There is significant uncertainty regarding the finances of leading AI laboratories, but as Anthropic approaches its IPO, transparency is increasing. The company is revealing details about the economics of compute, specifically highlighting the portion of its rapidly expanding revenue that is being consumed by these costs.
In a Monday update to its S-1 filing, Anthropic disclosed data indicating that margins on compute are actually declining. While the previous filing covered financial results through June 30th, the new data extends coverage to September 30th. Despite a strong quarter where revenue surged to $17 billion from $11.5 billion, the operating loss simultaneously widened, driven primarily by escalating compute expenses.
The situation becomes more concerning when examining compute margins. In the latest quarter, compute accounted for 58% of the company's revenue, up from 48% in the preceding quarter. This contrasts sharply with the goal of winning over IPO investors by demonstrating improving margins, as the trend is currently moving in the opposite direction.
A chart included in the filing illustrates this trend by tracking revenue generated for every dollar spent on compute. Since the beginning of 2025, this ratio has steadily increased, peaking at just over 2x in the recent quarter before settling back down to 1.72x.
While these shifts often indicate temporary fluctuations rather than permanent issues, they provide crucial insight into running a major laboratory. To meet surging demand, Anthropic's compute bill for the quarter reached $10 billion. The filing notes that "$5.0 billion of compute under short-term commitments currently costs significantly more than comparable compute procured under multi-year commitments." Essentially, the rapid expansion of demand forced Anthropic to scramble for resources, incurring higher rates.
Several of these arrangements have been made public. Anthropic is paying SpaceX $1.25 billion monthly for compute, alongside multi-billion dollar agreements with Nscale, Akamai, and Volta. In the industry, these are classified as "neoclouds"—specialized providers offering flexible compute deals at premium rates compared to traditional hyperscalers.
Margins for Anthropic contracted in Q3 due to such agreements, and there is no immediate indication that the company can operate without them. As stated in the filing, "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 actively attempting to reduce these expenses, particularly through its "Anthropic Direct" project aimed at building its own compute infrastructure. However, given the speed of demand growth, these efforts are insufficient to keep pace, leaving neoclouds as the most viable immediate option.
From one perspective, these are "champagne problems." As long as revenue is growing by nearly 50% quarterly, most investors are unlikely to be overly concerned about high compute costs. However, should demand eventually slow, investors might find comfort in a compute landscape that eventually becomes more favorable. Nevertheless, this serves as a stark reminder of how heavily AI laboratories depend on the trillion-dollar infrastructure buildout and their vulnerability to disruptions in that supply chain, such as skittish bond markets or local opposition to data centers.
Consequently, the future of Anthropic and its peers remains heavily contingent on the stability and pricing of compute resources.
The Editorial Staff at AIChief is a team of professional content writers with extensive experience in AI and marketing. Founded in 2025, AIChief has quickly grown into the largest free AI resource hub in the industry.