Summary

Gavin Baker sees the FT’s $2 trillion Anthropic IPO figure as potentially banker theater, not a reliable clearing price. Losers of the lead-left mandate “leak to the press” and set the bar high to make the winner look bad; he thinks $2T may be where you’d price to absorb the lockup, while his earlier 3–4T call is “where it might trade” — against a $100–120B exit run rate (16–20x sales) after three straight years of 10X growth.
Both Sacks and Baker converge on roughly $400–500B exit ARR for Anthropic in 2027, constrained by physical supply rather than lack of demand. Sacks: “the over-under really is 500 billion exit ARR… and I think they think they’ll do the over”; Gavin heard that at $60B ARR the company believed it could reach $600B within a year — and that Dario has said Anthropic “might be the only private company in the world,” hubris Gavin flags as “getting into SBF land a little bit.”
Sacks says OpenAI has pivoted to coding and reportedly reaccelerated to over 20% month-over-month growth with GPT-5.6 — a 10X annual pace matching Anthropic’s. Anthropic’s original coding bet came from watching Cursor’s utilization on its own platform (“We’re gonna take that business”), and moving vertically into what users build on you is now the industry playbook.
The episode’s ideological core: Gavin says Anthropic and the effective altruists think AI is “too dangerous to distribute”; Zuckerberg — and, he thinks, Elon and Jensen — think it’s “too dangerous to centralize,” and Gavin says “history has spoken.” Sacks adds the commercial irony: had Dario won his “FAA for AI” in Washington, the six-month frontier lead underwriting Anthropic’s pricing power “will be gone like that. Your model will be commoditized.”
Gavin’s counter to the bubble crowd: the overwhelming majority of tokens are profitable across the chain, and the Anthropic S-1 “is gonna break a lot of people’s brains.” Macro and value investors assuming subsidized tokens are like someone bearish because “oil’s at $500 a barrel… It’s just not.” Sacks frames Anthropic’s quarterly prints as the industry’s pace car — ~$100B revenue per gigawatt funding SpaceX at ~$50B/GW and NVIDIA at ~$30B — where a demand wobble means a pile-up.
NVIDIA’s $500B financing effort with Goldman, BlackRock, Blackstone, KKR and Apollo turns GPU compute into an asset class — Gavin calls NVIDIA “the central bank of AI.” Residual-value guarantees plus revenue shares above the floor could make it “a very large cloud with a capital light business” (Morgan Stanley); CoreWeave renting 2020 Amperes profitably out to 2029 implies nine-year GPU lives. Sacks’ lone risk: dark GPUs à la dark fiber if buildouts assumed $30–50/watt spot — though anti-data-center politics ironically insure against a glut.
Grok 4.6 breaks Sacks’ “frontier duopoly” frame: Pareto-dominant on intelligence-per-cost, only 1.5T parameters, with Grok 4.7 weeks away. Gavin’s positioning tell: SpaceX investors hold nuanced views on Starlink and orbital compute, but “very few of these investors talk about Grok at all” — while Elon runs a call option on the frontier and a put option of selling compute at “spot minus 90.”
A rare genuine bestie fight closes the show: JCal calls Amazon’s DSP subcontracting capitalism gone “a little too clever” and predicts “Mamdani and New Jersey are gonna win these lawsuits”; Sacks defends freedom of contract at $5.20 per package and $664 per household per year. Plus breaking news: Silver Lake may be buying Workday (+17%) — Gavin guesses the returning private-equity bid reflects the fact that open source is “an absolute godsend for the American software industry.”

Deep dive

1. The $2T IPO number may be a banker leak, not the clearing price

The setup: FT reports investors expect an October IPO, Polymarket puts 80% odds on this year (and 67% on Anthropic holding the top model at year-end), with the year ending at a $100–120B annualized run rate — so $2T is 16–20x sales, a fraction of SpaceX’s or Palantir’s exit multiples. Also breaking: talks to buy Decart for $6B, whose software cuts training and inference costs by making chips more efficient.
Gavin’s mechanics, worth internalizing: everyone is jockeying to be lead left, and whoever loses “leaks to the press and says, ‘Hey, we’re in the room. The bar is a two trillion dollar IPO.’” A responsible pricing absorbs the lockup — “you do not want anyone distracted at this moment” — so maybe it prices at two and trades to three; his prior 3–4T call is where it might trade. “If it comes out at two trillion, it probably means the testing of the waters or the roadshow just went incredibly well.”
Disclosure up front: Gavin isn’t a shareholder — “I don’t generally invest in Elon’s competitors” — though the SpaceX–Anthropic détente (Colossus compute rented to Dario) came once Elon got comfortable with Dario’s morals: “They don’t set off my evil detector.”

