Summary

Gerstner’s core claim: this is an earnings-driven market, not a bubble — multiples have actually contracted. The market is up 15% this year and 39% since January last year, with earnings up 26% while the Nasdaq and S&P multiples fell; NVIDIA trades at “14 times next year’s fully taxed GAAP earnings.” “This is no bubble like it was in 2000” — but semiconductors are 70% of the Nasdaq’s return, and consumer discretionary, software and financials have barely moved.
The single most important data point in the market is monthly AI-lab revenue. After Claude Opus 4.5 and Claude Code in December, Anthropic’s monthly revenue went from $2B (January) to $4B (February) to $11B (March), lighting “the fuse” for the April–May rally. The top three — Anthropic, OpenAI and SpaceX — have roughly $100B of collective run-rate revenue, based on July rumors, and Gerstner says they must reach at least $180B by year-end, adding another $80B “just to keep the AI trade intact.”
The offtake math is the whole game: ~$1.5T/year of capex needs someone to pay the rent. Microsoft, Google and Amazon are “building it to rent it,” so offtake revenue must go from roughly $200B exiting this year to $450B, then $800B–$1T over the next few years, or “we can’t build this much capex.” He’s not worried about demand: knowledge work is “the largest TAM in the history of the world,” and capturing about 4% — roughly $1.2T — would cover the capex.
He calls Dylan Patel’s 43GW-of-new-compute-next-year forecast too aggressive — expect about 25GW. Permitting, grid interconnection, labor shortages and sold-out power equipment constrain the buildout, but half of that 25GW going to Anthropic and OpenAI is still “enough to generate the revenue”: Anthropic’s reported $100B–$110B this year is being generated with about 1.5GW, so adding 4–5GW could add another $100B.
Three risks: regulation, power and rates — with nuclear as the cautionary tale. Activist-driven fear shut down 67 fission reactors and “we unilaterally disarmed against China”; he says the same must not happen to AI. He puts the odds of rate hikes tomorrow above 90% and warns a 5.5% 10-year yield would be “a big burden” on equities — “interest rates are to stocks what gravity is to matter.”
Positioning: medium, mentally flexible and emphatically no leverage. In 2023–25, “you only had to get one thing right” — that AI would be the biggest technology supercycle — and shove chips into the AI trade. In 2026, “it’s all priced now. It’s about facts and circumstances.” If monthly lab revenue approaches $8B and oil retreats, he would add; otherwise, “we’ll reserve the right to go even smaller.” “Don’t YOLO” and do not go 4x levered in this market.

Deep dive

1. The scoreboard says earnings, not euphoria

Gerstner’s opening case against bubble talk: the market is up 15% this year and 39% since January last year despite tariffs, geopolitics and AI-regulation fears — while gold, a favorite in the “bestie” group, is flat and Bitcoin is down 10%. NVIDIA revenue and hyperscaler capex doubled; OpenAI and Anthropic valuations doubled; SpaceX rose 2.5x.
The load-bearing fact: “This is not about multiple expansion” — earnings are up 26% while the Nasdaq and S&P multiples contracted, and NVIDIA sits at 14x next year’s fully taxed GAAP earnings, below historic averages. “This is no bubble like it was in 2000.”
But the breadth warning is explicit: semiconductors are 70% of the Nasdaq’s return — “that’s both good and bad” — while consumer discretionary, software and financials have barely moved. Token makers capture the money; hyperscaler capex is “almost dollar-for-dollar” the free cash flow of semiconductor companies, producing public stocks with venture-style returns. Dell is up 5x, including a 9x rise in just 18 months.

