Elon's Anthropic Deal, The Next AI Monopoly?, "FDA for AI" Panic, Trading the AI Boom

The episode opens with sound checks and teasing about appearance and tiredness. One voice complains that the audience is "brutal" about the bags under their eyes, and another jokes, "It's a good thing I'm rich."

The host introduces the episode as the All-In podcast. The panel is named as Chamath Palihapitiya, David Sacks and "our fifth bestie" Brad Gerstner (transcribed as "Gershuner"). David Freeberg is described as out sick and expected back the following week with two interviews.

The first topic is Spencer Pratt's campaign for mayor of Los Angeles. Participants praise his campaign ads; one calls them "one of the best political ads I've ever ever seen" and credits his social media and ad production teams. One speaker predicts Pratt will win and says the ads are the reason. Participants also praise his debating, calling him funny and "so chill."

The race as described pits Pratt against Mayor Karen Bass, characterized as "extremely left wing," and a city councilwoman described as further left still. In the account given, the councilwoman criticized the mayor over homelessness. Pratt countered that she already runs the city's homeless programs, and argued that the core problem is addiction and mental illness rather than a lack of housing. A participant says Skid Row is a place "you would not wanna walk through." A clip is played in which a candidate calls Pratt's description "a MAGA Republican's idea of what Los Angeles looks like."

A participant predicts that a California ballot initiative, the retirement protection and savings act, will pass "with big numbers." They describe it as a referendum that would effectively knock out the wealth tax by amending the state constitution to protect retirement savings and personal assets. They call this a "very nonconsensus view" they are increasingly optimistic about, and argue that passage would send a message to the rest of the country. Another participant is skeptical until that happens and says they will stay in Texas for now.

The discussion then turns to a figure transcribed as "Mondami," who made what is described as an attack video in front of Ken Griffin's house. A participant calls this dangerous given a recent ideologically motivated shooting of a CEO and the targeting of Sam Altman's house. As reported on the show, Griffin then said in an interview that he would put his efforts into Florida, said he felt offended and worried for his personal safety, and called the video out. The participant says the response was "mealy mouthed" and did not apologize. Another participant dismisses the matter: "New York is becoming a flyover city."

The main story is that Elon Musk has leased all of his Colossus 1 data center to Anthropic. The host credits a co-host, addressed as "Jamal" in the transcript, with saying on the previous episode that Elon and Dario should do a deal. The deal came five days later. The figures given: Anthropic added over 220,000 NVIDIA GPUs and over 300 megawatts of energy.

The panel says the deal is already showing up in the product. After Claude users ran into rate limits, Anthropic doubled the Claude Code rate limits, removed peak usage caps for paid users, and raised API volumes for Opus models. The host says xAI also trains its models at Colossus and still has more than enough compute. The host credits Elon's bet on building data centers quickly and asks whether this marks the emergence of "Elon Web Services," a hyperscaler competing with Google Cloud, AWS and Azure.

The first response argues that the frontier labs' revenue is limited by supply rather than demand.

Anthropic and OpenAI's revenue performance has nothing to do with demand. Zero.

The speaker says that with infinite power, Anthropic's and OpenAI's revenues would probably be "even more parabolic." In their view, coverage of whether the companies beat or miss forecasts means nothing, and the five-year outlook is strong. What the companies need is more compute and more power.

The same speaker raises a second problem: poor leadership at most AI firms. They say these leaders come across as "untrustworthy or too self interested," and that political and community reactions are turning negative. They cite about nine gigawatts of capacity due online this year, almost 50% of which is now being protested. They expect that, "if history holds," most of the protested capacity will be stopped, which would tighten supply further.

Later, another participant pushes back on the idea that the protests are spontaneous. They call them "highly organized activists" moving around the country, and compare them to the campaigns that stopped fission reactors from being built in America thirty years ago. They say America now builds no nuclear reactors while "China's got a 100 of them," and call for scrutiny of who funds the activists. They grant that some concerns are real but describe claims about water and electricity bills as misinformation. Their argument is that bills are rising in New York and California, which have not built supply on the grid, and falling in Texas, which is building the most data centers. These are presented as the speaker's claims and no data is given on the show.

The first speaker links the deal to the SpaceX IPO. When skeptics "red team" the SpaceX valuation, they argue, the biggest target is the speculative value of orbital data centers. Leasing out a large amount of capacity on the ground weakens that critique: even if orbital data centers are delayed by months or quarters, there is now "a structural core business" that can subsidize training Grok, a theme the speaker calls "underreported." In this account, Elon read the signs early, built scale and secured power before others, and now holds the critical asset, which makes him a kingmaker as SpaceX goes on its roadshow.

Another participant agrees, saying there is "nobody better on planet Earth than Elon at converting electrons to tokens." They cite a tweet from Sean Maguire describing SpaceX as a five-layer cake: launch, connectivity, compute (hyperscaler and space data centers), applications and models, and other bets. The open question on the roadshow, they say, had been that xAI is not on OpenAI's or Anthropic's revenue trajectory despite huge commitments. This deal, in their words, is the "ace card."

This speaker estimates the deal adds $4 to $5 billion of revenue this year, on top of analyst estimates they place "in the mid twenties." They describe three facilities, Colossus, MacroHard and MacroHarder, with 1.2 gigawatts and Blackwell chips at the newer sites. The facility given to Anthropic is described as "less connected" and its H100s as "great for inference." They say the deal removes pressure on xAI to produce revenue immediately, makes Elon an immediate hyperscale competitor, and is probably not the last such move.

the chaos that is American capitalism somehow finds a way.

The same participant welcomes what they call the détente and friendly exchange between Anthropic's team and Elon, arguing that all of this capacity is needed to keep American frontier models at the frontier.

A third view, which builds on points it credits to Chamath and Brad, calls the deal "highly complementary." SpaceX has a very profitable Starlink and launch business, while xAI has had "huge losses": frontier training clusters are expensive, and a lab earns no revenue until its model can compete at the frontier. The speaker adds that the revenue is currently in enterprise, meaning coding, where xAI has no product yet despite its Cursor deal. So it bears the costs without the revenue. Leasing capacity lets Elon keep a frontier model company without "massive unpaid for CapEx commitments," which fixes a balance-sheet problem. For Anthropic, the deal eases its compute constraint.

The same speaker then jokingly congratulates Dario on "winning the AI race." The host notes that this co-host has previously been critical of Anthropic and Dario on the show and on X, and asks for their view now.