Anthropic’s compute bill rises as revenue reaches $17 billion
Anthropic’s updated S-1 filing shows quarterly revenue of $17 billion for the period ending September 30, up from $11.5 billion in the preceding quarter. At the same time, compute represented 58% of revenue, compared with 48% a quarter earlier, while the company’s operating loss increased.
The disclosure offers a sharper view of the economics facing a large AI lab as Anthropic moves towards an initial public offering. Revenue growth remained strong, but the added demand required capacity that was more expensive than the company’s longer-term arrangements.
Short-term capacity raised the cost of growth
Anthropic said its compute bill reached $10 billion in the quarter. That figure included $5.0 billion of compute under short-term commitments, which the filing says currently cost significantly more than comparable compute obtained through multi-year commitments.
The company’s revenue generated per dollar of compute had been rising since the start of 2025, reaching slightly above 2x before falling to 1.72x in the newly disclosed quarter. The movement illustrates how quickly additional infrastructure demand can affect margins even as top-line revenue rises.
Anthropic has publicly committed to capacity arrangements with SpaceX, Nscale, Akamai and Volta. These providers are described as neoclouds: smaller infrastructure suppliers that can offer flexible deals, albeit at higher rates than large hyperscale cloud platforms.
Longer-term infrastructure remains central
Anthropic expects compute and infrastructure expenses to rise in absolute terms and remain its largest operating expense for the foreseeable future. The filing ties that expectation to further multi-year commitments with cloud service providers and continued investment in additional compute.
The company is also developing an internal compute base under a project called Anthropic Direct. Yet fast-growing demand means that externally sourced capacity remains necessary. The pressure is consistent with Anthropic cloud capacity constraints highlighting constrained cloud capacity under contracts exceeding $750 billion, while Anthropic continues to secure resources from several providers.
What the filing means for AI buyers and investors
The figures do not establish that the deterioration in compute margins is permanent. They do show, however, that AI revenue growth can require expensive short-term capacity before lower-cost, committed infrastructure becomes available.
For businesses assessing AI suppliers or building their own AI services, the practical implication is to evaluate demand forecasts alongside capacity contracts, pricing duration and reliance on external infrastructure, rather than treating revenue growth alone as evidence of sustainable operating economics.