2. Anthropic’s revenue lit the fuse — and must keep surging

In his October podcast with Sam Altman and Satya Nadella, Gerstner asked Sam how he could commit to $1T in capex with $13B of GAAP revenue. “Instead, he told me to sell my shares.” Then Claude Opus 4.5 and Claude Code arrived in early December, and Anthropic’s monthly revenue reached $2B in January, $4B in February and $11B in March: “an exclamation-point answer” to whether AI revenue would show up.
The summer consolidation followed Anthropic’s statement that its annual run-rate revenue was $65B, below the $75B some had expected, alongside concerns about open source catching up. The top three labs — Anthropic, OpenAI and SpaceX — have roughly $100B of collective run-rate revenue, based on rumors from July, and Gerstner thinks they need to reach at least $180B by year-end, adding another $80B “just to keep the AI trade intact.”
His framing of the scale: “these revenues have never happened before in the history of capitalism” and are on “parabolic, double-exponential curves.” A software company reaching $1B in revenue over four or five years used to be top-5%; now the key question is whether monthly lab revenue is $4B or $8B.

3. The offtake equation: $1.5T of capex needs a rent payer

The mechanism: Microsoft, Google and Amazon are not simply paying for the buildout — “they’re building it to rent it” — so offtake revenue must climb from roughly $200B of run-rate revenue exiting this year to $450B, then $800B or $1T, “just to keep up. Otherwise, we can’t build this much capex.”
TAM is not the constraint: knowledge work — consumer, advertising, coding, white-collar workflows and millions of enterprises — is “the largest TAM in the history of the world,” and capturing about 4%, or $1.2T, would pay for the capex. Evidence of demand includes 47 quadrillion tokens expected this year, Codex users up 40x in eight months and median enterprise AI spending up about 17x in 18 months.
The productivity dividend as margin math: from 2015 to 2025, Nasdaq EPS growth was about 10%, reflecting 6% revenue growth and roughly 38 basis points of annual margin expansion. Can AI turn 38 basis points into 100? “The answer is obviously yes,” he says, citing Uber growing 20% and Snowflake 30% without headcount growth. Humans and engineers are the largest cost input, and consumer agents — he cites Muse and Instinct in connection with his hotel-booking bet with Bill — could become another trillion-dollar category while consuming massive numbers of tokens.

4. Three risks: regulation, power and rates

On regulation, his answer to being accused of taking both sides is that neither extreme wins: the goal is “common-sense, pragmatic solutions that get my mom, my sister, and my brother off the cliff.” He points to Elon’s peer-review suggestion and warns against repeating the history of activists shutting down 67 fission reactors and leaving the country having “unilaterally disarmed against China.” “We can’t allow this to occur with AI.”
On power, he directly disputes SemiAnalysis’s Dylan Patel forecast of 43GW of new compute next year. About 19GW was added in 2026, while total US compute is currently below 40GW; permitting, grid-interconnection delays, skilled-labor shortages and sold-out power equipment make 43GW too aggressive. Gerstner expects closer to 25GW, with half going to Anthropic and OpenAI — still sufficient, since Anthropic’s reported $100B–$110B in revenue this year is being generated with about 1.5GW, and another 4–5GW could add another $100B.
On rates, he sees more than a 90% chance of hikes tomorrow. That raises the hurdle rate on the borrowed money used to build data centers, while a 5.5% 10-year yield would be a major burden on equities. Buffett’s formulation: “Interest rates are to stocks what gravity is to matter.”

5. The flight path: medium, mentally flexible, no leverage

His fan of outcomes: if monthly AI-lab revenue is closer to $8B, “it’s takeoff,” and he thinks an IPO will occur this year. Rates, oil prices and the election are key variables; regulation and the Anthropic IPO are additional concerns. The market traded down the previous day on fears that the IPO might be halted or postponed, which Gerstner does not expect but calls a major issue if it happens.
The regime change he wants remembered: from 2023 to 2025, “you only had to get one thing right” — that AI would be “the biggest supercycle in the history of technology” — and put chips into the AI trade. “That is not where we are in 2026. Everybody knows about AI. It’s all priced now. It’s about facts and circumstances.”
His current position is medium and mentally flexible. If revenue comes in strongly and oil retreats, “we’re going to put more chips on the table”; otherwise, “we’ll reserve the right to go even smaller.” The closing warning is categorical: “Don’t YOLO,” and do not go 4x levered in this market, as in his example of the friend up north who gave all his money to Citadel.

Satya Nadella on the AI Doomer Slowdown, Microsoft’s Master Plan & Who Wins AI