NVIDIA Predicts $1TRN in Revenue: Everything You Need to Know From GTC & Anduril Lands $20B Contract

Summary

• Nvidia’s “$1 trillion in demand” headline moved the stock less than 1% because, as Rory O’Driscoll decomposed it, the number is just analyst forecasts restated: $215B revenue last fiscal year, mid-$300s forecast this year, mid-$400s analyst forecast for ‘27 — add them and “apply a salesman’s round up, you get to a trillion.” The real GTC statement is that unprecedented capex continues four or five more years, and Rory assigns “at least some probability, pick a number 30%, that doesn’t happen this way.”
• Jason Lemkin’s bet on top: cumulative $10T in Nvidia revenue within ~5 years of the first trillion, requiring roughly three orders of magnitude more inference — “might be 3,000 times more tokens, not 3x.” Rory’s caveat cuts the other way: “You could have 3x more tokens, but if the price per token goes down by 6x, the revenue will decline.” Jason’s theory is that Nemo Claw and likely the Grok acquisition help keep tokens burning — “at least three agents running 24 hours a day.”
• Vibes are now a spread trade: “this is summer at Nvidia… firing on about 13 cylinders” versus “you can smell OpenAI struggling right now” — code red, refocusing on enterprise, stopping side projects.
• On layoffs (Atlassian 1,600; Meta reportedly 20%, or 16,000 of 79,000), the tradeable insight is Rory’s fourth category: Meta’s operating margins are still 40% but free cash flow honestly accounted for capex is almost zero, so depreciation forces the swap of humans for GPUs. “Today compute eats jobs. You literally can’t afford to have Nvidia and people.”
• Jason’s 2026 hiring test: “What commercial AI tool have you brought into your organization this month? That’s the test.” His coinage — agentic deployment expert (ADE), “from C-level to junior. Don’t hire anybody else” — and the democratizing flip side: “You do not need to be technical to win with AI agents in Q2 of ‘26. You do not need to be even 1% technical.” Or: “you’re going to get laid off because you’re not going to matter.”
• Anduril’s $20B, 10-year Army contract (5-year base + 5-year option, consolidating 120+ contracts) is procurement crowning “the clear new prime” for Lattice, its real-time connectivity layer. But Rory caps the euphoria: defense is only a bit over 3% of GDP, almost half of it people — “I think there’ll be five or six big winners in defense. I’m not sure there’ll be 100.”
• Seed math is the episode’s grimmest call: Jason has abandoned his decade-old small-TAM-plus-great-founder thesis, and with YC “productized to 60 million dollar post,” a 100X net of dilution means ~250X — a $13-20B outcome, when fewer than 50 public tech companies clear $20B. “The math is grim,” which is why $50-100M seed funds could be the worst-performing size of this vintage — and why Barton Biggs applies: “There’s no investment opportunity so good that excess capital won’t destroy it.”
• Travis Kalanick’s return with Atoms drew agreement on substance (robots on wheels, not humanoids) and a bold counterfactual from Jason — “Travis Uber would be a trillion-dollar company today because it’d be 5 years ahead of where it is today” versus $160B now. But neither would fund Atoms at the ~$20B he’s reportedly seeking from their own vehicles, while both would at General Catalyst/Coatue scale: “fund size is strategy.” Adobe rounds out the bear file — Shantanu’s announced exit before a successor plus the verdict, “I see no evidence that Adobe will grow. Nothing.”

Deep dive

1. Nvidia’s trillion was already in the analysts’ spreadsheets

• Rory’s decomposition of the headline: Nvidia did $215B revenue last fiscal year (up from ~$130B), the forecast is mid-$300s this year, and the analyst forecast for ‘27 is mid-$400s. A year ago the soundbite was half a trillion of demand over two years; now it’s a trillion — but it includes ‘27. “It turns out if you add 500 and 400 and then you apply a salesman’s round up, you get to a trillion dollars.” Translation: “the analyst forecasts for the next two or three years look roughly right” — so the stock moved less than 1%. No new information.
• The context that makes flat guidance still remarkable: five years ago this was a $20B company — 10X growth over four years, ~60% growth last year, 60% forecast this year, attenuating to 20-30%. “Not insane… if past performance is predictive.”
• Jason’s read of the tape was about energy, not numbers: data centers in space, Nemo Claw (“their version of Open Claw” — per Jensen, the most GitHub stars per unit time, faster than Linux), an open-source LLM partnership with Thinking Machines, Grok already integrated, a trillion in bookings crossed — “firing on about 13 cylinders.” The contrast: “You can smell OpenAI struggling right now… the code red,” refocusing on enterprise and stopping side projects. “Man, this is summer at Nvidia.”

