Anthropic S-1 Shocker, Greatest Business Ever? Google's $500B Wipeout, Midterm Predictions

The hosts open by joking that this is “the number one podcast in the world” and “this week in superintelligence,” responding to a reported claim that they had stopped using the term “superintelligence” and reverted to “AI.” David Sacks says the story was “fake news,” pointing out that a photograph used in the coverage was from an episode he was not even on. He says he will now keep the others “in line” on the terminology. The group also jokes about their recent card game: Jason Calacanis says David Friedberg talked continuously for roughly seven or eight hours, while Friedberg says he had accumulated a lot of material because he had not played in a long time. Ben Affleck reportedly texted that he never wanted to play with Calacanis again. Friedberg briefly discusses Ohalo’s work in Kenya, where farmers are testing true potato seed. Chamath Palihapitiya says his company’s fourth quarter is off to a good start, though he does not want to “count the chickens.” The others tease him for forcing his sales team to pursue only giant deals in the previous quarter. Chamath concedes that he “butchered” Q2 and says the lesson is to pursue singles and doubles—smaller customers who pay on time and complete projects—because a large business can be built from dependable smaller deals.

The main discussion begins with Anthropic’s IPO filing. The hosts say Goldman Sachs and Morgan Stanley are expected to lead the offering, reportedly targeted for November 16, with the ticker ANT and a potential raise of $100 billion. They discuss an estimated valuation between $1.5 trillion and $2 trillion, with roughly $1.75 trillion to $2 trillion described as the current range. Chamath highlights projected revenue of $16.3 billion in the first half of 2026, described as 14 times the first half of 2025, and a projected $11.6 billion second quarter. Anthropic also describes a token and services market with a claimed $40 trillion total addressable market and potentially $100 billion in annual recurring revenue or revenue run-rate by year-end. Sacks calls it potentially “the greatest business in the history of capitalism” and perhaps the fastest value creation ever achieved by a venture-backed company. He says the company is only about six years old and could produce a venture outcome comparable to SpaceX’s despite SpaceX having existed for more than two decades. He also emphasizes the possible employee wealth creation: hundreds of employees could receive hundreds of millions of dollars collectively, while thousands could receive more than $10 million each. Seven founders reportedly retain 50.1% of voting power through a Class F share, while about 80% of those founder shares are expected to flow into Anthropic’s nonprofit-related structure or benefit corporation.

Sacks argues that the widely reported $80 billion Anthropic loss is misleading rather than a cash loss. He says the number came from convertible notes issued to infrastructure partners such as Amazon being marked as losses as Anthropic’s valuation rose. If the company had issued equity instead, he says, that accounting loss would not have appeared in the same way. The filing instead suggests that Anthropic has raised approximately $110 billion, spent only about $9 billion to $10 billion, and still holds roughly $100 billion or more in cash and cash equivalents. Sacks says gross margins are now around 85% and that the company is becoming more efficient rather than being commoditized away by open models. He explains Anthropic’s framework of raw compute plus “compute multipliers.” Raw compute comes from more and better chips; multipliers come from algorithmic improvements, better data curation, more efficient architectures, and mixture-of-experts systems. For a given task, Anthropic may deliver roughly ten times more capability or efficiency year over year, allowing it to cut prices by more than half while retaining or even improving margins. Friedberg adds that the filing shows a sharp increase in customers spending at least $100,000: about 200 at the end of 2024, 1,500 at the end of 2025, and 11,900 by the end of September. He estimates that roughly one million businesses globally could spend $100,000 while keeping that below 1% of total spending, implying that Anthropic may have penetrated only about 1% of its potential customer base.

Friedberg argues that the entire AI value chain is already generating substantial economic value. He describes a chain in which Nvidia turns roughly $1 of costs into $4 of chip revenue, infrastructure operators such as SpaceX earn high margins over the useful life of their data centers, and Anthropic earns approximately 85% incremental gross margin on customer revenue. He estimates—explicitly labeling it an assumption—that customers may be receiving two to ten times the value of what they spend on AI. Sacks agrees that his own organization gets more done with Claude, but says the customer-return estimate is difficult to measure. He believes Anthropic has enough margin to lower prices, segment the market, and charge more for premium intelligence while making simpler tasks cheaper. Chamath says Anthropic’s risk factors identify compute as its biggest concern, especially data centers, power, electricity, hardware, and financing commitments. Regulation is the second major concern, followed by privacy, intellectual property, AI safety, monetization, competition, and governance. He says Anthropic is not primarily worried about customer traction or employee departures. The discussion then turns political: Chamath argues that the AI complex is now a major driver of U.S. economic growth and that Democrats could slow the Trump economy by restricting power and data-center construction, while Republicans need to solve permitting, electricity, and infrastructure. Friedberg offers a broader theory: governments may eventually regulate compute allocation itself because AI companies are beginning to compete with governments for capital, influence, and organizational power.

