On Tuesday, October 6, Google and Constellation Energy announced the largest nuclear uprate agreement ever signed between a technology company and a nuclear operator: a 20-year power purchase agreement backing 890 megawatts of additional generating capacity at eleven existing Constellation reactors. Constellation will invest more than $4.3 billion to squeeze that capacity out of plants that are already licensed, already staffed, and already connected to the grid. The first additional megawatts are expected in 2028, with the full 890 online by the end of 2032.
The structure is worth understanding precisely, because it is really two deals in one. The headline contract is the 20-year PPA covering the 890 megawatts of new capacity created by the uprates. Alongside it sits a separate 15-year energy supply agreement covering 2,700 megawatts from Constellation’s existing nuclear fleet in PJM — not new capacity, but revenue certainty that keeps those operating reactors economically viable and serving the market. Together, the package totals roughly 3,590 megawatts, though only about a quarter of it is newly created power.
The plants
The uprates span six sites across three states: the Braidwood, Byron, LaSalle, and Quad Cities plants in Illinois; the Limerick plant in Pennsylvania; and the Salem plant in New Jersey, which Constellation partly owns. Eleven units in total. Constellation says the work involves replacing or upgrading turbines, steam generators, and digital control systems — the heavy, non-nuclear hardware that determines how much electricity a reactor’s heat can actually produce. The program is expected to sustain roughly 4,400 existing jobs and create about 7,200 construction jobs across the host communities.
For readers keeping score on the hyperscaler-nuclear scorecard, this is the second such Constellation deal in about a week. Days earlier, Constellation signed a 20-year agreement with Amazon covering 690 megawatts from the 1,790-megawatt Calvert Cliffs plant in Maryland — including an approximately 190-megawatt uprate expected online between 2030 and 2032, backed by more than $3 billion of investment and supporting the plant’s relicensing for another 20 years. Combined, the Google and Amazon agreements put about 1,080 megawatts of planned Constellation uprates under long-term tech-company contracts. Uprates are becoming a product Constellation can sell on repeat.
What an uprate actually is
A power uprate increases the licensed output of an existing reactor without building a new one. The Nuclear Regulatory Commission recognizes three categories. Measurement uncertainty recapture uprates add less than 2 percent by installing more precise instruments — typically ultrasonic feedwater flow meters — that let operators safely claim thermal margin previously held back as measurement error. Stretch uprates, in the 2 to 7 percent range, adjust instrumentation and operating settings to use the plant’s original design margins, without major modifications. Extended power uprates go further — the NRC has approved increases as high as 20 percent — and require significant replacement of major non-nuclear equipment: high-pressure turbines, condensate pumps and motors, main generators, transformers.
The Google-backed program is squarely in that third category. New turbines, new steam generators, new digital controls across eleven units is the textbook definition of an extended power uprate campaign. Each one requires a license amendment, with the NRC’s technical staff reviewing the licensee’s safety analyses before approval. That regulatory step is real schedule risk — but it is a fundamentally different kind of risk than licensing a new reactor, because the site, the containment, and the operating history already exist.
Why uprates beat new builds
For a hyperscaler, the comparison that matters is uprates versus the alternatives for firm, carbon-free power. Start with speed. The first Google-backed uprate megawatts are expected in 2028 — roughly two years from announcement — with the program completing by the end of 2032. A new large reactor, by contrast, faces 6.5 to 8 years of construction alone on current AP1000 estimates, before site selection, licensing, and interconnection are even counted. As we detailed in our piece on bringing your own power, small modular reactors remain a better-part-of-a-decade proposition from agreement to operation. Uprates skip the site hunt, the greenfield licensing, and the generator interconnection queue entirely, because the plant is already there.
Then there is cost. Dividing Constellation’s disclosed $4.3 billion investment by 890 megawatts of incremental capacity implies roughly $4,800 per kilowatt of new capacity. That sits at the high end of historical uprate costs — Idaho National Laboratory’s 2025 assessment puts extended power uprates in the $1,500 to $2,500 per kilowatt range, with very large modifications reaching up to $5,000 — but it is roughly half the cost of new nuclear construction, where recent AP1000 estimates run $8,300 to $10,375 per kilowatt and Vogtle Units 3 and 4 came in around $11,000. The cheap uprates, it is worth noting, are mostly gone: 73 measurement-uncertainty-recapture projects and 65 stretch uprates have been completed since the 1970s, and the industry long ago harvested that low-hanging fruit. What remains is the harder, more expensive extended-uprate work — which is exactly what Google is now financing.
The product Google is actually buying
To understand why Google would anchor a $4.3 billion nuclear construction program, return to the load profile. As our load growth explainer laid out, AI data centers draw power nearly constantly — 70 to 90 percent of contracted capacity, around the clock. That flat, always-on demand is the worst possible match for wind and solar PPAs, whose output follows the weather rather than the servers. It is also increasingly awkward to serve with natural gas, which collides with corporate climate commitments and, in many jurisdictions, with permitting reality.
The product Google needs is firm, carbon-free, 24/7 electricity — widely considered the hardest product to procure in a decarbonizing grid. Existing nuclear is the only source that delivers all three attributes at scale today. An uprate PPA doesn’t just buy electrons; it buys additional firm capacity that would not otherwise exist, with a carbon profile of zero and a delivery timeline measured in years, not decades. For a company racing to deploy AI infrastructure, that combination is worth a 20-year commitment. (The companies did not disclose the PPA pricing, so the per-megawatt-hour economics of the deal itself can’t be evaluated from public information — the $4,800-per-kilowatt figure above describes Constellation’s capital investment, not the contract price Google will pay.)
