Power availability, rather than chip supply, is becoming a constraint on AI infrastructure. New data centers can face years of waiting for permission to connect to congested electrical grids, encouraging developers to build where electricity is already available.
French AI infrastructure company Antimatter is pursuing that approach with a network of 1,000 modular micro data centers beside energy sources in the US, Europe, and GCC regions. The facilities turn electricity that existing transmission connections cannot deliver to customers into on-site AI computing capacity.
Available power drives deployment
Each container-style unit accommodates up to 400 GPUs and can be deployed in roughly five months. Conventional hyperscale facilities often take more than two years to reach comparable readiness.
Wind, solar, hydro, and biogas sites are the primary targets. Transmission constraints can prevent those installations from delivering all their generation to customers; neighboring data centers can instead consume power that would otherwise be restricted, without waiting for new transmission lines.
Antimatter says it has secured more than 1GW through grid connection agreements and reserved locations. It reports over 160MW already operating in Texas and Oregon, with an initial footprint of ten units at eight sites and hundreds of further installations in development.
Expansion through 2030
The first major phase calls for 100 deployments in 2027, supporting more than 40,000 GPUs and about 3.6 exaFLOPS of computing capacity. By the end of 2030, the company aims to reach 1,000 sites across dozens of countries, providing more than 400,000 GPUs and roughly 36 exaFLOPS.
David Gurlé, Antimatter's Cofounder, Executive Chairman, and CEO, argues that energy—not intelligence—is the limiting factor in AI. He describes centralized infrastructure as a model built for the first cloud and AI era, while inference requires distributed facilities that deploy faster and are sovereign by design.
Inference demand comes from trained models running continuously in copilots, automated services, and real-time decision systems. Shared software connects the smaller facilities into one network while locating processing closer to users.
