Google’s 3.59GW Constellation deal turns nuclear uprates into AI infrastructure

Alphabet Inc. (Nasdaq: GOOGL, GOOG) subsidiary Google has entered a sweeping 3,590-megawatt power arrangement with Constellation Energy Corporation (Nasdaq: CEG), giving the technology group access to additional electricity while financing upgrades that will increase output from existing United States nuclear reactors. The agreement combines a 20-year arrangement supporting 890 MW of genuinely incremental nuclear capacity with a separate 15-year energy-supply agreement covering another 2,700 MW across the PJM Interconnection market. Constellation expects to invest more than $4.3 billion to deliver the nuclear expansion, demonstrating how artificial intelligence infrastructure spending is increasingly moving beyond servers and chips into generation assets themselves.

Constellation Energy shares surged more than 13% after the announcement, providing one of the clearest market signals yet that investors are attaching substantial value to utilities capable of supplying firm electricity into the artificial intelligence buildout. The nuclear operator plans upgrades across 11 generating units in Illinois, Pennsylvania and New Jersey, with the first incremental capacity expected to become available from 2028. Google will not remove the newly generated electricity from the public grid through a private physical connection; instead, the structure is designed to add supply to PJM while financially supporting the capacity needed for Google’s expanding data-centre load.

Why is Google paying to increase output from nuclear plants that already exist?

Building a new large nuclear reactor in the United States can require many years of permitting, engineering, financing and construction, which is badly mismatched with the speed at which technology companies want to deploy artificial intelligence data centres. Nuclear uprates offer a different route because operators improve turbines, generators, cooling systems and other equipment at plants that already possess operating licences, sites, grid interconnections and skilled workforces. The resulting power increases are smaller than constructing multiple new reactors, but they can reach the grid far more quickly while avoiding much of the first-of-a-kind development risk associated with entirely new nuclear projects.

Google and Constellation estimate that the planned investments can unlock 890 MW of additional generation, an amount broadly comparable with the output of a large new reactor. The first uprate is expected in 2028, which places new capacity much closer to Google’s immediate artificial intelligence infrastructure timetable than most greenfield nuclear alternatives. That sequencing is strategically important because data-centre developers increasingly report that electricity availability, rather than availability of land or computing equipment, is becoming the factor determining when projects can actually operate.

The economics also extend beyond Google. The additional nuclear generation is intended to enter the PJM market rather than being dedicated exclusively behind the meter to individual Google facilities, meaning other users should continue to benefit from a larger pool of available electricity. Google argues that this model allows it to support new supply associated with its own growth while reducing the risk that residential ratepayers are effectively asked to finance the infrastructure required by enormous data-centre loads.

What is the difference between the 890MW nuclear deal and the extra 2.7GW agreement?

The distinction is important because describing the entire 3.59 GW as new nuclear capacity would materially overstate the agreement. Only 890 MW represents incremental nuclear output unlocked through upgrades to Constellation’s existing fleet, while the additional 2,700 MW comes through a separate 15-year energy-supply arrangement involving existing resources in the PJM system. The larger figure therefore combines new capacity with long-term supply support rather than describing 3.59 GW of newly constructed generation.

That 2.7 GW agreement still matters strategically because long-duration contracts can improve the economics and operating certainty of power plants serving a market confronting rapidly rising demand. Data centres increasingly need both additional generation and confidence that existing dispatchable plants remain economically viable during the infrastructure buildout. Google is consequently financing two sides of the same power problem: increasing physical output from established nuclear assets while supporting continued availability of existing capacity across the grid.

The structure illustrates how corporate power procurement is becoming more sophisticated. Technology companies historically purchased renewable-energy credits or long-term wind and solar contracts to offset electricity use, but artificial intelligence computing requires enormous amounts of power around the clock. Intermittent renewable generation remains an important part of the supply mix, yet data-centre operators increasingly need firm generation, energy storage, demand response and flexible loads to work together if they are going to expand without worsening grid reliability.

Why has PJM become ground zero for the AI electricity problem?

PJM operates the largest competitive electricity market in the United States, covering all or parts of 13 states and the District of Columbia. The region also contains some of the country’s most important data-centre clusters, particularly northern Virginia, where hyperscalers and cloud providers have concentrated enormous computing capacity. That combination has produced a difficult mismatch between rapidly rising electricity demand and the pace at which new generation and transmission can be connected.