2. Losing share while 10X-ing: it’s the pie, not the slice

Gavin’s paradox: on the margin Anthropic is probably losing share to OpenAI, open source and Grok, “and they’re still growing so fast” — AI broadly accelerated in July even as public equities wobbled. He resurfaces Eric Vishria’s two-year-old thought experiment: OpenAI and Anthropic keep enormous value even if they lose the model layer, because “the product, the harness, the user familiarity” all compound.
Sacks’ math: three consecutive years of 10X growth implies a trillion dollars of ARR by end of next year if it held — the TAM supports it, “but you start to get into physical constraints” on compute and energy.
The coding origin story, per JCal and Sacks: Anthropic watched Cursor’s utilization on its platform and moved vertically to take the business. OpenAI has since “basically pivoted to coding” with GPT-5.6, going from 3–4X annual growth to reportedly over 20% month-over-month for two months — extrapolated, a 10X pace.

3. Dario’s hubris and a $25–65T knowledge-work TAM

Gavin, from multiple trusted sources: Dario has said Anthropic “might be the only private company in the world at some point” — Anthropic and governments, that’s it. Gavin takes the under and flags the tell: “I might interpret that as a negative signal because it’s so hubristic… This is getting into SBF land a little bit. Pride cometh before the fall.”
Internal confidence is real, though: at roughly $60B ARR they believed they could reach $600B in one year; now over $80B and not slowing, with “more advanced checkpoints than Fable up their sleeve.” Even “merely” $400–500B makes them the biggest software company; $200B is an amazing outcome.
Demand side: $25–65T of knowledge work depending on the count — and the head of the AI institute at one of the three largest investment banks told Gavin $25T was “way low.” So far it looks like accelerating growth, not labor substitution: more job openings for software coders than a year ago, “in the epicenter of the blast zone,” though he concedes some evidence of harm to fresh graduates. If it’s growth, “we will start to see that in the national accounts over the next 18 months.”

4. Supply is Landman country: turbines, jet engines, Texas audits

Gavin’s frame for the buildout: “it’s atoms, not bits” — go watch Landman, because you need Billy Bob Thorntons orchestrating thousands of people in 110-degree heat in remote locations, “actually worried about narcos.”
The turbine bottleneck is being brute-forced: the blades come from roughly two facilities with 40-ton presses now running 24-hour shifts; a news report said Elon bought a turbine company (JCal: “that speaks to me that he just wanted to go fast”); and the wildest detail — jet engines are being pulled off old planes and repurposed as data-center turbines, which is why private-jet residuals are “extremely strong these days.” Caterpillar, Cummins, GE Vernova and Siemens Energy are all expanding fast.
Politics and mix: Abbott’s Texas energy audit isn’t a moratorium — BYOE and you’re fine. Gavin, even as Elon builds a $10B solar plant in Texas: “the world will eventually run on sunlight, but we’re gonna run on natural gas for a while… it’s the greenest carbon fuel.” Sacks agrees green can’t scale in a year or two.

5. The data-center panic is wrong on every count

Sacks’ verdict: “a testament to the power of the media to create hoaxes and hysteria over literally nothing.” Water use is less than a golf course and recirculates; data centers lower electricity costs when they stand up their own generation and sell excess back; blue-state price rises come from decarbonization regulations, not compute.
Gavin: the water claim traces to a book error off “by a factor of 100,000”; the WSJ’s Ellendale story shows a dying town revitalized with tax revenue 10x’d, “the best thing that’s ever happened” — but nobody tells those stories, while the CCP runs “a very active anti-AI, anti-data center campaign.” His refrain: “The truth shall set you free… Can’t set you free if nobody tells the truth.”
JCal keeps the honest caveats: noise pollution is legit, siting matters, and at scale you can raise a valley’s temperature a couple of degrees — all manageable. “If you really wanna get obsessed about water, please, for the love of God, stop eating almonds.”

