· AI
Anthropic's New S-1 Has an 80 Billion Dollar Loss and an 85 Percent Margin. Only One of Those Numbers Tells You What Is Actually Happening.
A loss that big and a margin that good cannot both be the headline, and figuring out which one actually describes the business is the whole exercise.
Anthropic filed its public S-1 with the SEC on October 26, and the two numbers getting quoted most do not point the same direction. Reuters reported, via a draft of the filing, a net loss of $80.7 billion for the first half of 2026, against $13 billion in the same period a year earlier. At the same time, an independent analysis of the S-1's own financial notes puts gross margin, revenue minus cost of revenue, at roughly 85 percent in the first half of 2026, up from an implied 40 percent for full-year 2025. A company does not both collapse and get radically more efficient at serving its customers in the same stretch. It can do both at once if the two numbers are measuring different things, and here they are.
Start with the loss, because the headline writes itself and the headline is misleading if you stop there. A detailed breakdown of the filing and Reuters's own reporting agree that about $72 billion of that $80.7 billion is a non-cash accounting charge tied to convertible notes Anthropic issued to raise money. When a private company's valuation jumps the way Anthropic's has, instruments that convert into its stock become more valuable to the people holding them, and GAAP accounting requires the company to book that increase as an expense against itself. Strip that charge out and the loss left over, from running the actual business, is closer to $8 billion for the half. That is still a real loss and a large one. It is not $80.7 billion, and a number that size circulating without the breakdown tells readers almost nothing true about whether Anthropic's core operations are getting healthier or worse.
Here is the steelman for treating the margin number with real suspicion, because it deserves one. Anthropic does not report "gross margin" as a line item at all. The 85 percent figure is an analyst's calculation from a cost of revenue note buried in the filing, not a number the company chose to headline, and the comparison period matters: measured against full-year 2025 the improvement is 40 to 85 percent, but measured against the first half of 2025 specifically, one outlet reported the prior margin closer to 12 percent, which makes the jump look even more dramatic depending on which "before" you pick. A skeptic is right to note that a company two weeks from pricing a $1.5 to $2 trillion IPO has every incentive to let the flattering comparison travel further than the complicated one, and that cherry-picking the base period is a classic way to make a trend look cleaner than it is.
I think the skeptic is right about the salesmanship and still wrong about the substance. However you slice the base period, Anthropic's cost of running its current customer base, as a share of what those customers pay, is falling hard, from compute spending of $2.3 billion in the first half of 2025 to about $9 billion a year later while revenue grew more than fourteen-fold in the same stretch. That ratio does not move that far by accounting choice. It moves because token costs per unit of useful work keep dropping across the industry, Anthropic's own price cuts on commodity-tier models included, and a lab that did not pass efficiency gains through to its margin would be the one losing share to Gemini, GPT, and the open-weight field this blog covers every week. Falling unit costs showing up as better margins instead of evaporating into waste is exactly what competition is supposed to produce.
The number that should worry you more than either headline is the one nobody is fighting over: infrastructure commitments north of $600 billion, spread across Google Cloud, AWS, Microsoft, Broadcom, and chip-rental deals with Nscale and Lambda, most of it payable whether or not Anthropic ends up needing the capacity. An 85 percent margin on today's revenue means little if a chunk of that committed capacity sits idle because growth slows or because a cheaper competitor takes the workload instead. That is a real risk, and it is mine to flag as a read rather than a fact: nobody, including Anthropic, knows yet whether the next decade of demand fills those commitments.
The practical lesson for a business signing or renewing an AI contract has nothing to do with Anthropic's stock and everything to do with reading a vendor's numbers the way you would read your own. A scary total and a flattering ratio can both be true at once, and neither one, by itself, tells you what to do next. Before you take either number from a vendor's pitch at face value, ask us to run the actual math on what you're being quoted.
Sources
References used in this article. Links also appear alongside the relevant claims.
Let's make it happen.
You don't need a polished brief. A couple of lines about where you want your company to go is plenty, and we'll come back with what we'd tackle first.
We scope the work around your goals and timeline before quoting anything, so you know exactly what you're getting.