AI leaders lease mid-sized data centers to bridge capacity gap during mega-project buildout
OpenAI and Anthropic are negotiating leases for 20-30 MW data centers in the UK, Nordics, and US to quickly add compute capacity while their larger gigawatt-scale projects remain under construction. The companies have already committed tens of billions to massive AI infrastructure but need immediate capacity to meet surging demand. An OpenAI spokesperson said the firm is building a diversified compute portfolio to address varying workload needs.
The lease talks target existing facilities in the UK, Nordics, and US, with both companies seeking 20-30 MW of capacity. These smaller deployments handle inference workloads, which can spread across multiple clusters, unlike training that requires tightly interconnected GPU arrays. The approach also avoids community resistance — 45 US projects worth $68 billion were blocked in Q2 2026 over land, noise, and resource concerns.
Anthropic's reported $45 billion Nscale deal covers 460 MW in West Virginia, while OpenAI's Stargate commitments exceed 10 GW, plus additional Georgia and Ohio projects. Such gigawatt campuses take years to complete due to grid, substation, and cooling requirements, leaving a capacity gap that smaller facilities can fill.
This dual-track strategy could accelerate AI service availability for businesses and consumers, who may see faster rollout of new inference capabilities while mega-campuses remain under construction. However, scattered mid-sized facilities could still strain local power grids and draw scrutiny from nearby communities over resource use. The approach may also signal that AI infrastructure demand is broad enough to support both distributed and concentrated compute models, potentially influencing how regional markets plan energy capacity.