Brad Gerstner: No Bubble, Semis Carry the Nasdaq and AI's Takeoff Test

The All-In stage felt less like a celebration and more like a balance-sheet review. Brad Gerstner, founder of Altimeter and a five-time company builder, walked in with a rapid-fire market and tech check: where we are, where we are heading, and what has to prove true for the market to keep working. He nodded quickly to two personal moonshots — making every child a direct owner of capital and a $100, 15-minute heart scan he wants to turn into a heart mammogram — then moved straight into the speed round. The takeaway was plain: 2023-2025 was won with one call, 2026 will be won month by month on verified receipts.

Scoreboard: a rally paid by earnings

The scoreboard looks strong and narrow. The market is up about 15% this year and 39% since January of last year, despite tariffs, geopolitics, and AI regulation noise. Gold is flat, Bitcoin down about 10%, while Nvidia revenue doubled, hyperscaler capex doubled, OpenAI and Anthropic valuations doubled, and SpaceX rose two and a half times — all against a rough backdrop. Gerstner's frame is deliberate: this is an earnings-driven expansion, not multiple expansion. Earnings are up 26% on AI infrastructure, while multiples on the Nasdaq and S&P have actually tightened.

Not a bubble: 14 times GAAP

His bubble test fits in one line: Nvidia trades at 14 times next year's fully taxed GAAP earnings. The Nasdaq, the S&P, and the SOX all sit below their long-run average multiples; the Mag 7 sits about in line. In 2000 the market priced revenue multiples on companies with no revenue; today it anchors to taxed, auditable earnings. The other side of that strength is concentration. Consumer discretionary, software, and financials have barely moved. Semis account for about 70% of the Nasdaq's return — concentration that is both engine and fragility.

Who is getting paid? The token makers

That is why he follows the money. On his chart, blue is hyperscaler capex and orange is semiconductor free cash flow, tracking dollar for dollar — every dollar spent turns into cash for the infrastructure makers. The makers of the tokens are getting paid while the buyers of the tokens ride along in a tight market. The result looks like venture returns inside public equities: Dell up 5x and 9x in just 18 months. The build is the largest capex and super-cycle in tech history, and its rent must be covered persistently.

The break: monthly receipts

He returns to a conversation from last October with Sam Altman. The question was simple: how do you commit to a trillion in capex on $13 billion of GAAP revenue. The answer was blunt: sell your shares. Then early December brought Opus 4.5 and Claude Code, and Anthropic's revenue printed $2 billion in January, $4 billion in February, and $11 billion in March. For Gerstner that March print lit the fuse — proof that revenue had shown up — and powered the historic rip off the bottom in April and May. June and July then consolidated when the annualized run rate came in at $65 billion against expectations near $75 billion and open-source catch-up fears resurfaced. The curve itself is without precedent: hitting $1 billion once took years and put you in the top 5% of software; now the debate is billions per month.

That is why he calls monthly lab revenue the single most important data point in the market today. The top three labs together, on July chatter, run at about $100 billion. To keep the trade intact Gerstner argues they must reach at least $180 billion by year end — another $80 billion across the group. Is the monthly number closer to $4 billion or $8 billion. The market is pricing an answer that would have sounded impossible just a year ago.

Blue and orange bars: the rent math

The next slide makes the math visual. Blue is the Mag 5's expected capex, orange is the offtake revenue needed to pay the rent on that capex. Microsoft, Google, and Amazon do not build to own; they build to rent. If the renter does not pay, the build stops. If the year exits around $200 billion of run-rate revenue, Gerstner argues the path must then climb to roughly $450 billion, then $800 billion, then toward a trillion to stay in balance. Otherwise the scale cannot be built.

The physical counterpart is gigawatts. Citing SemiAnalysis and Dylan Patel, 2026 additions were about 19 gigawatts, roughly 7 gigawatts to the two leading labs, and 2027 is forecast at 43 gigawatts, about 14 gigawatts to the leaders — adding in one year as much compute as the entire United States holds today, with more than half of national compute sitting with two labs by 2028. That is why he pushes back on 43 gigawatts and marks a realistic 25 gigawatts for 2027, half to Anthropic and OpenAI. Anthropic is said to do $100-110 billion this year on about 1.5 gigawatts, so adding four to five gigawatts should be enough to add another $100 billion of revenue. The binding constraints are permitting and local opposition, grid delays, skilled labor shortages, and power equipment sold out; he invokes the shutdown of 67 fission reactors as a warning not to repeat that mistake on AI.

Demand: token flood and margin lever

On demand he is unconcerned. The total addressable market for knowledge work — consumer plus ads plus coding plus millions of enterprise white-collar workflows — is, in his words, the largest in history, where capturing just about 4% or $1.2 trillion would fund the build. Tokens tell the same story: about 47 quadrillion tokens this year, Codex users up 40x in eight months, median enterprise spend up about 17x in 18 months. The productivity lever shows up in margins. From 2015 to 2025 Nasdaq EPS grew about 10%, six points from revenue and about 38 basis points per year from margin expansion. The question now is whether AI can lift that 38 to 100 basis points. Every management team he cites — Uber growing 20% without adding headcount, Snowflake 30% without adding headcount — points to margin expansion through slower hiring, not layoffs, plus a new consumer agent category in every pocket that could be a trillion-dollar market and a massive token sink, from booking a hotel with Muse or Instinct onward.

Three brakes: rules, atoms, rates

He stacks three risks clearly. Rules are a tug of war that must land in pragmatic middle ground — his line is that the answer will not be at either extreme — and he praises a peer-review idea floated by Elon Musk the prior day, while stressing that fear plus activism can overregulate as it did with nuclear. Atoms and energy are hard: standing up 43 gigawatts would be the largest build in the country's history. Rates are gravity: he puts the chance of hikes the next day above 90%. All this capex is borrowed, so a higher hurdle rate pressures both data centers and equities. He quotes Buffett that rates are to stocks what gravity is to matter — if risk-free pays 5.5% to 6%, equities face a headwind, and a 10-year at 5.5% would be a heavy burden.

Flight path and fan of outcomes

He closes with a flight path, not a forecast. Up about 15% on the Nasdaq, the question is whether to be small, medium, or large. Monthly lab revenue near $8 billion means takeoff and likely an IPO this year; near $4 billion means the trade breaks. Rates, the election, and oil are the second axis — rates track oil to some degree — and regulation plus the Anthropic IPO are the third. Recent softness on Anthropic IPO timing already weighed on the tape. His own posture is medium and mentally flexible: if receipts print strong and oil pulls back, he will add chips; if not, he reserves the right to go smaller. The last line is a regime reminder: 2023-2025 rewarded anyone who shoved chips into the AI trade. In 2026 everybody knows the thesis and it is priced. What is left is facts and circumstances — stay flexible, follow receipts, do not go leveraged and do not gamble on foreign exchange.