Anthropic delays its IPO to November: leaked prospectus seeks $2 trillion despite $42 billion in losses in 2025

By Zendoric · October 7, 2026.

Anthropic's IPO, expected in October, is being pushed back to November. The article, by Robert Izquierdo at The Motley Fool and based on a leaked confidential prospectus, does not spell out the reason for the change of date, but it does reveal the scale of the offering: a valuation pointing to $2 trillion, double the $965 billion set in May of this year. If confirmed, it would top the $1.8 trillion that SpaceX, Elon Musk's space company, has reached this year.

Anthropic was founded in 2021 by former OpenAI employees, among them Dario Amodei, today its chief executive. The launch of ChatGPT in late 2022 sent demand for generative AI soaring across the market, and Anthropic capitalized on that momentum with a bet different from its rival's: the enterprise market and AI coding tools. The leaked prospectus portrays that bet as a success: 2025 revenue grew twelvefold over 2024, to $4.6 billion, according to the figures cited from the document.

That growth coexists with losses of a magnitude that is unusual even in this sector: $42 billion in 2025, almost ten times the year's revenue. Anthropic also plans to spend more than half a trillion dollars over the coming years on computing infrastructure to train ever more powerful models, the report notes. It is the same pattern we already know from OpenAI or xAI: the race for the AI frontier is won, or lost, on capital spending, not quarterly profits.

The leaked prospectus adds two risks that rarely appear together in an IPO prospectus. The first is financial and familiar: a quarter of revenue comes from just two customers, and many of the large accounts lack long-term contracts, so losing a single significant customer could hit revenue hard. The second is unusual: among the risk factors, Amodei warns about "recursive self-improvement", the process by which an AI system could autonomously improve versions of itself. In his words, quoted in the report, "unchecked, it could outpace our ability to understand and control these systems". The article also notes that a former Anthropic researcher who left the company this year has reportedly raised the possibility of a catastrophic outcome; this is a personal opinion reported by the outlet, not a verified fact, and should be treated with that caution.

Our reading: that a company includes the existential risk of its own product as a risk factor in the document defining its legal obligations to investors is, in itself, a notable development. It is not just reputation management: it is an acknowledgment, in writing and with legal implications, that the company does not rule out its technology getting away from it. That fits the line Anthropic has held since its founding: growing fast in the enterprise market while publicly arguing for safety guardrails.

In the short term, the case the market has to weigh is a demanding one: losses almost ten times revenue, infrastructure spending that commits half a trillion dollars in the future, and a concentrated customer base that can destabilize the books from one quarter to the next. These are the real risks of building the next generation of models at the pace that competition with OpenAI, Google or the Chinese labs demands. But if the leap in capabilities this investment is chasing translates, as we argue at Zendoric, into medium-term abundance — models capable of accelerating medical research, automating routine work and freeing up human capacity for what really matters — then this IPO is not just a financial event: it is proof that the market is beginning to put a price, doubts and all, on that promise.