2. The real GTC bet: five more years of unprecedented capex — with a ~30% chance it breaks

• Rory’s arithmetic on what the trillion implies: Nvidia captures about half of AI capex, so $200B of Nvidia revenue means $400-500B of capex; $600B implies ~$1.2T. “These are unprecedented levels of CapEx spend, and now we’re forecasting them to keep going for four or five years. There is at least some probability, pick a number 30%, that doesn’t happen this way.” Nothing yet says it won’t — but it’s “a pretty heroic assumption,” railway-boom scale.
• Jason’s order-of-magnitude wager: cumulative revenue goes from $1T to $10T in about five years, which needs “three orders of magnitude more inference” — “might be 3,000 times more tokens, not 3x.” Rory’s inverse: “You could have 3x more tokens, but if the price per token goes down by 6x, the revenue will decline” — and whether cost declines eventually hit an inverse Moore’s law is an open question (“maybe that’s why we need data centers in space”).
• Why does Nvidia even ship Nemo Claw? Jason’s theory is that it burns tokens — same reason “everyone in China is giving away open claw… you’ve got grandma and grandpa lined up on the streets outside of Tencent Alibaba with their free open claw.” Jason’s paraphrase of Jensen’s implicit aim: “we want you to be burning tokens at least 72 hours a day. We want at least three agents running 24 hours a day.”
• The message behind the message: whichever models win, everyone needs gigawatt data centers — “most rolls of the dice lead to Nvidia winning… most of the roads lead back to Nvidia.”

3. Layoffs decoded: five kinds, and the Meta kind is compute eating jobs

• Neither Atlassian (1,600 cuts) nor Meta (reportedly 20% — 16,000 of 79,000) has to do this; both throw off cash. Jason: “This is a purposeful decision” happening in every boardroom — “I just don’t know what to do with half of these people… I do need people. I need different people.”
• Rory’s taxonomy, worth keeping whole: (1) we over-hired and AI is the excuse (Block, growing 2%); (2) growth fell from 20% to 2% and “Wall Street is simple. If you give them growth, they’ll leave you alone. If you don’t give them growth, you better give them profitability. And if you don’t give them either, they’re going to bust your chops”; (3) genuine AI efficiency, “probably true in coding,” less sure elsewhere; (4) Meta — margins still 40% but “the free cash flow when you honestly account for the CapEx is almost zero,” so “today compute eats jobs… You literally can’t afford to have Nvidia and people”; (5) Jason’s addition — the reshuffle: fire 20 C++ engineers, rehire “eight engineers who are just really awesome at AI,” maybe at twice the salary.
• The emotional register matters: “everyone’s stressed AF right now, including OpenAI, including Anthropic… You’re going to be obsolete in 18 months.” Even the hot names — “unless you’re likely Legora, likely Sierra, or likely Lovable, but even they know they’re running out of time. In 12 months their current products will be obsolete.”
• Jason’s contrarian defense of 2021: teams weren’t bloated — they were right-sized for a world where 60% growth required humans for everything. Had that world continued, “every office in Soma would be triple packed.” What changed is the need itself: “We used to need warm bodies. That’s what’s changing.” The new companies are efficient because “they’re just not brute forcing things with humans.”

4. The 2026 hiring test: what tool did you bring in this month — hire only ADEs

• Jason field-tested this on a favorite exec at a startup crossing $100M and found the answer six months stale. Looking at what tools people “play with” was the summer-2025 answer; the 2026 answer: “What commercial AI tool have you brought into your organization this month? That’s the test.” One tool a month, deeply evaluated even if not bought, in every role — sales, marketing, product, QA. Candidates either can’t keep their AI usage “from spewing out of their mouth or they’re staring at you. There’s nowhere in the middle.”
• The cautionary tale: a “super hot AI company valued at 6 billion” whose agent told Jason’s team to quadruple their spend — because nobody had trained the agent on its own pricing. Beta launched a year ago. “These are people asleep at the wheel.”
• Title inflation is accelerating — prompt engineer died, go-to-market engineer “is going to die,” forward-deployed engineers are insatiably demanded (Palantir just cut deployment times over 90% with them) — so Jason coined the durable one live: “agentic deployment expert, ADE… That’s your job from C-level to junior. Don’t hire anybody else. You’re going to regret it.” By his count, maybe 30% of management teams at even his best companies meet the bar; in general interviews, single digits.
• The democratizing flip side both agreed on: “You do not need to be technical to win with AI agents in Q2 of ‘26. You do not need to be even 1% technical” — if you’ve personally deployed Salesloft, Outreach, or HubSpot in the last 3-5 years, “you can deploy any agentic tool. Any.” The only non-intuitive part is training the agent. Rory extended it to investors, via his partner Andy’s line: “if you decide you want to stop learning new things, you probably should retire within 6 to 12 months” — now “maybe that’s down to 3 to 6 months.” The alternative: “you’re going to get laid off because you’re not going to matter.”

