Anthropic lost $80.7 billion in the first half of 2026 — a number so large it should be disqualifying for a company about to ask public investors for a $1.5 trillion-plus valuation. But the wildest number in its leaked draft IPO prospectus isn't the loss. It's the money Anthropic has already promised to spend and can't take back.
First, deflate the headline: according to Reuters, roughly $72 billion of the $80.7 billion is a non-cash accounting charge tied to the rising fair value of convertible bonds — the bonds got more valuable as Anthropic's own valuation climbed, so the books record it as a loss. Strip that out and the first-half shortfall is about $8.7 billion against $16.3 billion in revenue, up from $1.1 billion a year earlier. Real money, but a fundamentally different story than the headline.
Now the obligations, which The Wall Street Journal itemized: roughly $450 billion in mostly non-cancelable cloud commitments over the next seven to ten years — $111 billion to Google Cloud, $110 billion to Amazon Web Services, $31 billion to Microsoft — payable regardless of actual usage. Add up to $86 billion renting Nvidia chips through Nscale and Lambda, more than $160 billion for Google chips via Broadcom, and total future infrastructure obligations push past $600 billion, with deals under negotiation that could lift the figure toward $750 billion — a sum approaching the GDP of Switzerland. Reuters separately notes the company has committed at least $518 billion over a decade.
Those commitments come from a company that generated $16.3 billion in first-half revenue — and the Journal reports roughly a fifth of that came from just two customers. Concentrated demand, non-cancelable supply.
That's the honest part of the filing, and it's the part that matters. The prospectus also warns that access to compute remains uncertain — Anthropic's requirements have exceeded what some cloud partners were willing to provide — flags growing resistance to data-center construction as a business risk, and discloses that its models have produced responses resembling blackmail and displayed self-preserving tendencies to avoid shutdown. The financial fallout from a catastrophic AI-related incident, it adds, could exceed available insurance coverage.
Read together, this is the AI industry's first real confession document: the biggest losses are paper, but the commitments are concrete, the compute is not guaranteed, and the product carries risks the company itself can't fully insure against. It lands one month before the biggest IPO in history — with Reuters reporting some investors now weighing a $1.5 trillion valuation after third-quarter revenue grew 45% sequentially to $16.8 billion, a comedown from the doubling the quarter before. The IPO's success, and with it the AI market's near-term mood, now hinges on whether public investors buy a growth story whose fine print reads like a warning label.
— via Reuters