Anthropic's new filing shows compute eating a bigger share of revenue
An updated S-1 from AI company Anthropic shows compute costs took 58% of revenue in the latest quarter, up from 48% the quarter before. Revenue rose to $17 billion.
Anthropic, the AI company preparing for an initial public offering (IPO), filed an updated S-1 on Monday, the registration document submitted to US securities regulators. It reveals that compute costs are consuming a growing share of its revenue, a point that matters to prospective investors.
The S-1 published last week covered financial results through June 30. The new update adds the quarter ending September 30. In that period Anthropic's revenue reached $17 billion, compared with $11.5 billion in the previous quarter. The operating loss also grew, largely because of higher compute costs.
A larger cost share
Compute took up 58% of revenue in the quarter, versus 48% before. A chart in the filing tracks revenue per dollar spent on compute: the figure rose steadily from the start of 2025, peaked at just over 2x last quarter, then fell to 1.72x.
Total compute spending reached $10 billion. Of that, $5.0 billion was under short-term commitments, which the filing says cost significantly more than comparable compute bought under multi-year commitments. Demand grew so quickly that the company had to scramble for capacity, paying higher rates.
Suppliers and outlook
Anthropic pays SpaceX $1.25 billion a month for compute, and has multi-billion-dollar deals with Nscale, Akamai and Volta. These are so-called neoclouds, smaller providers that offer more flexible but pricier terms than the large hyperscale cloud companies.
The company expects compute and infrastructure expenses to keep rising in absolute terms and to remain its largest operating expense for the foreseeable future. Anthropic is trying to lower costs by building its own compute base under a project called “Anthropic Direct”, but TechCrunch assesses that this cannot keep pace with demand. The outlet views the margin dip as likely temporary, while noting it shows how dependent AI labs are on the pace of data center construction.