Who pays — and who doesn’t
The most politically significant feature of the deal may be its cost-allocation design. A Google spokesperson said the company is covering Constellation’s costs to invest in the new capacity, make necessary transmission upgrades, and manage ongoing operations — in other words, Google pays the full cost of the new generation its demand requires, rather than socializing it across the rate base. That framing is deliberate. Across the country, utilities and regulators are fighting over who pays for the grid infrastructure that data center load growth demands, and residential ratepayers are the constituency everyone claims to protect.
The deal is explicitly positioned against that backdrop: the companies aligned it with the White House’s “Ratepayer Protection Pledge” and with PJM’s “bring your own power” framework for large load growth, under which big new loads are expected to bring their own supply rather than lean on the system. Whether or not one finds the pledge’s politics persuasive, the commercial logic is clean. By financing the capacity itself, Google preempts the cost-shift fight entirely — and buys goodwill with the regulators and legislators who will decide the fate of its future interconnection requests.
A complement to the fence, not a substitute
Readers of our bring-your-own-power piece will notice something important: this deal is the opposite of defecting from the grid. Google is not building turbines at the data center fence or islanding its load. It is financing upgrades to shared, grid-connected generation inside PJM — capacity that strengthens the regional system serving 67 million people, not just Google’s servers. The electrons aren’t even necessarily delivered to Google’s data centers; what Google buys is the PPA, the clean-energy claim, and the knowledge that its load growth is matched by new firm supply on the same grid.
That distinction matters for the grid-planning questions we raised earlier. Behind-the-meter self-generation raises cost-shift and duplication concerns precisely because it can bypass the shared system. A grid-connected uprate PPA does the reverse: it adds supply to the shared system, funded by the large customer, with the benefits — reliability, emissions, jobs — spread across the market. If bring-your-own-power is the data center industry learning to think like a power developer, this deal is the industry learning to act like a responsible grid citizen instead.
The hourly-matching angle
There is a second, quieter reason nuclear fits Google’s procurement strategy better than anything else on the menu: hourly matching. As we covered in our piece on data center power deals, Google has committed to operating on 24/7 carbon-free energy — matching every hour of consumption with carbon-free generation on the same grid, not just buying enough annual renewable credits to cover the yearly total. Wind and solar PPAs, even generous ones, leave the nighttime and calm hours exposed; filling those gaps with storage or overbuilding adds cost and complexity without fully solving the shape problem.
An uprate PPA is, in effect, the purchase of 8,760 matched hours. Nuclear runs at capacity factors above 90 percent, producing through every night, every calm spell, every winter peak. For a buyer with an hourly-matching commitment and a flat load profile, that is the closest thing the market offers to a perfect hedge — one contract that simultaneously satisfies the sustainability ledger and the operations team. It is no accident that the companies most aggressive on 24/7 carbon-free goals are the ones writing the biggest nuclear checks.
The runway beyond this deal
How much more of this is available? More than most people assume. Idaho National Laboratory’s 2025 assessment of uprate opportunities estimated untapped potential on the order of 1 to 2 gigawatts across the boiling-water-reactor fleet and 3 to 4 gigawatts across pressurized-water reactors — several large reactors’ worth of capacity, achievable without pouring a single new foundation. The federal UPRISE program, aimed at increasing output from existing nuclear infrastructure, is pushing in the same direction. The Google and Amazon deals together capture roughly a fifth of that estimated runway, which suggests the uprate market is just getting started rather than nearly exhausted. And Constellation is not alone in the queue: the NRC’s published list of expected uprate applications shows Duke Energy preparing extended power uprates at its McGuire units, PSEG filing stretch uprates for both Salem units, and Southern Nuclear pursuing extended uprates at Hatch and Vogtle — all within the next few years. The pipeline of licensable uprate projects is real, multi-operator, and already moving through the regulatory system.
What to watch
Three things will determine whether this deal becomes a template or a one-off. First, execution: extended power uprates are complex construction projects inside operating nuclear plants, and the 2028 first-delivery date is the proof point the whole thesis rests on. NRC license amendment reviews and refueling-outage scheduling will set the real pace.
Second, replication: Constellation has now sold the same basic structure to two hyperscalers in a week, and every other large nuclear operator is watching. If Microsoft, Meta, and Oracle follow with their own uprate PPAs, the existing fleet becomes a meaningful source of new firm capacity this decade. Third, the policy environment: PJM’s evolving framework for large loads will shape whether deals like this one remain voluntary corporate leadership or become the regulatory expectation for any data center above a certain size.
The deeper significance is the emergence of the hyperscaler as anchor financier of clean firm capacity. A 20-year PPA from a creditworthy buyer is what makes a $4.3 billion capital program bankable; without Google’s commitment, Constellation’s chief executive acknowledged, the company had known the uprates were possible for years but couldn’t justify the investment. That is project finance logic applied to the existing nuclear fleet — and it suggests the fastest new clean megawatts of the next decade may come not from technologies we haven’t built yet, but from reactors we’ve been running for forty years.