6. JCal’s contrarian call: open source slows Anthropic considerably

JCal’s position: large companies are embracing GLM 5.2, while the cited pricing comparison is GLM 2 versus Claude Opus, with GLM 2 about 90% cheaper. He and his founders use it, and corporate America will re-run the Linux playbook — “We don’t want to get rug pulled… open source is the safer bet.” So “there’s no way they get to a billion in revenue next year”; he says maybe a triple to $300–400 million. One hyperscaler told him it did a nine-figure reversal, abandoning a $100B frontier-model infrastructure plan for open source.
Sacks’ Apple/Android rejoinder: a meaningful subset will always pay a huge premium for true frontier intelligence — if the top 20% pays 10X, they’re golden — but the frontier labs are “on that hamster wheel”: stay six months ahead or the premium dies.
The Zuckerberg threat, per JCal: “Go ask Myspace, go ask Snapchat what it’s like to have Zuckerberg in the rear view mirror… He’s gonna have the best open source model in the next year, I guarantee it.” Gavin: “I might take the under” — his candidate for the American open-source frontier is Jensen and NVIDIA’s Nemotron stack.

7. Zuck’s 6,500-word manifesto: decentralize or else

“The Future Is for Everyone: The Path to a Positive AI Future” — open models, a free agent for everyone, a tutor for every person, superintelligence as “invention, not automation,” and safety grounded in balance of power. JCal’s framing: Zuck picked up “the abundance crown, which Sam Altman dropped when he went for the private company.”
Sacks found himself nodding, especially at the shot at the doomers — why rush to build a dystopia you don’t believe in? — and the load-bearing quote: “The notion that AI is so dangerous that the only safe path is an extreme concentration of power seems inherently problematic. Historically, hoping that an absolute power will benevolently provide for humanity if sufficiently enlightened has not led to safer positive outcomes.” Straight out of Sowell’s The Vision of the Anointed.
Sacks’ Washington account makes it concrete: former Biden officials moved into Anthropic government affairs, and the think-tank clique wanted AI limited to the US, cartelized into two or three companies under an Atomic Energy Commission-style body merged with state power. He heard versions of the “don’t bother competing” message Andreessen reported.
Gavin’s distillation — the episode’s sharpest line: Anthropic and the EAs believe this technology is “too dangerous to distribute”; Zuckerberg, Elon and Jensen believe it’s “too dangerous to centralize. And history has spoken.” He credits Dworkash’s point that Anthropic’s constitution wires its AI to do what Anthropic thinks is best for humanity — genuinely good people, but “there are always unintended consequences.”

8. Guns, not nukes — and the regulation that would have killed the IPO

Gavin reaches for the Second Amendment: the right to bear arms is “the last line of defense against tyranny,” and “I want the right to my own AI” whose values align with him — “I do not want Dario Amodei or Anthropic deciding what’s in my best interests.” His patriot version: ban open source and “we will lose AI, we will lose geopolitics, we will lose everything to China. It’s just that simple.” And: “If we’re going to have a Terminator thing, I want my own Terminators who like me.”
Sacks explains why the EAs pushed the nuclear-bomb analogy — you have no right to a nuke — and demolishes it via Jensen: “no consumer needs a nuclear weapon. Every consumer, every business needs AI.” It’s a consumer technology before a military one. JCal, blunter: they’ve “deluded themselves while taking peyote or LSD at Burning Man… You could unplug it at any time. Delusions of grandeur, period, full stop.”
Sacks’ commercial irony: had Dario’s “FAA for AI” happened, the six-month lead behind Anthropic’s entire pricing power would evaporate under a regulator that takes five-plus years to certify anything — “Your six months will be gone like that. Your model will be commoditized.” His victory lap: “Are they gonna thank me for preventing Dario from shooting himself in the foot?”

