Anthropic IPO filing reveals 85% gross margin on AI models

The debate around Anthropic's IPO filing has focused on its risk factors, its $80 billion headline net loss, and its governance structure. The 85% gross margin on running existing models cuts against a specific worry: that consumption-based, pay-as-you-go pricing for AI wouldn't be profitable. As The Information put it, the number is "a reality check" for skeptics.

The comparison set the filing invites is telling. At roughly 81-82%, Meta's ad machine sits in the same range, while analysts' estimates for Google Search ads land at 75–80% and the iPhone in the mid-40s. That places the core economics of operating AI models alongside some of the most profitable businesses in tech.

The caveats are built into the number itself. Anthropic is still losing a lot of money overall, and the margin excludes the cost of training new models — so it describes the economics of running current products, not the full cost of keeping them at the frontier. For OpenAI, which runs a similar model but is considered less efficient in part because of its massive non-paying user base, the figure offers a comparable data point rather than a verdict.