All-In Podcast: Anthropic's S-1 Reveals 85% Gross Margins and a $100 Billion Cash Pile, but the Hosts See Treasury Crowd-Out as the Risk Nobody Can Price
Episode 295, October 30, 2026: the hosts dissect Anthropic's IPO filing, Google's post-earnings selloff and the midterms
The All-In hosts spent most of episode 295 on Anthropic's S-1, which they described as the most consequential IPO document of the cycle. As discussed on the show, Goldman Sachs and Morgan Stanley are leading the offering, targeting November 16 under the ticker ANTN. The company is reportedly seeking to raise about $100 billion at a valuation of $1.5 trillion to $2 trillion, which would be the largest IPO ever. Polymarket, a show partner, puts a 78% probability on a raise of at least $95 billion. Its market-cap brackets cluster just under $2 trillion, with 31% on the $1.75 trillion to $2 trillion range.
The "$80 Billion Loss" Was an Accounting Artifact, and Capital Efficiency Is the Real Story
The most important correction to the prevailing narrative, per David Sacks, concerns the roughly $80 billion loss that critics have cited as evidence of a bubble. Sacks said it is not a cash loss. Anthropic issued convertible notes to infrastructure partners such as Amazon. As the company's valuation rose, those notes were marked up and the change ran through the income statement. "If they had just issued equity, that loss wouldn't be there," he said. "It was not a cash loss. It's basically a fake number."
What the filing shows instead, according to Sacks, is a company that has raised roughly $110 billion and spent only $9 billion to $10 billion over its life. It holds about $100 billion in cash or equivalents. Chamath Palihapitiya said he had assumed Anthropic held $30 billion to $50 billion of cash and was wrong. He credited CFO Krishna Rao with "an unbelievable job." Sacks added that an estimated $80 billion debt facility would give the company more than $300 billion of dry powder once the IPO closes. In his view that backs up the multiyear compute commitments bears have called unfundable.
The reported revenue trajectory is steep. Jason Calacanis cited $16.3 billion in the first half of 2026, up 14x year over year, with Q2 alone at $11.6 billion. Anthropic is projecting $100 billion of annualized revenue by year-end, though the hosts noted the filing does not make clear whether that is recurring revenue or a run rate. The company also cites a $40 trillion addressable market, which Calacanis said he could not reverse-engineer. Seven founders reportedly hold 50.1% of voting power through a single class F share.
Margins Are Rising While Prices Fall
Sacks said gross margin is now about 85%. He said the filing shows a given task getting roughly 10x cheaper year over year. Anthropic passes on about half of that as lower prices and keeps the rest as margin. Anthropic calls the drivers "compute multipliers": algorithmic gains, better data curation and architecture improvements that sit alongside raw chip scaling. Sacks argued this undercuts the thesis that open models are putting Anthropic on a "hamster wheel" toward zero margins. He added that if recursive self-improvement is real, as the company claims, "their lead could actually grow."
Freeberg built a value-stack analysis, which he said was prepared with Claude, to argue that everyone in the chain is profitable. Nvidia turns $1 of cost into $4 of revenue at a 75% margin. Anthropic reportedly earns about 85% incremental gross margin on new revenue, which he calculated as roughly $18 of Anthropic revenue per Nvidia dollar. His assumption that customers earn two to 10 times what they spend was, by his own admission, made up. He also noted that customers spending $100,000 or more rose from 200 at the end of 2024 to 1,500 at the end of 2025 and 11,900 at the end of September. He estimates only about 1% of the roughly one million global businesses that could plausibly spend that much are customers.
Revenue Mix: Broad Experimentation, a Thin Layer of Large Deals
Palihapitiya, drawing on his experience selling enterprise software at 8090, offered the most skeptical read of customer quality. By his breakdown, about 40 million individuals and small businesses spend under $100,000. About 10,000 spend $100,000 to $1 million, about 1,650 spend $1 million to $10 million, and only about 250 spend more than $10 million. He attributed the strength below $1 million to decentralized budgets. "There are millions of humans running around with kind of million-dollar budgets," he said, which encourages experimentation. Multimillion-dollar deals require cross-functional sign-off and a demonstrable return on investment.
He estimated that about 1,900 companies have made that case. After stripping out venture-funded AI natives, he thinks "a couple of hundred" Global 2000 type buyers remain. "I think the ROI case is still a little speculative," he said. He views the quarterly migration of customers up the spend ladder as the first of two key risks for investors in the IPO, and the one that can be priced.
What Management Fears: Compute
Palihapitiya ran a word analysis of the risk factors. Compute dominates, particularly data centers, financing commitments and power. Regulation and export controls come second, followed by privacy, AI safety and competition. Customer traction and talent retention barely appear. In his view this shows a company with a clear binding constraint. He argued it also sets up the political fight into 2028, with Democrats gaining from slowing power and data center buildout and Republicans needing to get it right. Sacks went further and said the party's agenda amounts to either an outright ban or "a DMV for models." Calacanis disputed this, noting that Anthropic's founders are among the party's most important potential donors. Sacks and Calacanis also traded sharp words over the state of the economy, with Sacks citing 3.7% tracking GDP growth and Calacanis pointing to Trump's approval ratings.
