Anthropic’s IPO case is colliding with pressure to release another model

Anthropic executives met prospective public investors last month after a sharp change in operating economics: The Information reports that the company went from spending about $2.30 on operations for every dollar of revenue in spring 2025 to a slight operating profit in the June quarter. Reuters says annualized revenue passed $65bn by the end of July.

The growth is exceptional, but investors are testing its durability. OpenAI’s GPT-6 Astra represented about 13% of enterprise AI spending tracked by Ramp, against roughly 8% for Claude Fable; OpenRouter said spending on OpenAI models exceeded Anthropic’s for the first time in more than two and a half years. Anthropic is therefore considering another model release even as Dario Amodei calls for slower capability development. The company says safety evaluation is part of the timing decision.

There are deeper concerns than one release cycle. The FT reports that only 22.5% of Anthropic customers remain after a year—better than rivals, but still evidence of easy switching—and that cheaper open models have cut some AI costs by as much as 40%. Meta, a major Anthropic customer, is also building more capability internally.

The IPO could move until after the November midterms. Delay would give Anthropic another quarter of results, but it cannot settle the structural question: whether proprietary labs can keep enough performance and workflow advantage to hold customers while models improve, prices fall and large buyers develop alternatives. A new model can defend share; it also illustrates why a safety-led laboratory finds it hard to slow when public-market investors are watching weekly usage data.