Data as of 2 November 2026 ยท from the S-1 and Anthropic’s own statements
In short
Anthropic spent $5,558 million on compute and infrastructure in Q2 2026, according to its Form S-1. That was 48% of the quarter’s revenue, down from 251% in Q1 2025, because revenue grew faster than the compute bill. The bill keeps rising: fixed or minimum compute commitments were about $265.6 billion at 30 June 2026, and Anthropic signed about $245 billion more after that date.
The S-1 reports “compute and infrastructure” as Anthropic’s largest operating expense. The quarterly table covers Q1 2025 to Q2 2026. Compute cost rose every quarter, from $944 million to $5,558 million, about 5.9 times (our arithmetic). Revenue rose about 30.7 times over the same quarters (our arithmetic), so compute fell from 2.5 dollars per dollar of revenue to under half a dollar.
Anthropic S-1 (26 Oct 2026), quarterly results. Compute = 251% of revenue in Q1 2025, 48% in Q2 2026.
How we source this page. Anthropic is private until its IPO, so the primary sources are its Form S-1 (filed with the SEC on 26 October 2026) and its own announcements on anthropic.com. Figures that only the press has reported are labelled “reported by” with the outlet named. Ratios, sums and multiples are our arithmetic from those figures and are labelled as such. Every quote below was checked against its source on 2 November 2026.
By year
Compute and infrastructure cost $433 million in 2023, $2,522 million in 2024 and $7,326 million in 2025, and $8,945 million in the first half of 2026. Against revenue that is about 10.6 times revenue in 2023, 6.5 times in 2024, 1.6 times in 2025 and 0.55 times in the first half of 2026 (all our arithmetic). The S-1 explains why the line moves in steps: Anthropic says it typically buys compute 12 to 18 months before it is available, so the cost follows contracted capacity, not same-quarter demand.
Anthropic Form S-1 (SEC EDGAR, filed 26 Oct 2026), consolidated statements of operations. Our arithmetic: compute / revenue = 433/41, 2,522/386, 7,326/4,585, 8,945/16,286.
The commitments behind the bill
The expense line shows compute already delivered. The commitments show what is already contracted. Non-cancelable purchase commitments were $54,559 million at 31 December 2025. At 30 June 2026 the S-1 puts expected fixed or minimum guaranteed commitments at about $265.6 billion, with terms of one to 20 years. After 30 June, Anthropic entered new and amended agreements with about $245 billion of future minimum payments, plus about $21 billion of short-term minimum payments.
Anthropic Form S-1 (SEC EDGAR, filed 26 Oct 2026), notes on commitments and subsequent events. The S-1 leaves the total of all obligations blank.
The total of all commitments is blank. The S-1 sentence that sums contractual obligations, short-term and later commitments, and leases not yet commenced leaves the dollar amount empty, and the contractual obligations table is unfilled.
Training versus inference. The S-1 reports one compute and infrastructure line; it does not split it by training and serving in the passages we read, so we do not either.
Agreements with no fixed dollar amount are left out of the $265.6 billion; the S-1 says payments under them are expected to be significant.
Frequently asked questions
How much does Anthropic spend on compute?
$7,326 million in 2025 and $8,945 million in the first half of 2026, of which $5,558 million in Q2 2026, per the S-1 line “compute and infrastructure”.
No longer. Compute was 251% of revenue in Q1 2025 and 48% in Q2 2026, per the S-1.
How much compute has Anthropic committed to buy?
About $265.6 billion of fixed or minimum commitments at 30 June 2026, plus about $245 billion of new and amended agreements after that date, per the S-1.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