The Risk That Cannot Be Priced: The AI Complex Versus Treasuries
Palihapitiya's central warning was structural. The AI complex, which he defined as seven to 10 companies, now owes roughly $340 billion to $400 billion. By his math, its funding needs through 2027 equal about 30% of upcoming Treasury issuance. Microsoft is the exception, having cut its debt. While discussing this, the hosts noted that Meta appeared to carry negative net cash, which Palihapitiya initially suspected was an error. It was ultimately reconciled as about $80 billion of cash against $82 billion of debt.
He laid out two scenarios. If hyperscalers must pay higher coupons, those rates cascade to neoclouds and raise default risk. If Treasuries must pay more to clear auctions, the cost of borrowing rises across the economy. "So many roads lead to downside risk more than upside risk," he said, adding that he has not seen a handful of companies compete with governments for capital in his career. Sacks took the opposing view, arguing the spending reflects returns the companies can see. Calacanis said utilization could rise 500x and the real risk is too little compute. Freeberg predicted governments will eventually seek to regulate compute allocation.
Valuation View and the Q3 Question
On pricing, Palihapitiya expects a valuation between $1.3 trillion and $1.7 trillion, with a midpoint around $1.5 trillion. He called that "fairly priced," meaning "no deal" but an entry point into the future. He said the smaller raise he now expects, roughly $80 billion to $120 billion, gives Anthropic more pricing power and easier float management. Calacanis hopes the IPO comes after November 11 so the company must include Q3 figures.
Whispers about Q3 are less flattering. Calacanis said revenue growth reportedly moderated to a range near 50%. Sacks relayed unverified chatter that the slowdown partly reflected Anthropic cutting off 99% of distillation, a voluntary hit. Demand then reaccelerated after recent token price cuts. The hosts stressed that this is scuttlebutt, not data.
RSI, Regulatory Capture and the Open-Model Threat
Sacks named open models as the biggest risk, if they saturate intelligence at low price points, and recursive self-improvement as a second-order one. Anthropic's efficiency gains increasingly come from using its own models to find algorithmic improvements. An August company blog showed 26% of tasks in developing next-generation models already involved RSI. "There's no single company in the world that is doing more to race towards full RSI than Anthropic," Sacks said. He said the company has to keep pushing activity it describes as risky to stay ahead. He sees a possible "divorce" between Anthropic and the doomers. He called Dario Amodei's pacing letter "virtue signaling" and "an alibi in advance." Sacks also predicted the IPO will clarify Anthropic's status as a big tech monopoly deserving suspicion, and criticized its government affairs hiring and political spending as regulatory capture.
Calacanis argued that Anthropic's first seven years of cumulative revenue compare with the best franchises ever built, including Amazon retail, Google search and the iPhone. He suggested the business could eventually exceed those three combined.
Google's $500 Billion Wipeout: Search, Auctions and the Talent Question
Alphabet fell 10% to 15% after Q3 earnings, erasing about $500 billion of market value. Calacanis said search revenue missed for a second straight quarter. Search grew 12%, down from 17%, rising from $56 billion to $63 billion by Sacks's figures. Cloud revenue doubled to $30.7 billion. Freeberg noted Cloud operating income rose to over $11 billion from $3.5 billion a year ago. He said the question is whether Google's consumer AI products can replace any erosion in core ad revenue. He cited internal tension over whether Google is a model company or a compute landlord, and the reported departure of Jeff Dean.
Palihapitiya flagged a more serious concern. A widely shared post by advertiser Max Anderson claimed Google changed its ad auction away from a second-price format and is "over-earning" per click. If true, advertisers will eventually cut spend. "The market will react badly to it," he said, urging Google investor relations to address it directly. He said the company's only other task is "How do you keep Demis at Google?" Sacks argued the selloff is overdone, since search is still growing in absolute terms and its share fell below 50% of revenue only because Cloud is growing faster. Calacanis said he bought shares during the show. Sacks noted Alphabet owns about 15% of Anthropic and Amazon also holds a large stake.
Midterms: Polls Distrusted, Polymarket Leans Democratic
With the election on Tuesday, Calacanis cited Polymarket odds of 68% for Democrats winning the Senate and 92% for the House. He expects impeachment efforts and sweeping subpoenas. Sacks argued polls are biased or fraudulent and that a normal out-party midterm gain is being framed as a rout. Palihapitiya expects Republicans to hold the Senate and Democrats to take a narrow House majority, and thinks the Democratic playbook will be to create uncertainty and slow power and data center permitting ahead of 2028.