Meta secures 6.6 GW nuclear power to fuel AI superclusters

Meta is effectively turning nuclear power into an AI input. By locking in up to 6.6 gigawatts of firm, low‑carbon electricity, it’s building a moat not just in model quality but in the physical capacity to train and serve them. This move directly addresses the emerging constraint that everyone from data‑center operators to regulators is worried about: there simply isn’t enough reliable power near where hyperscalers want to drop multi‑gigawatt clusters.

Strategically, the mix of legacy plants (Vistra) and next‑gen SMRs (TerraPower, Oklo) hedges timelines. Existing reactors cover Meta’s near‑term AI growth, while advanced units are a bet that nuclear manufacturing can scale just as AI workloads explode in the 2030s. It also turns Meta into a cornerstone customer for two of the most important Western advanced‑nuclear developers, giving it influence over how and where those reactors get deployed.

In the race to AGI, this underscores a hard reality: the winners may not be whoever has the cleverest architecture, but whoever can guarantee cheap, 24/7 megawatts at scale. Meta is clearly signaling that its AI roadmap assumes orders‑of‑magnitude more compute – and it’s rearranging parts of the U.S. power sector to make that plausible.