Anthropic and OpenAI hunt for smaller data center deals

Anthropic and OpenAI are seeking smaller-scale data center capacity - roughly 20-30 megawatts per deployment - in markets such as the U.K., the Nordics and parts of the U.S., supplementing the multi-hundred-megawatt and gigawatt deals they already hold. Both firms have signed massive long-term arrangements (Anthropic with a roughly $45 billion, ~460 MW deal and OpenAI with multi-gigawatt commitments across several U.S. projects), but are adding smaller allocations because existing large builds face community resistance, land and power limits, and long lead times. Securing modest blocks of capacity at already powered sites lets them bring usable compute online faster and spread workloads across multiple locations.

The practical driver is a shift from large, tightly coupled training clusters toward distributed inference, which can run on many smaller clusters. Industry analysis and JLL real estate projections show inference growing from a minority of data center workloads today to a plurality by the late 2020s, increasing demand for dispersed, lower‑capacity sites. Hardware and data-center players are responding: Nvidia is studying smaller-scale centers for inference and neoclouds like Crusoe are pivoting into faster, cheaper small builds while raising large funding rounds. The strategy lets AI labs diversify their compute portfolio, matching different infrastructure types to distinct performance, timing and cost requirements.