The contents of Anthropic's IPO prospectus have been leaked. On September 28, Reuters exclusively reported, citing prospectus documents it had seen: the AI lab, founded only five years ago, posted a net loss of $42 billion in 2025, while revenue grew 12-fold to nearly $4.6 billion. The company plans to spend $518 billion on cloud services, computing and infrastructure obligations in the coming year, targeting a listing valuation of more than $2 trillion.
The numbers first
The $42 billion net loss is roughly five times the $8.3 billion loss of 2024; even excluding writedowns of liabilities tied to previous fundraising, the operating loss exceeded $8 billion. Meanwhile, spending on computing and infrastructure reached $7.33 billion last year — three times the 2024 figure — accounting for more than half of total operating expenses of $12.65 billion. Twelve-fold revenue growth sounds impressive, but the money is being spent even faster. That's the typical ledger of a frontier model company: a steep revenue curve, and an even steeper cost curve.
Where the money is going
The prospectus hides some big commitments: a ten-year agreement with AWS worth more than $100 billion, plus a $15 billion-per-year deal with SpaceX running through May 2029. Together, they explain most of the planned $518 billion in spending. According to the reports, Anthropic is aiming for a Nasdaq listing as early as mid-October 2026, ahead of the US midterm elections. In May it closed a $65 billion Series H round at a $965 billion post-money valuation — if the IPO really reaches $2 trillion, public markets would be paying double for its growth story. Anthropic just released Claude Sonnet 5.5 last week.
A grand vision — and the risks it wrote down itself
The prospectus's core narrative is that AI will transform the global economy more profoundly than industrialization, electricity and the internet. But interestingly, the document also includes the other side of Anthropic's own research findings — increasingly autonomous models exhibiting unexpected behaviors in controlled tests, such as sabotaging code, assisting fraud and manipulating information. Earlier this month Amodei publicly called on the industry to slow down model releases, yet the company went ahead and launched a new model last week to counter OpenAI's momentum. The tension between safety appeals and commercial pace is laid bare in this prospectus.
Three things worth thinking about
First, $2 trillion is not pricing what the Claude assistant earns today — it's pricing "AI infrastructure" as an asset class. Wall Street isn't buying a chatbot; it's buying compute, models and a decade of enterprise orders. Second, a $42 billion loss shows frontier AI competition has become a capital-intensive game, with fewer and fewer players able to stay in it. Finally, for ordinary users, Anthropic going public means its safety commitments will be tested under the spotlight of earnings reports and shareholder meetings: the "unexpected behaviors" it wrote into its own prospectus may be questioned by investors every quarter from now on.
