Google needs more electricity for its data centers, and on Tuesday it found 890 megawatts without pouring a single new foundation. Under a 20-year agreement with Constellation Energy , Google will buy the added output from upgrades at 11 existing reactors in Illinois, Pennsylvania, and New Jersey, giving Constellation the certainty to invest more than $4.3 billion. The new capacity roughly equals one large new reactor, and it lands on PJM Interconnection, the grid that serves 67 million people across 13 states and Washington, D.C.

Colorado tells the other half of the story. Comanche 3, Xcel Energy’s largest coal unit in the state, returned to service only last month after a turbine failure kept it dark for 13 months. To cover the gap, regulators kept the 50-year-old Comanche 2 unit at the same site running past its scheduled retirement, and Xcel bought power from other utilities. And Colorado has no reactor to upgrade. Its only nuclear plant, Fort St. Vrain, shut down in 1989 .

The two examples illustrate the divide. Firm power, meaning electricity available around the clock regardless of weather, is becoming a geographic asset. States with nuclear fleets can stretch them, increasingly with tech companies paying the bill. States without them must wait for small modular reactors that are not yet running anywhere in the U.S., bet on newer carbon-free options such as advanced geothermal , or fall back on fossil fuels: aging coal plants kept alive past retirement and more natural gas. As AI demand and coal retirements converge, that split will shape where data centers, factories, and capital go.

The goal should be to steer that capital toward carbon-free power: wind and solar for volume and to meet escalating demand, and nuclear for the steady output that keeps a renewables-heavy grid reliable.

Constellation CEO Joe Dominguez said that his company’s deal with Google “can serve as a model for how technology companies and the energy industry can work together.” The question for policymakers is whether that model extends to states with no reactors to upgrade.

The Fastest Megawatt Is One You Already Own

An uprate is a retrofit—a government-approved increase in how much power an existing reactor may produce, achieved by upgrading turbines, steam generators, and controls at a plant that is already licensed, staffed, and connected to the grid. That largely sidesteps the slowest steps in building power: finding a site and waiting years in line to plug in. Constellation expects its first uprate by 2028 , and Google says all 890 megawatts will be online before the end of 2032.

The deal also tests PJM’s proposal that large new users bring their own power . “Our offtake agreements cover 100% of the cost of uprate construction at each facility,” Patrick Taylor, Google’s commercial lead for energy origination, told me.

It is Google’s largest uprate agreement, Taylor says, and part of a pattern. Google says it has now enabled more than 1.5 gigawatts of nuclear capacity through uprates and restarts. The Energy Department wants 5 gigawatts from uprates and restarts by 2029. Most of that is still in the pipeline; the bulk of uprate applications isn’t expected until 2027 and 2028 .

China has 36 reactors under construction that will add roughly 39 gigawatts, nearly half of all nuclear construction worldwide. Washington’s near-term goal is mostly about squeezing more from plants built a generation ago. That buys time, but it is not a strategy for keeping pace.

The Other Side of the Map

That option exists only in the 28 states with reactors . The other 22 have nothing to uprate, and they include nearly the entire interior West: Colorado, Utah, Nevada, New Mexico, Wyoming, Montana, and Idaho.

The North American Electric Reliability Corporation, the continent’s grid watchdog, rates the western grid’s Northwest subregion at high risk of power shortfalls within five years . The Rocky Mountain subregion that includes Colorado fares better for now, but NERC flags its aging fleet of fossil plants and projects shortfalls by the mid-2030s . John Moura, NERC’s director of reliability assessment and performance analysis, said in January that the grid is changing faster than the infrastructure needed to support it .

Colorado has built large amounts of wind and solar on its Eastern Plains. What it lacks is carbon-free power that runs when the wind and sun don’t.

Comanche 3 shows what that means. When the 750-megawatt unit failed, Xcel had no quick replacement. It has since asked to keep Comanche 2 , less than half Comanche 3’s size, running until March 2028. In a March filing, the utility told regulators that replacing Comanche 3’s capacity would cost billions and could not arrive before 2029. Advanced nuclear is a 2030s answer to a problem that has already arrived.

Until then, Colorado is leaning on coal it had planned to retire.

None of this makes uprates a cure. Each adds a modest slice: the 890 megawatts here is spread across 11 reactors, or roughly 80 megawatts apiece. The fleet is finite: regulators have approved 171 uprates since the 1970s and expect only about 30 more applications through 2030. Every deal shrinks the remaining headroom.

Google also signed a second, larger deal : it will buy up to 2,700 megawatts over 15 years from Constellation plants already operating in PJM. Taylor says that contract “is not asset specific and primarily serves as long-term revenue certainty for Constellation’s operating plants.” It keeps existing power flowing and may prevent retirements, but it adds no new supply.

Paying for the upgrades also doesn’t settle every cost question. All PJM customers pay for capacity, the grid’s guarantee that enough plants will be available at peak demand, and big new users can push that price up for everyone. Google expects the new megawatts to count against its own demand and to be offered into PJM’s capacity auctions at $0, adding supply without raising the price, Taylor says. But those rules are still pending before federal regulators, and PJM’s market monitor and state consumer advocates will want to test them.

Small modular reactors have their own record: NuScale’s first U.S. project was cancelled in 2023 after rising costs left too few utilities willing to buy its power. Skeptics argue that the West’s realistic bridge is natural gas.

Gas can fill gaps, and in the near term it will. But it ties buyers to volatile fuel prices and carbon emissions for decades; a long-term nuclear contract is the opposite bet.

Google’s own portfolio points to a better path. It buys uprates and restarts where reactors already run, and it is also anchoring new technology, backing Kairos Power’s planned advanced reactor at Oak Ridge, Tenn., through the Tennessee Valley Authority. In the West, where reactors are scarce, it has turned to geothermal, signing a 396-megawatt deal with Fervo Energy . “Which technology we pursue in a given region and market depends on a variety of factors,” Taylor says.

That is the lesson for states without reactors. “When Google serves as an anchor customer, our contract provides the certainty that projects require in order to stand up supply chains, invest in a skilled labor force, and secure the most efficient sources of capital,” Taylor says.

A creditworthy buyer signing a long-term contract is what absorbs first-of-a-kind risk, whether the asset is a new turbine in Illinois or a small reactor at a retired coal site in the West. Hyperscalers increasingly want carbon-free power. Non-nuclear states must provide a regulatory and siting path that makes the deal worth signing.

In Pueblo, Colorado, home to Comanche 3, local leaders have already made the case for the technology. Frances Koncilja, co-chair of an advisory committee Xcel created and a former Colorado public utilities commissioner, said in 2024 that only advanced nuclear “is going to make Pueblo whole after Comanche 3.”

In the AI era, firm power is joining land, water, and transmission as a deciding factor in where investment lands. States with reactors are already cashing in . States without them need more than a technology preference. They need a financing and regulatory plan that gets carbon-free firm power built before more fossil plants fill the gap.