Anthropic, OpenAI hunting 20–30 MW DC deals

Opening

Anthropic and OpenAI are hunting smaller AI data-center capacity deals in the ~20–30 MW range, not only multi-hundred-MW / GW campuses. CNBC (Kai Nicol-Schwarz), citing sources, says Anthropic has sounded out agreements in that range across the U.K. and the Nordics (four people); OpenAI has explored similar smaller deployments in the Nordics (two of those sources); one source also cited U.S. talks at that scale for both. An OpenAI spokesperson said the company builds a "diversified compute portfolio" and assesses partners on requirements, performance, reliability, timing, and cost, without commenting on specific commercial discussions. Anthropic did not comment.

The numbers

~20–30 MW — smaller capacity deal size under discussion (CNBC sources). Anthropic–Nscale — ~$45B / ~460 MW West Virginia (CNBC sources, prior August reporting). OpenAI Stargate — original 10 GW commitment surpassed; +3 GW Georgia and +8 GW Ohio stated by OpenAI (CNBC). Inference share of global DC workloads — 9% in 2025 vs training 14%; projected 37% inference vs 13% training by 2030 (JLL via CNBC).

Why it matters

Structure Research's Jabez Tan frames the attraction as "speed to usable capacity" — a few megawatts at an existing powered site can beat waiting for a larger block — and notes many inference workloads can serve across separate smaller clusters. CNBC ties the shift to community pushback on huge campuses and European land/power scarcity. Same window as Crusoe's modular/smaller-site push (news).

Positioning

AI capex durability — Labs still racing for deployable watts; the mix tilts toward distributed smaller blocks for inference without cutting the large-campus path.

Connects to

DC moratoriums map — Local pushback is one reason smaller/faster sites matter (news). Crusoe Series F / Spark modular — Same morning theme: neocloud capital for large campuses and truckable smaller AI factories (news).