Anthropic's Amended S-1 Puts Compute Commitments at $895B, 7x Its Cash, With $70B Due Within a Year
Anthropic filed an amended S-1 on November 2 that, for the first time, puts one number on what it has promised to spend: about $895 billion in commitments, nearly all of it for computing and data centres. That is almost seven times the roughly $132 billion in cash and short-term investments it held at the end of September.
The original filing on October 26 had left the figures scattered across footnotes, which produced competing tallies. The New York Times counted up to $807 billion. Bloomberg Tax described nearly $900 billion in off-balance-sheet obligations. The amendment settles it.
The schedule
- Total: about $895B in commitments.
- Next 12 months: about $70B.
- By end of 2030: nearly $450B.
- Purchase commitments: about $314B, up from $55B at the end of 2025.
The vendors
- Broadcom: about $161B in equipment leases, the largest line.
- Google: about $121B, up from $114.5B in the first filing.
- SpaceX’s AI unit: up to $120.5B for Nvidia GPU capacity.
- Amazon: $110B.
Two further deals are excluded because neither has a fixed dollar commitment: another $125B to $150B of Google chips, and more than $20B with AMD. If both materialise, the real exposure is higher than $895B.
The short-term capacity penalty
Anthropic books most of its compute 12 to 18 months ahead. Last quarter, demand ran past that capacity and the company bought $5 billion of short-term computing that, in the filing’s words, “currently costs significantly more” than its long-term deals. The quarter also included a $2 billion non-cash charge tied to a donation-matching programme. Anthropic’s first operating profit lasted one quarter.
Context
An earlier report on the prospectus cited $518B in commitments, about 80% of it non-cancelable, so the headline figure is far larger than the one cited then. The structural risk is the same: revenue growth in the filing has to keep outrunning obligations that, for the tail of them, extend into 2045.