There's considerable opacity surrounding the financials of leading AI labs, but as Anthropic moves closer to its IPO, some clarity is emerging. The company is now revealing more details about the economics of compute—specifically, how much of its rapidly growing revenue is being consumed by compute expenses. A new filing released on Monday showed that compute margins are actually deteriorating. This filing updated the S-1 that came out last week; while the earlier financial disclosures only went through June 30th, the new data extended results through an additional quarter, to September 30th. On the whole, Anthropic had a strong quarter: revenue hit $17 billion, up from $11.5 billion the prior quarter, indicating sustained growth. However, operating losses grew simultaneously, driven largely by rising compute costs. Examining compute margins paints an even grimmer picture. In the new quarter, compute accounted for 58% of the company's revenue, up from 48% the previous quarter. At a time when the company could attract IPO investors by demonstrating improving margins, they're moving in the wrong direction. A chart in the filing illustrates this even more starkly, tracking the company's revenue for every dollar spent on compute. Since the beginning of 2025, that ratio has been climbing steadily, peaking at just over 2x this past quarter before dropping back to 1.72x.

Many signals suggest this shift in compute margins is more of a temporary fluctuation than a long-term issue, but it still reveals important truths about running a major lab. In an effort to keep pace with demand, Anthropic's compute bill surged to $10 billion for the quarter, including—as the filing notes—"$5.0 billion of compute under short-term commitments, which currently costs significantly more than comparable compute procured under our multi-year commitments." Essentially, demand was growing so rapidly that Anthropic had to scramble for compute, which meant paying premium rates. Some of those arrangements have already become public. Anthropic has been paying SpaceX $1.25 billion a month for compute, alongside multi-billion dollar deals with Nscale, Akamai, and Volta. In industry terminology, these are neoclouds rather than hyperscalers—smaller providers that offer more flexible compute arrangements at higher prices. Anthropic's margins worsened in Q3 because of deals like these, and there's no indication they'll be able to avoid them anytime soon. As the new filing states, "We expect compute and infrastructure expenses to increase in absolute dollars and remain our largest operating expense for the foreseeable future as additional committed compute becomes available to us under multi-year arrangements with CSPs, and as we continue to invest in additional compute."

Anthropic is still working to reduce its compute costs. In particular, the company discusses at length its efforts to build its own compute infrastructure under a project called "Anthropic Direct." But with demand growing so quickly, those efforts won't be able to keep up, and the neoclouds will remain the best option available. In a sense, these are champagne problems. As long as revenue is growing at nearly 50% per quarter, few investors will fret about high compute costs—and once demand slows, investors may take comfort in a compute picture that has brightened considerably. But it's a reminder of how heavily AI labs still depend on the trillion-dollar infrastructure buildout—and how exposed they are to disruptions in that process. It's easy to envision the data center buildout decelerating over the next year if bond markets become jittery or local opposition intensifies. There's little Anthropic can do to mitigate either of those risks, and either one would make an already messy compute situation far messier. In the meantime, the future of Anthropic and other labs will hinge on compute prices.