Buying megawatts from plants that already exist
Google's latest power deal with Constellation is notable less for its size than for its method. Instead of funding new reactor construction, which routinely takes a decade or more and faces well-documented cost overruns, Google is paying to extract more capacity from 11 reactors that are already operating. The 20-year power purchase agreement backs more than 4.3 billion dollars in Constellation investment to upgrade turbines, steam generators, and digital control systems across plants in Illinois, Pennsylvania, and New Jersey, adding a combined 890 MW without a single new reactor vessel.
Constellation says the first uprate will be delivered by 2028, a timeline that would be almost unthinkable for new nuclear construction in the United States but is realistic for incremental equipment upgrades at plants with decades of existing operating history and regulatory familiarity. That speed is precisely the point: Google needs firm, round-the-clock power for AI data centers now, not in the mid-2030s, and uprating existing capacity is the only nuclear option that plausibly delivers on that timeline.
Three agreements, one underlying bet
The uprate deal is only one piece of a broader arrangement between the two companies. A separate 15-year energy supply agreement covers 2,700 MW of capacity in the PJM region, Google's largest such commitment to date, while a five-year technology agreement has Constellation adopting Google Cloud and AI services across its own operations. Structuring the relationship this way, power purchase, broader energy supply, and a cloud and AI services commitment, ties Constellation's technology modernization directly to the same company funding its generation upgrades.
That reciprocal structure is becoming a recognizable pattern among hyperscalers negotiating power deals: utilities and generators increasingly receive not just revenue certainty from long-term power contracts but also a commercial relationship as a cloud and AI customer. For Constellation, adopting Google's AI and cloud tools is a comparatively small commitment next to a 4.3 billion dollar capital program, but it is a tell that these negotiations now routinely bundle power supply with technology sales in both directions.
Why PJM specifically, and why now
The PJM Interconnection, covering much of the Mid-Atlantic and parts of the Midwest, has become ground zero for the collision between AI data center demand and grid capacity. It is home to Northern Virginia, the largest data center market in the world, and has already seen capacity prices spike as a result of surging large-load requests from hyperscalers. Google's decision to concentrate its uprate investment in exactly this grid region is a direct response to where its own capacity constraints and those of its data center neighbors are most acute.
Both companies frame the deal as compatible with the current US policy environment, citing alignment with the administration's Ratepayer Protection Pledge and the Department of Energy's UPRISE program, and stressing that the added output does not pass costs to other ratepayers. Whether that claim holds up under scrutiny from state utility regulators remains to be tested, but the political framing itself signals how carefully hyperscalers are now managing the optics of competing with residential and commercial customers for scarce grid capacity.
A pattern other hyperscalers are already following
Google is not alone in reaching for existing nuclear capacity rather than waiting on new construction. Microsoft, Amazon, and Meta have each pursued some combination of uprates, life-extension deals, and small modular reactor investments over the past two years, all converging on the same conclusion: the fastest nuclear megawatt available to a hyperscaler today is one that already exists inside an operating reactor, not one still on a drawing board. That convergence is turning uprate capacity at operating plants into a genuinely competitive resource among the handful of companies with the balance sheet to fund it.
Operators like Constellation, which controls the largest fleet of US nuclear plants, are consequently gaining real negotiating leverage, since they are choosing which hyperscaler's capital to accept for which reactor's upgrade program. Google's agreement to also commit Constellation to using Google Cloud and AI services should be read partly as a bid to lock in that relationship before a competing hyperscaler courts the same operator for its next available uprate slot.
What this signals for enterprise power strategy
For any enterprise technology leader whose AI roadmap depends on cloud GPU capacity, this deal is a useful signal of where that capacity bottleneck actually sits. It is not primarily about chip supply anymore, as Nvidia's shipment volumes have scaled dramatically over the past two years. It is about whether a hyperscaler can secure firm power fast enough to energize the data centers those chips sit in, and Google's uprate strategy shows the company treating that as urgent enough to fund billions in upgrades to someone else's power plants rather than simply bidding for more capacity on the open market.
Enterprises negotiating long-term capacity commitments with any hyperscaler should ask directly how that provider is securing the power behind its roadmap promises, not just its chip allocation commitments. A cloud provider with a credible, contracted power strategy like Google's uprate deal is less likely to face the delivery delays that have already hit data center buildouts elsewhere in the industry when grid interconnection queues or local power constraints intervened unexpectedly.
