Anthropic is still grappling with the economics of compute costs

As artificial intelligence labs inch closer to major public offerings, more details are emerging about their internal finances and operational economics. Anthropic has released updated financial data showing that despite impressive revenue growth, a massive portion of its earnings is being consumed by skyrocketing compute and infrastructure costs.

Key highlights from the Anthropic financial update - Quarterly revenue reached $17 billion, up significantly from $11.5 billion in the previous quarter. - Operating losses grew concurrently due to expanding compute expenses. - Compute costs accounted for 58% of the company's revenue, rising from 48% in the prior quarter. - Total quarterly compute spending hit $10 billion, including $5 billion under short-term commitments.

Short-term deals and the reliance on neoclouds To keep pace with the overwhelming demand for its AI models, Anthropic has been forced to secure expensive short-term compute deals. The company has partnered with smaller neocloud providers like SpaceX—paying $1.25 billion a month—alongside agreements with Nscale, Akamai, and Volta. These flexible contracts come at a significantly higher rate compared to multi-year commitments with traditional hyperscalers.

Efforts to lower costs through Anthropic Direct Anthropic is actively attempting to mitigate these high expenses by building its own compute base through an initiative called "Anthropic Direct." However, because market demand is growing at such a rapid pace, internal infrastructure efforts cannot keep up, leaving neoclouds as a necessary option. The company expects infrastructure expenses to remain its largest operating cost for the foreseeable future.

What this means for the AI industry Anthropic's financial disclosures highlight the immense capital dependency that major AI labs have on the broader trillion-dollar infrastructure buildout. While rapid revenue growth keeps investors confident for now, any bottlenecks in data center construction or shifts in the bond market could expose vulnerabilities in the business models of top AI developers.