Chamath revises his earlier estimate that Anthropic needed to raise $200 billion to $250 billion. Because the company already had roughly $112 billion or more in cash and has burned only about $9 billion to $10 billion, he now estimates that it may need to raise approximately $80 billion to $120 billion to reach a quarter-trillion-dollar capital base. He places a rough IPO valuation around $1.5 trillion and says the company may soon have more than $300 billion in combined cash, debt facilities, and other capital available to support its compute commitments. Sacks says this undermines the argument that Anthropic cannot finance the infrastructure orders it has made. The hosts then identify a larger risk: major AI companies are spending so heavily that they may compete with U.S. Treasuries for capital. Chamath says the AI complex could need to raise an amount equal to roughly 30% of Treasury issuance through the end of 2027. If AI companies must offer higher interest rates, financing costs could spread through the “neo-cloud” ecosystem and increase defaults. If Treasuries must offer higher rates instead, the borrowing cost of the entire U.S. economy rises. Sacks counters that the underlying businesses may be so productive that investors will still want both American government debt and American AI equity. The group also debates Anthropic’s pursuit of recursive self-improvement. Sacks says Anthropic used its own models in 26% of certain model-development tasks as early as August, creating a possible conflict with the “doomer” community that worries about runaway AI progress. He calls a future Anthropic–doomer split plausible.

The conversation shifts to Alphabet after Google’s stock reportedly fell 10% to 15%, wiping out approximately $500 billion in market capitalization. The hosts say overall revenue still grew about 25% year over year, and Google Cloud revenue roughly doubled to $30.7 billion, but search revenue growth slowed from about 17% to 12%. Search revenue nevertheless increased in absolute terms from approximately $56 billion to $63 billion. The concern is that AI systems such as ChatGPT, Claude, and Grok increasingly answer questions directly or complete tasks through agents, weakening the cost-per-click advertising model that has historically financed Google. Friedberg calls the situation complicated. He says Google Cloud has improved dramatically, moving from a business that was not making meaningful money to one generating more than $11 billion in operating income for the quarter, compared with about $3.5 billion a year earlier. The unresolved question is whether Google can turn its infrastructure and models into a consumer AI business that replaces or supplements search advertising. He also suggests that Google’s internal scientists may face tension over whether scarce compute should be used to develop models or rented to customers. Chamath discusses claims by a heavy Google advertiser, Max Anderson, that Google changed its ad-auction mechanics and may be over-earning on clicks. He says Google should publicly refute or explain those claims before the market concludes that the company is pulling revenue from the future to smooth current earnings. Sacks is more constructive: search revenue is still growing, while Google Cloud, YouTube, subscriptions, Android, Waymo, and Alphabet’s stakes in companies such as SpaceX and Anthropic diversify the business. He says Alphabet looks oversold and discloses that he bought more shares during the recording.

In the final major segment, the hosts debate the upcoming U.S. midterm elections. Chamath says he is confused by the polling and expects what Sacks calls a “spritz,” though Chamath characterizes it as more likely a Democratic spritz. He predicts that Republicans may retain the Senate while Democrats take a narrow House majority. In his view, a Democratic House would create uncertainty by issuing subpoenas, pursuing investigations, and releasing documents, while also potentially using power permitting and data-center policy to slow the AI-led economy. He says governors’ races and state-level control may matter as much as the federal result. Sacks warns that polls have repeatedly been biased, fraudulent, or badly calibrated, citing allegedly fake polls and a history of pollsters oversampling progressives. He argues that the economy is not a disaster, pointing to estimated growth of 2.5% in the first half of the year and 3.7% in the latest quarter, median household income near $90,000, and poverty near an all-time low. He says Republicans should vote rather than become demoralized by media narratives. Calacanis instead trusts Polymarket’s probabilities, gives Democrats roughly a 68% chance of winning the Senate and a 92% chance of winning the House, and predicts impeachment efforts, legislative conflict, and a chaotic two years. He believes the Republican Party may eventually move beyond Donald Trump, JD Vance, and Marco Rubio in search of a different populist champion. The episode ends with promotion of an All-In midterm livestream featuring guests including Tim Dillon, Rahm Emanuel, Taylor Lorenz, and Basel Lorenz, followed by the usual joking, interruptions, and invitations to drink together.