Google Locks In 3.6 GW With Constellation to Power Its AI Buildout
A single power agreement covering 3.6 gigawatts—roughly a quarter of it nuclear—signals how seriously hyperscalers are treating electricity as a hard constraint on AI growth.
Electricity is no longer a procurement afterthought. It is the binding constraint on how fast AI infrastructure can scale—and Google just moved to secure a large block of it.
On October 6, 2026, Google signed a long-term power agreement with Constellation Energy covering 3.6 gigawatts of capacity. The deal is explicitly aimed at supporting Google's expanding AI and data-center footprint. It is one of the largest individual power contracts a hyperscaler has disclosed, and it arrives at a moment when the industry is quietly reorganizing itself around energy access as much as compute access.
What 3.6 GW Actually Means
To put the number in context: 3.6 gigawatts is not a rounding error. That is the kind of capacity figure associated with regional power grids, not individual corporate supply chains. For a single company to contract that volume in one agreement reflects just how dramatically AI workloads have changed the demand curve for electricity.
Approximately 25% of the contracted capacity is expected to come from new nuclear generation. That fraction matters. New nuclear—as opposed to existing capacity—implies construction timelines, regulatory processes, and capital commitments that extend well beyond a typical procurement cycle. Constellation committing to bring new generation online for this deal is a signal that the supply side is treating hyperscaler demand as durable enough to justify long-horizon bets.
The remaining roughly 75% of capacity comes from other sources, though the specific generation mix beyond the nuclear component was not detailed in reporting on the agreement.
The Long-Term Contract as Infrastructure Strategy
The Google-Constellation deal is not an isolated transaction. It reflects a broader pattern: hyperscalers using long-term power purchase agreements to effectively pre-buy electricity capacity before the infrastructure to generate it fully exists. This is a meaningful strategic shift.
Historically, large technology companies negotiated power deals primarily to hit sustainability targets—locking in renewable energy certificates or matching consumption with clean generation on an annual basis. What's happening now is different. These contracts are being structured to guarantee physical delivery of electrons at scale, at specific locations, over extended periods. The distinction matters for grid planning, for energy developers' financing, and for the competitive dynamics between hyperscalers themselves.
When one major player secures 3.6 gigawatts, that capacity is no longer available to competitors or to the broader market. Power, like compute, is becoming a resource to be locked up early.
Separate from the Larger Alliance
It is worth being precise about what this deal is and is not. The Google-Constellation agreement is a distinct development from Google's previously reported 100-gigawatt alliance involving Nvidia and Anthropic. The two efforts operate on different scales and serve different structural purposes—one is a bilateral supply contract, the other a broader industry alignment around AI infrastructure capacity.
But read together, they describe the same underlying pressure: AI workloads are scaling faster than existing energy infrastructure anticipated, and the companies closest to those workloads are taking aggressive action to get ahead of the constraint. The 3.6 GW Constellation deal is Google managing a known bottleneck at the operational level. The larger alliance is an attempt to reshape the infrastructure landscape at the industry level.
Neither initiative would exist if electricity were abundant and cheap relative to demand. The fact that both are happening simultaneously is the tell.
The Bigger Shift
What the Google-Constellation agreement ultimately represents is the moment when AI infrastructure strategy became inseparable from energy infrastructure strategy. Founders and operators building on top of hyperscaler platforms should understand what that means downstream: the cost, availability, and geographic distribution of compute is increasingly determined not by chip supply chains or data-center construction timelines, but by who controls large blocks of power generation capacity—and whether that capacity includes firmer, always-on sources like nuclear.
The companies that secure energy now are the ones that can build without interruption later. That dynamic is reshaping competitive advantage in ways that have nothing to do with model architecture or training algorithms—and everything to do with who signed the right contract in 2026.