Anthropic's Revenue Reaches $17 Billion as Compute Takes a Larger Share
Anthropic reported $17 billion in revenue for its latest quarter, up from $11.5 billion in the quarter before, according to an updated S-1 filing that extends its financial disclosures through September 30. The prior S-1 covered only through June 30. TechCrunch reported the figures on November 3, 2026.
The filing's cost disclosures are drawing the closest attention. Compute, the processing capacity that runs the company's models (typically GPU clusters rented from cloud and data center providers), accounted for 58% of revenue in the latest quarter, up from 48% in the quarter before. The quarterly compute bill was $10 billion, and $5.0 billion of that was purchased under short-term commitments. Short-term capacity costs significantly more than comparable compute bought under multi-year commitments, which is the rough equivalent of paying nightly hotel rates instead of signing a long lease. Revenue per dollar of compute spend fell to 1.72x, after peaking at just over 2x in the prior quarter.
Anthropic said it expects compute and infrastructure expenses to rise in absolute dollars and to remain its largest operating expense for the foreseeable future. It is also working to build its own compute base under a project it calls "Anthropic Direct." The company is at an early stage of that effort, and the filing gives no timeline for it.
Bloomberg reported on November 2 that Anthropic revenue reached nearly $17 billion ahead of an IPO, and that the company disclosed a twelve-fold increase in quarterly revenue. Bloomberg also reported that Anthropic intends to spend more than $895 billion in the years ahead. That is a stated intention, not a set of committed outlays already on the books.
On the bottom line, Yahoo Finance reported an operating loss of $567 million for the third quarter of 2026. Operating loss measures the shortfall from core business activity, before financing and other non-operating items. A separate Reuters report, citing the Wall Street Journal, put Anthropic's net loss for the first half of 2026 at $80.7 billion. That loss widened in large part because the fair value of bonds rose. A net figure driven by a balance-sheet revaluation says little about unit economics, which is why the operating line matters more to anyone modeling the business.
In my view, the central question in these numbers is not demand. Revenue has grown faster than the cost base, but compute costs are growing as a share of revenue, and the quarter-over-quarter decline in revenue per compute dollar points the same way. Short-term capacity is the most expensive way to buy inference at scale. Multi-year commitments and owned infrastructure are the levers that would bend that curve, and both carry their own risk, since they lock in capacity and capital before utilization is certain.
The broader context here is that the compute market is pricing Anthropic's growth at a premium, and the filing shows that premium being paid in real time. If the company moves a larger share of its load onto longer commitments or its own base, the same revenue would translate into materially better margins. If it cannot, a rising proportion of each revenue dollar will flow to the data center providers and chip supply chain beneath it. The next set of disclosures will show whether the 58% figure is a peak or a trend.
For the wider industry, the disclosure offers a rare line-item view of what frontier model providers pay for inference capacity, and what that capacity costs when bought on short notice. Whatever the IPO outcome, the cost structure in this filing is likely to inform how the market prices the compute strategies of every competitor.
