Google and Constellation Energy have agreed to support 890 megawatts of additional nuclear capacity for the PJM Interconnection grid, using upgrades at existing reactors rather than a new build. Announced on October 6, the arrangement combines a 20-year power purchase agreement, a separate supply agreement for existing generation and a five-year technology alliance. The Google nuclear deal with Constellation is a useful marker of how large technology companies are beginning to connect their power needs to physical grid additions—not simply to electricity procurement.
The additional 890 MW is expected to come from uprates at 11 Constellation-owned reactor units across six sites in Illinois, Pennsylvania and New Jersey. An uprate is an increase in output from an operating reactor, achieved here through equipment and technology improvements such as work on turbines, steam generators and digital control systems. Constellation says the first uprate is expected in 2028; Google says the full 890 MW is expected before the end of 2032.
What the Google nuclear deal includes
The 20-year agreement is designed to provide revenue certainty for the reactor upgrades. The companies say the programme represents more than $4.3 billion of new Constellation investment and will add firm capacity to PJM, the regional grid operator serving parts of the Mid-Atlantic and Midwest. The releases describe the 890 MW as incremental capacity available across the grid, not as electricity earmarked for a single Google data centre.
That distinction matters. The announcement also includes a separate 15-year agreement covering 2,700 MW from Constellation’s existing PJM fleet. The two figures should not be collapsed into one headline number: 890 MW is the planned increase in capacity from the upgrades, while 2,700 MW concerns continued supply from operating assets. Together, they give Google a longer-term power arrangement while giving Constellation a basis to invest in output and keep existing generation in the market.
The companies have also expanded their technology relationship for five years. Constellation plans to use Google Cloud and Gemini Enterprise workflows in areas including site selection, power-flow modelling, permitting and interconnection planning, outage and turnaround work, and operational-technology security. Those are stated deployment plans, not evidence that AI has already delivered the promised capacity or cost outcomes. The meaningful test will be the pace and reliability of the actual uprates.
Why upgrading existing reactors changes the conversation
New nuclear projects can take years to permit, finance and construct. This deal takes a different route: it seeks more output from plants that already exist and are connected to the grid. That does not make the work frictionless, but it shifts the near-term question from where to build a reactor to whether targeted upgrades can be delivered on time, within the announced investment level and without disrupting reliable operations.
For technology companies, that approach is increasingly consequential. Google’s own account explicitly links the agreement to its data-centre growth, yet the capacity is framed as a grid-wide addition. That is a more substantive model than claiming clean-energy progress through certificates alone: the commercial agreement is tied to physical work intended to increase available generation. It is still not a guarantee that all local transmission or affordability pressures disappear. The releases do not disclose a project-by-project capacity allocation, a power price or a full regulatory timetable.
What to watch between now and 2032
Three milestones will show whether the deal is delivering more than a compelling announcement. First, the first uprate due in 2028 will test the schedule. Second, public detail on the individual units, approvals and realised output will reveal how the 890 MW total is being assembled. Third, the companies’ demand-response and load-shaping commitments will matter during periods when the PJM grid is under stress.
The broader lesson is not that AI has found a quick fix for electricity demand. It is that the infrastructure conversation is becoming more concrete. In this case, Google is pairing long-term offtake with upgrades to existing nuclear assets, while Constellation is pairing those upgrades with a more extensive digital-operations partnership. The result is a real, time-bound capacity plan that readers can evaluate against 2028 and 2032 delivery points.
Sources
Primary reporting is based on the joint Google and Constellation announcement, Constellation’s company release, and Google’s explanation of the nuclear-uprate programme. Utility Dive independently reported the agreement’s core terms and expected first 2028 uprate.
