MOUNTAIN VIEW, California — Google is making one of its biggest electricity bets yet, locking in 3.59 gigawatts of power from Constellation Energy as artificial intelligence pushes U.S. data centers toward an increasingly uncomfortable reality: the next great AI shortage may not be chips. It may be electricity.
The agreement, announced October 6, includes a 20-year power purchase deal that will add 890 megawatts of new nuclear capacity across Constellation’s existing reactor fleet in Illinois, Pennsylvania and New Jersey.
That is roughly equivalent to adding the output of an entirely new large nuclear reactor to the grid.
A separate 15-year agreement covers another 2,700 megawatts of electricity, bringing the overall power arrangement to approximately 3,590 MW.
The scale is extraordinary.
But the bigger story is what Google is really buying.
It is not simply clean electricity.
It is reliable, round-the-clock power for an AI economy whose energy consumption is rising faster than parts of the U.S. grid can comfortably handle.
Google Is Paying to Squeeze More Power From Existing Nuclear Plants
Rather than waiting years for entirely new nuclear reactors to be built, Google and Constellation are pursuing a faster route.
Constellation will upgrade 11 nuclear units at six operating plants across Illinois, Pennsylvania and New Jersey.
The work will involve new equipment and efficiency improvements designed to increase thermal and electrical output — a process generally known as a nuclear uprate.
Those upgrades are expected to unlock 890 MW of additional capacity.
Constellation and Google emphasize that this is genuinely new generation, not merely electricity that already existed and is being reassigned to Google.
That distinction matters because U.S. utilities are struggling to add enough new supply to keep pace with rapidly rising data-center demand.
Building a large new reactor can take many years.
Improving existing plants can bring extra electricity onto the grid far sooner.
The Deal Runs for 20 Years
The nuclear portion of the agreement stretches for 20 years, giving Constellation long-term revenue visibility and Google greater certainty over future power availability.
Financial terms were not fully disclosed.
But The Verge reported the nuclear component could be worth more than $4.3 billion over its lifetime.
The arrangement is focused on the PJM Interconnection, the largest electricity market in the United States.
PJM serves all or parts of 13 states plus Washington, D.C., including some of America’s most important data-center regions.
It has also become one of the clearest examples of the pressure AI is placing on electricity infrastructure.
Data-center projects are arriving faster than transmission lines and power plants can be constructed.
That has raised concerns about:
higher electricity prices,
grid reliability,
long connection queues,
and whether planned AI data centers can actually receive enough power to operate.
AI Has Turned Nuclear Power Into a Technology Trade
A few years ago, nuclear utilities were hardly obvious AI stocks.
That has changed dramatically.
Constellation shares surged by roughly 12% to 15% on October 6 following the Google announcement, making the company one of the strongest performers in the S&P 500.
The broader utilities sector rose around 2.5%, its best session in months.
That reaction reveals how investors increasingly view electricity companies.
The AI investment boom began with semiconductors such as Nvidia.
Then it expanded into networking, memory and optical communications.
Now Wall Street is moving one layer deeper:
Who will generate the electricity required to run all of it?
Nuclear power has become particularly attractive because it provides steady output regardless of weather.
Unlike solar and wind, nuclear plants can run continuously.
That makes them appealing to hyperscale data centers that require power 24 hours a day.
Google’s Electricity Demand Is Already Surging
Google has good reason to secure power now.
The company’s electricity consumption rose roughly 37% last year, according to reporting on its sustainability data.
Its carbon emissions also increased approximately 18%, underscoring the difficulty technology companies face in expanding AI infrastructure while meeting climate targets.
Training and operating modern AI models require huge computing clusters.
Those computers consume electricity not only for processors but also for:
cooling,
networking,
storage,
backup systems,
and power conversion.
The result is that AI data centers can demand hundreds of megawatts individually.
A cluster of large projects can consume as much electricity as a small city.
Google Has Now Enabled More Than 1.5 GW of New Nuclear Capacity
The Constellation agreement is not Google’s first nuclear move.
Google says its recent nuclear deals have now helped enable more than 1.5 gigawatts of additional U.S. nuclear capacity through reactor uprates and restarts.
That strategy reflects an important shift in Silicon Valley.
Large technology companies once focused overwhelmingly on purchasing renewable electricity from wind and solar farms.
They are now increasingly seeking firm power — electricity that can be delivered continuously rather than only when weather conditions cooperate.
That has pushed companies toward:
nuclear,
natural gas with carbon controls,
geothermal,
long-duration storage,
and other technologies capable of supporting 24-hour operations.
Google Is Also Bringing Gemini Into the Power Plant
The partnership goes beyond electricity purchasing.
Constellation plans to use Google Cloud and Gemini Enterprise to create what the companies call an “AI for Energy” blueprint.
The idea is to use artificial intelligence to improve nuclear-plant and grid operations.
Potential applications include:
predictive maintenance,
operational optimization,
data analysis,
equipment monitoring,
and reducing infrastructure costs.
Google and Constellation say the goal is to make energy systems more efficient and resilient.
That creates an unusual circular relationship.
AI needs enormous amounts of electricity.
Now AI itself may be used to make the power plants supplying that electricity more efficient.
Constellation Gets a Powerful Long-Term Customer
For Constellation, Google is exactly the type of customer nuclear operators want.
