Anthropic’s Compute Costs Soar, Impacting IPO Prospects Despite Revenue Growth

As Anthropic approaches its IPO, more financial details about major AI labs are emerging, specifically concerning how much of Anthropic’s escalating revenue is being consumed by compute expenses.

In a recent filing on Monday, Anthropic disclosed data indicating a decline in its compute margins. This update to last week’s S-1 extended financial reporting an additional quarter, through September 30th, beyond the original June 30th coverage. Overall, Anthropic experienced a strong quarter, with revenue reaching $17 billion, up from $11.5 billion previously, signaling sustained growth. However, operating losses simultaneously expanded, largely due to rising compute costs.

Further examination of compute margins paints an even less favorable picture. In the latest quarter, compute expenses constituted 58% of the company’s revenue, an increase from 48% in the prior quarter. This worsening trend occurs just as the company would ideally be demonstrating margin improvements to attract IPO investors.

A chart within the filing visually reinforces this, tracking the company’s revenue generated per dollar spent on compute. After steadily climbing since early 2025, peaking just above 2x last quarter, it has now decreased to 1.72x.

While many suggest this shift in compute margins is a temporary fluctuation rather than a long-term issue, it still reveals important insights about managing a prominent AI lab. To meet demand, Anthropic’s compute bill surged to $10 billion for the quarter, including, as noted in the filing, “$5.0 billion of compute under short-term commitments, which currently costs significantly more than comparable compute procured under our multi-year commitments.” Essentially, rapid demand growth compelled Anthropic to quickly acquire compute, leading to higher expenditures.

Some of these agreements have been publicly disclosed. Anthropic has been paying SpaceX $1.25 billion monthly for compute, alongside multi-billion dollar contracts with Nscale, Akamai, and Volta. In industry terms, these are referred to as

Leave a Reply