9. Distillation hypocrisy, and the case for going direct

JCal on the manifesto’s buried clause — “you can learn from anything you can observe”: a preemptive license to scan the world’s data from a company that “has sued every startup… that has ever tried to index or use the social graph at Instagram or Facebook.” Hundreds of lawsuits against exactly the distillation Zuck now blesses.
Credit where due, from JCal: Zuck’s job-displacement framing — “there is no rule that AI must increase automation faster than it increases individuals’ capabilities” — is the essay-length version of Jensen’s you-won’t-lose-your-job-to-AI-but-to-someone-with-AI line.
Gavin’s rule for every CEO, prompted by Zuck, Nikesh from Palo Alto Networks and Jensen all activating X accounts: “If you do not have your own voice and your own brand on X, you are at risk… it’s protection, it’s power, it’s armor, and I think it’s a sigil.” Corporate PR won’t save you when Democrats’ talked-about “accountability agenda” calls CEOs before Congress.

10. Open source makes frontier tokens more valuable — and the 2027 over/unders

Gavin’s Oppenheimer analogy: a 250-IQ frontier intellect orchestrating 150-IQ open-source intellects becomes more valuable, the way the Manhattan Project needed Oppenheimer and 10, perhaps 30, effectively Nobel-level physicists plus 2,000 merely top-10,000 ones. His plausible endstate: frontier tokens capture “65 to 85% of the economic value,” open source ~80% of the volume, “and that’s good for everyone, and everyone wins.”
Pinned down on 2027 exit ARR off a $100B 2026 exit: Gavin says “400 to 500 billion is within the zone of realistic and achievable” and would take a 25–30% over on Wall Street consensus (absent a successful regulation campaign). Sacks lands identically — $400–500B exit ARR “seems very comfortable,” limited by physics not demand — “and I think they think they’ll do the over.” JCal’s zoom-out: they were discussing the company at $10B 10 months earlier, making $400B a 40X jump.
Two demand backstops: Gavin’s computational-efficiency-of-humans point — approximating one brain takes a data center drawing a million American homes’ power for six-to-nine months, so there will always be a market for “the most computationally efficient intelligences on planet Earth” — and JCal’s salary math: 150–160M US workers, $10–12T of pay, 5–10% spent on tokens ≈ $1T of US AI spend, “not even talking about the rest of the planet.”

11. NVIDIA becomes the central bank of AI

The mechanics as Gavin lays them out: Goldman, BlackRock, Blackstone, KKR and Apollo raise third-party capital; NVIDIA matchmakes and offers residual-value guarantees — a floor rental rate after three-to-four years — in exchange for a revenue share above that floor. Citing a Morgan Stanley note: those effective royalties could make NVIDIA “a very large cloud with a capital light business.” His frame: “NVIDIA is kind of becoming the central bank of AI, the Federal Reserve of AI” — monetary policy working through private banks.
Why the asset finances: CoreWeave is renting 2020-vintage Amperes at economically profitable rates for 2029 — a nine-year life — and the divergent architectures of the big Chinese open-source models (Qwen, Kimi, DeepSeek, GLM) favor exactly this flexible GPU compute.
Sacks on the brilliance: the TAM was becoming constrained by financing capacity — Elon wants 6–8 gigawatts next year, $300–400B of capex against $100B just raised — and Jensen is unblocking it. Not circular à la Gurley’s worry: it’s asset-backed like airplane financing, standardized via reference designs and securitized “like they did with mortgages.” His one risk: dark GPUs like post-dot-com dark fiber, especially if buildouts underwrote $30–50/watt spot — but “in a weird way, all the political headwinds insure against that outcome.”