Google’s agreement effectively adopts a “bring your own power” philosophy. Instead of assuming that utilities and grid operators will eventually build enough supply for every proposed data centre, hyperscalers are increasingly financing or contracting directly with companies capable of adding generation. This does not remove the role of regulated utilities, transmission developers or regional grid planners, but it moves part of the capital burden toward the technology companies creating much of the incremental demand.

The trend has implications well beyond Google. Amazon.com, Inc., Microsoft Corporation, Meta Platforms, Inc. and other hyperscalers have entered nuclear, renewable and gas-related power arrangements as they pursue multi-gigawatt computing pipelines. The power market is consequently developing into another competitive front in artificial intelligence, where a company with chips but no electricity may be unable to translate theoretical computing capacity into commercially usable infrastructure.

Could nuclear uprates become faster and cheaper than building new small modular reactors?

For the immediate AI boom, existing-reactor upgrades offer an important timing advantage. Small modular reactors promise factory-based construction and potentially lower project complexity, but most designs remain at earlier commercial stages and still need licensing, financing, supply chains and operating experience. Uprates, by contrast, modify plants already producing electricity, allowing investment to target specific engineering improvements rather than recreating an entire nuclear site.

The opportunity is nevertheless finite. Existing plants can only increase output so much before physical and regulatory limits become binding, meaning uprates cannot alone satisfy the enormous long-term load projections associated with artificial intelligence and electrification. The most plausible future system therefore combines nuclear extensions and uprates, renewable generation, battery storage, gas generation, transmission upgrades and eventually new reactors rather than relying on one technology to solve the problem.

Constellation is unusually well positioned because it operates the largest nuclear fleet in the United States. Its ability to extract incremental megawatts from existing plants creates an asset that did not command the same strategic premium before electricity demand began accelerating. The sharp share-price reaction suggests investors increasingly view nuclear operators not merely as defensive utilities but as scarce infrastructure providers participating directly in the artificial intelligence capital cycle.

Why is Google also giving Constellation access to Gemini Enterprise?

The companies are extending the relationship beyond electricity by creating what they describe as an “AI for Energy” blueprint using Google Cloud and Gemini Enterprise. Constellation plans to apply the technology to plant operations, capacity delivery, asset dispatch and infrastructure protection, creating an unusual feedback loop in which artificial intelligence both drives greater electricity consumption and is then used to improve the efficiency of the energy system supporting it. The commercial importance will depend on whether those tools actually lower operating costs, accelerate maintenance or improve output rather than merely adding another software layer.

Nuclear facilities produce enormous quantities of operational, engineering and maintenance data, making them plausible environments for advanced analytics. Artificial intelligence could help identify equipment degradation, improve outage planning and organise complex technical information, although safety-critical decisions will continue to operate inside tightly regulated processes. Google therefore gains an opportunity to demonstrate enterprise artificial intelligence in one of the most demanding industrial environments while Constellation gains technology intended to help operate an increasingly valuable fleet.

The partnership also provides a counterpoint to the simplistic argument that AI infrastructure consists only of semiconductor demand. Each additional cluster requires electricity, transmission, cooling, buildings and increasingly sophisticated financial arrangements around those assets. The biggest beneficiaries of the artificial intelligence boom may consequently include industrial companies that never manufacture a processor but control something the processors cannot operate without.

What does Constellation Energy’s share-price surge say about investor sentiment?

Constellation Energy rose more than 13% following the announcement, substantially outperforming the broader market and reflecting the size and duration of the new commercial relationship. Investors appear to be rewarding two characteristics simultaneously: tangible incremental generation from the nuclear uprates and long-term contractual support from one of the world’s largest technology companies. Those factors strengthen revenue visibility at a time when electricity demand projections are being revised sharply higher.

The reaction should not be interpreted as the present value of the Google contract alone because Constellation’s equity valuation reflects its entire generation portfolio, future transactions and broader electricity-market conditions. The agreement nevertheless creates an important benchmark demonstrating what scarce firm generation may be worth when hyperscale customers are willing to finance capacity directly. Future nuclear power agreements will increasingly be compared against the duration, capacity and investment commitments embedded in the Google arrangement.

The most important conclusion is that Google is no longer treating electricity simply as an operating expense purchased after a data centre has been designed. Power supply is becoming part of data-centre strategy itself, with technology companies participating earlier in generation investment and accepting longer commitments to ensure capacity arrives when computing infrastructure needs it. The artificial intelligence infrastructure race has moved from who owns the best chips to who can assemble chips, land, cooling, networks and electricity into a functioning system, and Constellation’s existing reactors have suddenly become one of the most valuable shortcuts in that equation.