Distributed compute capacity is being sought by Anthropic and OpenAI across the U.S., U.K., and Nordic regions to bring AI services online faster, sources tell CNBC. These deals target 20 to 30 MW of power per site, leveraging existing infrastructure to avoid the long wait times associated with multi GW campus builds. Such clusters are specifically designed for inference workloads, which can be spread across multiple sites rather than one massive facility.

The strategy complements OpenAI's multi GW Stargate projects and a $45 billion agreement between Anthropic and Nscale for 460 MW in West Virginia. JLL forecasts that inference will overtake training as a share of data center capacity in 2027 and reach 37% of global workloads by 2030 from 9% in 2025.