Break the $42 billion apart first
The headline number is startling: a company with under $5 billion in annual revenue lost $42 billion in a year. But according to Reuters, **about $34 billion of it is an accounting charge.** Anthropic has issued financing instruments that could eventually convert into company shares; the faster its valuation rises, the higher those instruments' book value, and the difference is recorded as an expense. The money wasn't actually spent, and it doesn't mean the business got worse; in a sense it's the opposite, a side effect of the rising valuation showing up on the books. **The number that reflects operations is a different one: an operating loss of more than $8 billion.** That is still large, nearly twice revenue, but it is a problem of a different order of magnitude from $42 billion. The first step in reading this prospectus is to keep those two figures apart.
The money goes mostly to compute
Total 2025 operating expenses were $12.65 billion, of which compute and infrastructure accounted for $7.33 billion, roughly triple the previous year and more than half the total. The bigger number comes next: the prospectus shows about $518 billion in cloud, computing and infrastructure obligations over coming years. Set against roughly $4.6 billion in 2025 revenue and about $20.3 billion in cash at year-end, **this prospectus is essentially explaining to public markets that the next several years of compute bills must be covered by revenue growth and IPO proceeds together.** We wrote around September 17 that Anthropic had signed about $517 billion in compute contracts over eleven months. The $518 billion in the prospectus is of the same magnitude; Reuters does not give a breakdown, so for now we can only say the figures line up, not that they are the same set of contracts.
Two risk factors worth noting
The first is customer concentration: nearly a quarter of last year's revenue came from two customers, and many large clients are not on long-term contracts and can cut or stop spending at any time. For a company targeting a $2 trillion valuation, that is a very concrete vulnerability. The second is that the prospectus puts the company's own safety research into its risk factors: increasingly autonomous models may exhibit harmful behaviors such as sabotaging code, aiding fraud and manipulating information. **Putting this into a legal document addressed to investors means it is no longer just a discussion in research papers, but a risk the company acknowledges could affect the business.**
Read it alongside last week's story
On September 26 we reported that Anthropic is asking shareholders to approve a structure giving its seven co-founders 50.1% of voting power on about 14% of the shares, when its secondary-market valuation was about $1.5 trillion and its reported IPO target was up to $2 trillion. Today's prospectus details fill in the other half: **the control structure settles who decides after the listing; the prospectus has to answer why the company is worth that price.** Limits of what is known: the prospectus has not been made public, all figures here come from Reuters' exclusive report, and the final public version may differ; no public response from Anthropic to the report has been seen.
via: Reuters exclusive via U.S. News, Reuters via Yahoo Finance, CNBC citing Reuters