Does Anthropic Lose Money on Inference or Monthly Plans

The relevant quote: Today, that ... compute spend patterns. This is being shared as proof that Anthropic is haemorrhaging money on inference....

Three questions, three answers. Is Anthropic losing money on you specifically, if you're a whale who pins the 20x plan to its limits? At the margin, possibly yes, and you should send them a holiday card. Is it losing money on the average Max subscriber's inference?

Anthropic’s theory: inference costs will fall 5 to 10 times annually; users who become dependent on Claude Max today will still be subscribers when the marginal cost of serving them drops to $20 to $50 per month; and the switching costs built ...

But there’s a question most people are conveniently ignoring: will it ever make money? Instead of just telling you “Anthropic loses money,” I’m going to explain to you how their business works, how it’s way more complicated than one might think, and end with my two cents on whether they’ll actually ever be profitable.

The physics of inference is democratic — cost per query is mostly a function of model size and hardware, and both companies run comparable frontier models on similar chips. Gross margins confirm it: Anthropic's 40%, OpenAI's 33%, despite radically ...

The firm, as reported by Cryptopolitan, has defined this situation as an “inference paradox” in that falling token rates do not imply lower costs; this is due to the fact that agents will use significantly more tokens now. That is what makes the SemiAnalysis numbers important. If power users can pull five figures of API-equivalent usage from a $200 plan, OpenAI and Anthropic eventually face a choice between absorbing the subsidy and tightening access.

The question confronting every allocator evaluating this name is whether six structural fractures, each independently capable of derailing the thesis, are adequately compensated by a multiple that requires near-perfect execution through 2028 in a market where six well-capitalized competitors are converging on capability parity, inference costs are declining by an order of magnitude annually, and Anthropic’s own CEO has publicly acknowledged that the margin between transcendent success and insolvency is measured in quarters. Inside this analysis: the precise mechanism by which Anthropic’s u

The bear case on AI labs was an analogy before it was an analysis. The reference point was the 2000 cohort, companies that sold every unit below cost and called the losses growth. The claim that followed was that every token went out the door at a loss, so scale could only deepen the hole. The mechanics of the model business never matched that description. Reporting around Anthropic's fundraising has repeatedly put the gross margin on its API sales above 80 percent, which means the company-level losses came from somewhere else.

The Information had previously reported that Anthropic had $4.5 billion in revenue in 2025, which I already found difficult to match with Rao's statements. While boosters may claim that “exceeding” could mean literally any number they want above $5 billion, I find it doubtful that the CFO of Anthropic would, under oath, lead the court to believe its business was 30% to 40% smaller than it was, especially when trying to convince it that the damage of being labeled a supply chain risk would ruin its business.