Google and Constellation sign 890 MW nuclear plus AI energy deal

This deal crystallizes how the energy and AI races are fusing. Google is not just buying renewable credits anymore, it is co‑funding nuclear uprates that add 890 MW of real, firm capacity to a constrained U.S. grid while locking in 2,700 MW of additional supply. At the same time, Constellation is standardizing on Google Cloud and Gemini Enterprise to design “AI for energy” workflows that speed site selection, permitting, uprate planning and outage management. In effect, Google is underwriting both the electrons and the optimization software its own data centers will need.

For the AGI race, the bottleneck has quietly been shifting from chips to power. Frontier training clusters are increasingly power‑limited inside regions like PJM, and regulators are starting to push hyperscalers to “bring their own power.” This structure is a template others can copy: marry decade‑long AI demand curves to baseload nuclear and then use agentic AI to wring more capacity out of existing plants. If it works, it blunts one of the more credible arguments for slowing large‑scale AI deployments, namely that the grids cannot handle the load.

It also deepens Google’s advantage over smaller labs and startups that cannot sign multi‑GW, multi‑decade power deals. As access to cheap, reliable power becomes a differentiator alongside access to GPUs, the AGI race tilts further toward the firms that can co‑design energy and compute at continental scale.