Anthropic and OpenAI Seek Smaller 20–30 MW Data Centers to Speed Up AI Deployment
Leading AI labs Anthropic and OpenAI are actively pursuing smaller-scale data center partnerships of 20 to 30 megawatts across North America and Europe. This shift complements their ongoing investments in massive, multi-hundred megawatt or gigawatt-scale hyperscale facilities. Securing smaller compute footprints allows AI providers to bypass long power grid connection queues and heavy construction timelines typical of giant hyperscale centers. This strategy enables faster deployment of computing power to satisfy surging end-user inference demand and regional workloads. Anthropic has begun exploratory discussions in the UK and Nordic regions, while OpenAI has evaluated small-scale capacity in the Nordics and the US. OpenAI stated that it is diversifying its compute portfolio to optimize deployment lead times, reliability, performance, and operational costs across varied AI tasks.
## BACKGROUND
Training frontier AI models requires massive hyperscale data centers consuming hundreds of megawatts to interconnect tens of thousands of GPUs over weeks or months. In contrast, serving AI inference—running existing models to respond to live user requests—requires lower latency and can be distributed across smaller data centers closer to regional power supplies and users.