Anthropic's AI Cheaper, But Business Faces Compute Cost Problem
Anthropic's AI is becoming cheaper to use, creating a problem for its business due to rising compute costs.
"AI's getting cheaper for you, but Anthropic's paying big. Their compute costs are going wild. Will they fix it before the IPO?"
Anthropic is grappling with the economics of compute as its AI becomes cheaper for users, leading to worsening compute margins. Despite skyrocketing revenue, a significant portion is being lost to compute costs. Recent data shows compute taking up 58% of the company's revenue in the latest quarter, an increase from 48% in the previous quarter. This trend is occurring as Anthropic approaches its IPO.
The company's compute bill reached $10 billion for the quarter, partly due to short-term commitments for compute, which are more expensive than multi-year agreements. Demand growth has forced Anthropic to procure compute from "neoclouds" like SpaceX, Nscale, Akamai, and Volta, which offer flexible but higher-rate deals. Anthropic's revenue for every dollar spent on compute fell to 1.72x this past quarter, down from a peak of just over 2x.
While revenue is growing significantly, the increasing operating loss due to compute costs presents a challenge. Anthropic expects compute expenses to remain its largest operating expense, even with efforts like "Anthropic Direct" to build its own compute base. The situation highlights the dependence of AI labs on infrastructure buildout and their vulnerability to price fluctuations and potential slowdowns in data center development.
This situation reveals the significant operational challenges and cost pressures faced by leading AI companies. Businesses relying on AI services may see future pricing influenced by these underlying infrastructure costs. It underscores the importance of efficient compute resource management in the rapidly expanding AI sector.
Find the right AI tool for your business
Chat with Insta and get matched to the right tool in seconds.
Try Insta Tool Finder →