is in capital preservation mode

“Selling tasks, no tokens. You heard it here first.”

Stan Druckenmiller, on where frontier labs are headed

“I think the big value creation over the next 10, 15, 20 years is going to be in the application layer, not the infrastructure and the initial models.”

David Solomon, on whether AI has peaked

“You should not become a useful idiot. You should become the smartest person you can be in the world because of AI and leverage it in everything you do.”

Guillermo Rauch, closing his session

Watch all of the sessions on the iConnections YouTube channel, starting with Brad Gerstner’s talk, and find the program on the TomorrowX Summit site.

Related Reading