What Anthropic's S-1 Filing Really Reveals

Buried inside Anthropic's initial public offering filing is a number that looks like proof of disaster: an $80 billion loss. Read the filing closely, and the loss turns out to be an illusion. It came from convertible notes Anthropic issued to infrastructure partners like Amazon. As the company's valuation rose, those notes had to be marked up on the books, which flowed through as a paper loss on the income statement, even though no cash actually left the building (07:07). Strip that away, and the business underneath looks almost absurdly healthy: Anthropic has raised roughly $110 billion and spent only about $10 billion of it (07:30), leaving more than $100 billion in cash sitting on the balance sheet (05:22).

That filing, the S-1 a company submits before going public, is what the hosts spend the first half of this episode taking apart. The numbers are startling on their own. Anthropic is targeting a $100 billion raise at a valuation between $1.5 and $2 trillion (03:56), with revenue of $16.3 billion in just the first half of 2026, fourteen times what it made in the same period a year earlier (04:55). Co-host David Sacks calls it a historic outcome:

So this is probably the fastest value creation of any venture-backed company in history. [06:16] — David Sacks

The margin story is just as striking. Anthropic is booking an 85% incremental gross margin, meaning for every new dollar of revenue, 85 cents drops to gross profit (07:30). Co-host David Friedberg traced the money further back, mapping how a dollar NVIDIA spends making a chip turns into roughly $18 by the time Anthropic sells the resulting intelligence to a customer, once you follow it through NVIDIA's own margin, then a cloud provider's margin, then Anthropic's (13:50). His read is that this is the first time a business has plausibly approached what former Google CEO Eric Schmidt once called Google's real total addressable market: global commerce itself.

Google's total addressable market is global commerce. And I think for the first time ever, we're seeing a business actually potentially starting to realize that theory. [14:47] — David Friedberg

The Microsoft comparison the hosts pulled up makes the same point from a different angle. Microsoft's revenue grew from $198 billion to $332 billion over four years, while headcount grew by only 2,000 people (16:58). That is not a story about job losses, the hosts argue. It is a story about the same number of humans producing far more output, which is the quieter and more defensible version of the AI productivity claim.

None of this means the business is risk-free, and co-host Chamath Palihapitiya spent a chunk of the episode reading Anthropic's risk-factor section the way an analyst would: by counting which words show up most. The single largest cluster of language is about compute, specifically data centers, power, chips, and the electrical grid (18:46). Regulation is a distant second. Customer churn and employee attrition barely register. In other words, Anthropic itself believes its biggest constraint is not demand, competition, or trust. It is whether America can build data centers and generate power fast enough.

A New Rival for the Treasury

The more unusual argument in the episode is about what happens when a handful of AI companies need more money than almost anyone else on earth. Chamath laid out a chart showing that the capital AI firms collectively need to raise through the end of 2027 equals roughly 30% of all projected US Treasury issuance over that same period (34:11). That means Anthropic, OpenAI, Meta, and a few others are not just competing with each other. They are competing with the United States government for lenders.

The entire United States economy is levered to the AI complex. So very narrowly, we're talking about seven or eight companies that sets the tone for the entire country. [33:00] — Chamath Palihapitiya

Chamath's worry is not that the AI companies are bad businesses. It is that when one narrow slice of the economy becomes this large relative to total credit markets, either side of that competition for capital can push borrowing costs up for everyone else. If investors flee to the safety of government bonds, AI firms pay higher interest. If investors chase the AI boom instead, the Treasury has to offer higher yields to attract buyers, and that higher rate ripples through mortgages, corporate debt, and everything else priced off government bonds (48:13). He compares the scale of this dependency to the run-up before the 2008 financial crisis, not because he thinks a crash is coming, but because so much of the economy now rests on so few companies.

The engine making Anthropic's margins keep improving, and the thing making that capital bet look reasonable, is something called recursive self-improvement, or RSI: using an AI model to help build the next, better version of itself. Anthropic disclosed that as of August 2026, 26% of the tasks involved in creating its next-generation models were already being done through RSI (44:25). Sacks argues this is both Anthropic's biggest technical advantage and a strange kind of tension, since Anthropic has also publicly urged the industry to slow down.

It's almost like creating an alibi in advance for themselves if this RSI thing goes really bad. [45:14] — David Sacks

Sacks's point is that Anthropic gets to say it tried to get the industry to agree to a slower pace, while continuing to push RSI as hard as anyone, which could matter a great deal if something eventually goes wrong with a model that improved itself.

Google's Wobble

If Anthropic's filing is the episode's bull case for AI, Google's third-quarter earnings are its cautionary tale. Alphabet's stock dropped enough after its Q3 report to wipe out roughly $500 billion in market capitalization (62:04), even though overall revenue grew 25% year over year and Google Cloud's revenue roughly doubled to $30.7 billion. The problem was narrower than the headline number suggests: search revenue grew 12%, down from 17% the previous year, and for the first time, search fell below 50% of Alphabet's total revenue (71:50).

Friedberg, a Google alum, argued the drop in search's share is mostly a sign of Google Cloud finally working, not proof that search is dying; Google Cloud's operating income jumped from $3.5 billion to over $11 billion in a single year (63:58). But a separate and more pointed accusation also surfaced during the episode: a viral post from an advertiser claiming Google had quietly abandoned its long-standing Vickrey auction, the pricing method where the winning bidder pays only what the next-highest bidder offered, in favor of a system that extracts more money from advertisers (68:20). Chamath, who helped build Facebook's original ad auction using the same Vickrey model, called the accusation a serious problem if it sticks:

If you are Google IR, please read this and refute it... the market will react badly to this idea that this franchise is essentially robbing Peter to pay Paul. [69:31] — Chamath Palihapitiya

Put the two companies side by side, and a theme emerges: the easiest way to misread either business right now is to look at one headline number. Anthropic's $80 billion loss said nothing about its actual cash position. Google's slipping search share said nothing about its cloud business. The episode's real argument is that the AI boom has made company balance sheets, auction mechanics, and even Treasury yields into a single connected system, and reading any one piece in isolation will mislead you.