5. Anduril’s $20B isn’t a new program — it’s the Army crowning a prime

• Jason’s deflation of the headline: the 10-year deal (5-year base, 5-year option) consolidates 120+ separate contracts into one enterprise contract — “more of a procurement thing.” And scale it properly: “they only have four or five customers. So you better get 20 billion from each of them if you want to be a big company.” What it really signals: “It kind of picks them as the clear new prime.”
• The product logic centers on Lattice, the software connectivity layer tying together hardware from Anduril and others. Recent conflicts show why real-time matters: “if you’re in the Strait of Hormuz, you’ve got literally seconds before you can take down an incoming drone. You don’t have time for a slow connectivity protocol. You definitely don’t have time for a human.” The bigger trend: the Pentagon has bought cost-plus for 40 years, “wildly inefficient,” and Anduril sells Silicon Valley-style — build the product up front, sell per unit.
• Rory’s sizing caution against defense euphoria: defense is “a little over 3% of GDP” and almost half of that is people; the Pentagon’s budget for new programs is fairly finite. “I think there’ll be five or six big winners in defense. I’m not sure there’ll be 100.” Jason’s retort: “But isn’t that the game of venture? It’s four or five, not 100.”

6. Jason kills his small-TAM thesis — and the seed math turns grim

• A ten-year investment thesis publicly abandoned: Jason had a smallish-TAM-plus-great-founder thesis — he says, “when I started in Ecity Ventures, the TAM was $2M”; he points to DocuSign as an example of a TAM that grew — but “in my heart and soul, I can’t do any of those investments anymore… I can’t even bring myself to take a meeting with a startup where I don’t believe the TAM will be utterly massive.”
• He concedes the failure mode this creates: everyone swings at mega-TAMs, funds the number three or four player, and pays up — “100 pre for these seed investments” — yielding “zeros after zeros because there isn’t a chance to stair-step your investment.” That’s the mechanism behind the episode’s framing that $50-100M seed funds could be the worst-performing size of this vintage.
• The arithmetic, done live: YC has “productized to 60 million dollar post”; a 100X after dilution “is 250X in today’s world”; 250 × $60M post is $13-14B, “round up to 20” — and fewer than 50 public tech companies have market caps north of $20B, fewer than when the podcast started. “The math is grim.”
• Rory’s pushback: “the thing about power laws is they get in your head” — like gamblers on tilt, missing OpenAI leads people to “fund eight next-gen foundation models,” and “we might be seeing that happen right now.” Beyond town size there’s town velocity and your ability to dominate the town: “you can make really good coin in a 4-5 billion dollar market with a great outcome” — just not at power-law entry prices. His standing quote from Barton Biggs: “There’s no investment opportunity so good that excess capital won’t destroy it.”

7. What still works: near-end traction plus a big-picture trend — and growth funds waiting for proof

• Jason’s honest concession about stage: “the growth fund is the winning strategy at the moment because they wait for the proof.” He wouldn’t have believed Decagon and Sierra’s numbers, or Intercom re-accelerating — but Rory doing the deal with proof points beats “hoping that a slide and a vibe-coded website proves it.”
• The counterargument to his own TAM pessimism: convergence. Per Owen [likely Decagon’s CEO, referenced from the prior week]: “there’s not going to be any difference between support and sales… all these agents are converging to a meta agent” — and if that’s true, “your TAM explodes.” The hard question: will you invest where there’s no evidence yet that AI is exploding the TAM?
• His humility case: Craft and others did Replit “at a billion in ‘21 pre-AI,” on roughly $8-10M revenue — “I ain’t that visionary.” But he distinguishes Replit-scale vision from today’s niche vibe-coding pitches — “this little nit fixing the fact that the icons all look like cloud artifacts, I just don’t believe.” The bar: “you better show me something hyper disruptive, like my jaw falls on the ground.”
• The operating rule for the middle stage: “big picture trends and near-end traction. We don’t want to compromise between the two.” Anduril in 2018 was exactly that — border watchtowers as near-end traction, cost-plus disruption as the trend. The catch: “anyone can use words… the trick in investing is to figure out which of those words is [real] and which is not” — squint one way, 2021 Replit is a niche tool; squint the other, it’s the democratization of software.

