Loeb bets the bulk of Third Point on AI — the week Gurley says Anthropic is midwifing a deity
Dan Loeb says oil and AI now trump every traditional macro signal and has the bulk of Third Point's capital in the AI complex — the same week the All-In panel argues over whether Anthropic believes it is building a god, whether Washington ends up banning open-weight models, and whether the AI jobs apocalypse was ever in the data. Underneath it all sits a live war, and Darren Farber's warning that adversaries can poison AI's foundation of fact.
May 25 – May 31, 20263 episodes2 shows113 claims on the record19 new predictions
01
Anthropic: most consequential company, or Dr. Frankenstein's lab?
Dan Loeb names NVIDIA, Anthropic, and Elon Musk's companies the three most consequential in the world and keeps the bulk of Third Point's capital in the AI complex, citing Anthropic's revenue growth as evidence enterprise adoption is barely scratching the surface. A day later, Bill Gurley reads the same company as a Dr. Frankenstein story: its founders, he argues, believe they are midwifing a deity, not writing software. David Sacks pushes a colder theory — Anthropic's safety branding is red capture, a Washington play that ends in an effort to ban open-source and open-weight models — while Chamath Palihapitiya warns that when the refs don't understand the game, the game gets run over. Same company, same week: the bulk of a $24B firm's capital, and the industry's most contested creation myth.
“I also think about what's going on in the three most consequential companies today, NVIDIA, Anthropic, and Elon World, all of his companies collectively.”
“in terms of the red capture agenda that you're seeing in Washington, I think where it's all leading to is an effort to ban open source models or open weight models”
The job-loss narrative flips — with one loud dissent
David Sacks claims vindication for his January call that AI would create job gains, not losses — nobody, he argues, can produce data for an apocalypse, and software-developer postings sit at a three-year high. Chamath Palihapitiya goes further: the layoffs are AI-washing of years of over-hiring, since nothing measurable from AI has shown up at end-consumption. Dan Loeb, a day earlier on Invest Like the Best, lands on the same side — net job creation, with humans permanently needed for high-touch work like private equity. Jason Calacanis is the holdout, taking CEOs at their word that they are cutting, and putting a ten-year clock on cab drivers, truckers, and warehouse sorters. The optimists argue from aggregates; Jason argues from the displaced, and demands humility on their behalf.
“my most contrarian take back in January on our prediction show is that AI would lead to job gains, not job loss”
The not-a-bubble bull case runs into model commoditization
Dan Loeb's case for the AI complex is that it is nothing like the dot-com era: hyperscalers fund capex out of enormous cash generation, and NVIDIA at 15 times 2027 earnings is a catch-up trade, not a mania. The same week, Chamath Palihapitiya reads Rogo's eval paper — the top frontier models separated by less than three-tenths of a percentage point — as the end of the single-best-model era, and expects training costs to collapse from $10 billion runs toward $10 million ones as stacks get rebuilt close to the bare metal. Both can be right, but they price different futures: Loeb is paying up for durable dominance; Chamath sees a complex whose core asset is commoditizing while enterprises torch token budgets for minimal results.
“You don't have the valuation bubble now on those companies that you had on those companies back in those days.”
Two reads on the same war: Farber's fragile dictators, Loeb's first question
Recorded mid-Iranian contingency, Darren Farber's master frame is that dictatorships are enormously strong and enormously weak at once — the IRGC controls half of Iran's economy while the regime caps ATM withdrawals at about $7 — and he predicts both China and Iran fall within his lifetime. Winning, he argues, is politically defined: an open Strait of Hormuz plus a durably degraded Iranian military, which he says will take an amphibious force ashore and real magazine depth. Dan Loeb, two days later, shows what that theater does to markets: oil and AI now trump every traditional macro indicator, and Third Point went into the war with more credit than equity exposure for the first time since 2009. The contingency Farber wargames is the first variable Loeb prices.
“Dictators are enormously strong and enormously weak at the same time. This is actually like a good frame for our entire conversation, which is to say, we have to be able to hold two competing ideas in the same vessel and both can be true.”
Darren Farber's defining Ukraine lesson: cheap commercial drones became a decisive article of war, iterating roughly 50 times in three years through garage-buildable commercial supply chains. ILTB →
Farber counts four or five continuing resolutions a year, each barring new program starts, and warns neo-prime defense startups will be lost to them unless Congress grants multi-year procurement authority. ILTB →
Third Point bought the resold Twitter buyout debt at 96-97 cents on the dollar for a roughly 12% yield when most credit investors were too scared to touch it — at the time the firm's largest credit position. ILTB →
Chamath Palihapitiya relays a report of a company that accidentally spent half a billion dollars on Claude usage in a single month — about $16.6 million a day — after failing to set employee limits. ALL-IN →
Pope Leo XIV's first encyclical — 235 pages on AI — warns technology takes on the characteristics of those who build, finance, and control it; Amazon, Google, and Meta lobbied the Vatican to soften it and failed. ALL-IN →
David Sacks's market-share math: a company growing 10x a year against a rival at 3x reaches about 90% share within two years — his ceiling case for Anthropic, absent a competitive response or compute constraints. ALL-IN →
The record
Every episode this week
The issue above is written from these structured records — nothing else. Each card is the full breakdown: claims, predictions, quotes, all timestamped.
AI competition will drive a productivity boom of cheaper goods and services rather than obscene profits, with gains whittled away by competitors; no one ends up with 70% operating margins.
We will have a Cambrian explosion in startups as laid-off, AI-enabled talent forms small high-output companies.
Jason Calacanis · ALL-IN · short to mid term · source 1:26:54
AI-driven job loss will increase in the short to mid term, after which displaced people will have to re-skill or leave the workforce, as in prior technological revolutions.
Jason Calacanis · ALL-IN · within 10 years · source 1:09:20
Self-driving and robotics will eliminate the jobs of cab drivers, truck drivers and warehouse package-sorters within the next decade.
David Sacks · ALL-IN · within 1 year · source 57:13
Token efficiency will be a big theme over the next year as enterprise token spend ramps faster than customers expected, driving a push for efficiency.
David Sacks · ALL-IN · within 2 years · source 55:11
If Anthropic keeps outgrowing OpenAI (10x vs 3x YoY), it will reach ~90% market share within two years, though a competitive response or compute constraints may prevent it.
First mentions
New to the graph this week
Subjects no indexed episode had discussed before this week.
AI washing×17open source AI×14Iran×10Department of War×10AI×8Third Point×7magazine depth×5neo-primes×5open weight models×5Hamas×4Meta×4technology×3corporate governance×3shareholder activism×3Japan×3Maxwell Taylor×2
How it's built
Written at ingest, never at query time
The week's episodes are diarized into structured records; one editorial pass writes this issue from those records when the week's ingest lands. A grounding gate greps every quote above against its episode record — a quote that isn't verbatim fails the build.
the data behind this issue
# this issue, as structured data — stories, citations, receipts
curl 'https://aipod-two.vercel.app/digest/2026-w22.json'# the records it was written from
curl 'https://aipod-two.vercel.app/api/episode/<id>'# any id in episode_ids