Large technology companies have enormous balance sheets.
They need power for decades.
And they are willing to sign long-term agreements to guarantee supply.
Those contracts can justify capital investments that utilities might otherwise hesitate to make.
Google’s commitment allows Constellation to upgrade plants knowing that a major buyer will be there for years.
The deal is also expected to support approximately 4,400 existing jobs and around 7,200 construction jobs, according to Google.
The 890 MW Is Only Part of the Story
Some coverage has focused entirely on the nuclear expansion.
But Reuters reports that the full arrangement is much larger.
Google has contracted for 3,590 MW in total from Constellation in the PJM region.
About one-quarter of that supply comes from the new nuclear capacity.
The rest comes through a separate 15-year power agreement.
That means this is not simply a nuclear announcement.
It is a massive power-procurement strategy.
Google is effectively securing enough capacity to support an enormous expansion of its computing footprint.
Why Not Just Build New Nuclear Plants?
Time is the problem.
A new large nuclear plant can take a decade or more to permit and construct.
AI demand is growing much faster than that.
Google and Constellation therefore chose to increase output from reactors that already exist.
Uprating can add hundreds of megawatts while avoiding some of the permitting, financing and construction challenges associated with entirely new facilities.
That doesn’t mean new nuclear has disappeared from the equation.
Technology companies are also investing in next-generation reactor designs and small modular reactors.
But existing plants offer something particularly valuable today:
speed.
The U.S. Grid Is Becoming AI’s Weakest Link
Semiconductor supply dominated the first phase of the AI boom.
Power supply could dominate the next.
Data centers cannot simply be built wherever land is available.
They need:
massive electrical connections,
substations,
transmission lines,
backup generation,
water or alternative cooling infrastructure,
and reliable power around the clock.
Those constraints are becoming severe enough that Wall Street analysts are warning of delays to data-center projects.
Morgan Stanley recently warned that insufficient electricity could slow parts of the AI semiconductor supply chain because data-center deployments may be postponed even if chips are available.
That means Nvidia can manufacture the GPU.
TSMC can fabricate it.
Lumentum can provide optical networking.
But none of that matters if there is nowhere with enough electricity to plug the servers in.
The AI Boom Is Transforming Utility Valuations
That realization has radically changed investor attitudes toward power producers.
Utilities were traditionally viewed as slow-growth defensive investments.
Now some are being priced like AI infrastructure companies.
Constellation is one of the biggest examples.
The company owns the largest nuclear fleet in the United States.
As technology companies compete for reliable low-carbon electricity, those plants have become strategically valuable assets.
Other nuclear and independent power producers have benefited from similar enthusiasm.
Utilities connected to data-center regions have also seen their growth expectations rise substantially.
In practical terms, Wall Street has discovered that megawatts may become as strategically important as semiconductors.
Google Isn’t Alone
Nearly every major technology company is pursuing long-term power strategies.
Microsoft has backed the restart of a U.S. nuclear reactor.
Amazon has made nuclear-related power investments.
Meta has signed long-term clean-power agreements.
Google has invested in several advanced nuclear and energy technologies.
Their motivations are similar.
AI spending is moving from tens of billions of dollars toward hundreds of billions.
Electricity infrastructure must expand with it.
The technology industry cannot build AI capacity faster than the power system that supports it.
Electricity Prices Could Become Politically Sensitive
The downside is that AI data centers are competing for electricity with ordinary households and businesses.
If demand rises faster than generation and transmission capacity, prices can increase.
That has already become politically sensitive in regions with heavy data-center development.
Critics argue technology companies should pay for the infrastructure required to serve them rather than forcing residential customers to absorb those costs.
Google’s agreement attempts to address that concern by financing new capacity, rather than simply contracting for existing supply.
That is one reason the companies emphasized that the 890 MW is incremental.
If tech companies can finance additional generation, they may reduce the risk that AI expansion simply reshuffles electricity away from everyone else.
Nuclear Power Is Having an Unexpected Comeback
Perhaps the biggest long-term implication is the revival of nuclear energy.
For years, nuclear plants struggled against cheap natural gas and subsidized renewable power.
Some reactors were closed because operators said they were uneconomic.
Now the economics are changing.
Data centers value qualities nuclear plants already provide:
high capacity factors,
round-the-clock output,
low operational carbon emissions,
and large blocks of power.
The AI boom has therefore given nuclear power something it had lacked for years:
customers willing to pay for reliability over decades.
The Next AI Arms Race May Be Measured in Gigawatts
Google’s Constellation agreement captures how quickly the AI competition is changing.
The first question was:
Who has the best model?
Then:
Who has the most GPUs?
Now another question is becoming just as important:
Who has enough electricity to run them?
Google has answered part of that question with one of the biggest long-term power deals yet announced by a technology company.
A total of 3.59 GW of contracted supply.
890 MW of brand-new nuclear capacity.
A 20-year nuclear agreement.
And six existing nuclear sites being upgraded rather than waiting for entirely new reactors to be built.
For Constellation, the deal turns nuclear assets into AI infrastructure.
For Google, it secures a crucial resource before the power shortage becomes even more severe.
And for the broader technology industry, it sends a warning.
The race for AI dominance is no longer only about who can buy the most advanced chips.
It is increasingly about who can secure enough power to switch them on.