12. The pace car: Anthropic’s S-1 will break brains

JCal’s dead-man-switch argument: the two biggest compute buyers could self-correct — OpenAI has already trimmed its buildout ambitions from roughly $1.4T to ~$600B — so hands could come off the buy button before a mortgage-backed-securities-style blowup.
Sacks maps the food chain: Anthropic makes ~$100B per gigawatt of compute, which is why it can pay SpaceX ~$50B/GW spot, which is why SpaceX pays ~$30B to NVIDIA, on down to TSMC, Micron, SK hynix. Post-IPO, Anthropic’s quarterly GAAP numbers become “probably the most important signal the entire industry has” — “If the pace car slows down, the whole race is gonna feel it.”
Gavin’s crucial nuance: a pile-up happens only if the brake is demand. If Dario is merely being passed by SpaceX, OpenAI or an American open-source champion, no crash. And the S-1 “is gonna break a lot of people’s brains”: macro and value investors confidently assume tokens are subsidized, but “the overwhelming majority of tokens are profitable for everyone in the chain” — Anthropic generating cash, open-source tokens profitable, and OpenAI, if it is not already generating cash, doing so imminently. Their bearishness is like saying “I’m very bearish on the world economy because oil’s at $500 a barrel… It’s just not. Like, they’re just wrong.”

13. The Amazon DSP fight: a real bestie disagreement

The setup: New Jersey’s AG sued over Amazon’s delivery-service-partner structure, and Mamdani wants municipal law forcing Amazon to employ drivers directly — the DSP layer shields Amazon from crash liability, unionization and labor obligations.
Sacks defends freedom of contract: DSPs enable holiday-surge scaling and flexible work, employ thousands, and the change would cost “$5.20 per package… $664 more per household per year” while putting something like 5,000 jobs at risk. Voters fall for socialism, “but when they’re wearing their consumer hat, they just wanna pay less for packages.” Gavin’s mischief: bill it as a line item — “This is the Zohran Mamdani fee.”
JCal — a Prime subscriber since the beginning and a long-term Amazon shareholder who keeps increasing his position — breaks with both: the structure was “created explicitly to benefit Amazon” and to shift benefits, unemployment, food stamps and health insurance onto taxpayers. He calls Sacks “misinformed” on flexibility — these are 10–12-hour shifts, and DSPs are a union-busting tool — and claims employment would cost about 25 cents per delivery, invoking Schultz’s Starbucks living-wage move. His prediction: “Mamdani and New Jersey are gonna win these lawsuits.”
Sacks presses how Amazon would decide which drivers remain DSPs once surge flexibility is retained; JCal says Andy Jassy and Amazon should make that decision proactively, without specifying a selection rule.

14. Grok 4.6 breaks the duopoly; private equity returns for software

Sacks openly revisits his frame — as of last week he’d called a frontier-lab duopoly of Anthropic and OpenAI. Gavin’s evidence: Grok 4.6 sits outside the Pareto frontier on intelligence-per-cost, on CursorBench (caveat: xAI — almost certainly SpaceX — has the right to acquire Cursor, “maybe this is grading your own homework”) and on the Databricks evaluation — Databricks having just raised at $190B — ahead of Fable 5, “the gold standard,” at slightly lower price. DHH posted positively; it’s only a 1.5-trillion-parameter model, with a significantly more capable Grok 4.7 “in a few short weeks.” GrokBot feels like “another OpenClaw moment,” but usable by non-coders unlike OpenClaw or Hermes.
JCal’s attribution: the Cursor team plus Elon reshuffling in “his killers from SpaceX,” all in six months — “He’s frequently late. He’s never wrong.” Sacks’ structure: the compute clusters are a call option on becoming a true frontier lab with a put option of selling compute to rivals at “spot minus 90” — both sides of the Anthropic deal hold 90-day cancellation, so Elon effectively charges near-spot. Gavin’s kicker for SpaceX investors: they have nuanced views on Starlink direct-to-cell and orbital compute, but “very few of these investors talk about Grok at all, and it is Pareto dominant on a lot of measures.”
Breaking mid-show: Silver Lake may buy Workday, stock +17% after riding from ~300 to ~100. Gavin’s read: software’s 12–18-month death spiral troughed three-four months ago, and he guesses the returning PE bid may reflect the fact that “the rise of open source… is an absolute godsend for the American software industry” — a world of one-to-three dominant frontier models is a hard world for software; open-source diversity is the reprieve. JCal appends the Bending Spoons playbook: cut 80% of staff, run the remainder AI-first, and print.

Swulinski: the e-com paid-ads playbook is SaaS's $100M growth engine Episode 193 - August 14, 2026