8. Travis is right about wheels — and the trillion-dollar Uber counterfactual

• The setup: Kalanick surfaced after eight years in stealth with ~1,000 employees, rebranding City Storage Systems/CloudKitchens into Atoms — “gainfully employed robots” for food, mining, transport — declaring “I bled, but I did not perish,” plus an investment in Pronto, the autonomous-driving company founded by likely Anthony Levandowski.
• Jason’s robotics call, which he endorses on the merits: don’t build humanoids — “It’s not clear if you’re building an industrial machine for a lot of use cases that you add legs.” Legs drain battery and are unstable; factory and warehouse work wants wheels. Evidence he cites: Sunday, the recently raised home-robot company, also chose wheels. It’s effectively a call against the humanoid cohort — “not saying it’s not going to happen ever, but it might take a lot longer than you think.”
• Jason’s counterfactual, stated without hedge: “I think Travis Uber would be a trillion-dollar company today because it’d be 5 years ahead of where it is today” — versus $160B now. Travis said from the beginning “our business is dead at its terminal state” and wanted autonomy; he’d also have weaponized capital and “some dark arts in your mobile app” to take 90% of food delivery rather than buying his way in. “That guy would be running a trillion-dollar company today if Gurley and buddies didn’t force him out.”
• The rebuttal, which Jason partly accepts: the claim refutes its own founder-worship — “implicitly what you’re saying is it’s a great founder plus a great opportunity.” Uber had to get public and cash-flow positive; autonomy was ten years away; and “Uber seemed feature complete about 2 years after it launched,” which is why outside management could run it. The synthesis both land on: “they should have done the Steve Jobs thing” — swap him out, get it public, “and sometime in ‘22 should have got him back and said, now is the time to do autonomy.”

9. Would you fund Atoms at $20B? Fund size is strategy

• First, the removal calculus, from Rory who’s done it: “It’s like open heart surgery and 50% of people die… It is easier to just lose money.” Only two justifications — business decisions literally heading to bankruptcy with a founder unwilling to change course, or behavioral issues rising to “a really systemic problem with a high bar.” Given Benchmark’s founder-friendly posture, “you got to believe that one or two of those issues was on the table.”
• On the round itself — last round at $15B, and “Claude thinks he’s looking for a flatter up round, so up to 20 billion” — Jason passes: parts of the interview “felt like I was back in 2017,” and the real question is “do I think he’s past it.” At $20B he’d need “100% conviction he can still do it” including that Travis is “a thousand percent in for doing this for 20 years” at 48. But if he were General Catalyst raising $10B, or Coatue? “Do I want to chuck a couple of hundred million into Travis, one of the greatest founders of our time? Yes.”
• Rory named the pattern: “everyone’s a victim of their fund. If you’re a seed stage fund, you assess the guy. If you’re a growth stage fund, you assess the opportunity to put quantities of money to work.” And the sharper version: if your logic reduces to “I’ve got to get rid of this much money this month and this entrepreneur is amazing… let me give you some advice: halve your fund size [likely]. That’s why you’re not playing the AUM game.”
• Jason’s closing flag carries real history — he once answered a Quora question naming Kalanick the best entrepreneur he’d ever met, from a Red Swoosh lunch (down to two employees, he thinks, pre-YouTube) where Travis “explained the entire future of video on the internet… My jaw was just on the ground.” His rule since: “my investing mistakes are when I’ve invested below that line.” His hesitation now: “He might be a little bit in the past as well as the future. That’s a flag for me at the rate that I have to win.”

10. Adobe: keys on the table, and no evidence of growth

• Rory insisted on sequencing precision: Adobe beat earnings and announced the CEO’s resignation without a successor simultaneously — then the stock tanked. Jason, an Adobe alum, reads it as voluntary: David Wadhwani being passed over again “suggests the odds that he becomes CEO are less than 100%,” and after 18 years Shantanu “left the keys on the table” — not fired, “for a million reasons.” The kicker stat: Adobe’s aggregate returns now sit just below the S&P over the last decade.
• Mid-conversation, Rory checked and revised: Shantanu stays as CEO until a successor arrives — so it’s “an announcement without an answer,” which fits a board pre-empting activists: “please don’t call us, Elliott… We know we need to make a change. Here it is.”
• The growth verdict is the tradeable part. Against the argument attributed to Sriram Rajaram that Adobe and Intuit are near-immune because SMB pricing and motion are so hard to replicate, the verdict is: “Just because your nominal churn is low does not mean you’re going to grow, and I see no evidence that Adobe will grow. Nothing.” Rory splits the pair: Intuit’s automation of tax and accounting has bounded AI risk, but Adobe is “the classic creator tool” in a market where “most of the AI traction so far has been individual users, creators” — “the Adobe AI risk over the next 5 years is pretty large. And that probably should be figuring into their search.”
• The wider epidemic, per Harry’s wager — “I bet 10 Shantanus stepped down in 2026” [likely; name garbled] — with Dustin Moskovitz quitting Asana with “no successor, no anything” as precedent. Rory softened it: “I don’t know if 10 quit this year, but I can totally see more than 10 saying, ‘Should I quit this year?’” The tell is Alex Karp, crushing it from a $50M Miami mansion: “the correlation between winning and having fun is pretty damn high” — and nobody running a low-growth public software company is having fun.