Google’s Nuclear Play: How One AI Giant Is Rebuilding the Power Grid It Depends On

When Google started signing long-term contracts for wind and solar more than a decade ago, it helped invent the modern corporate renewables market. Those early power-purchase agreements turned Big Tech from a passive utility customer into a force that could move entire segments of the grid.

Now the company is trying to do something far harder: repeat that trick with nuclear power—at the exact moment AI is blowing up its electricity demand.

In the dry language of a sustainability blog, Google framed the agreement as part of a “broad portfolio of advanced clean electricity technologies” that will complement its wind and solar purchases and help it reach 24/7 carbon-free energy and net-zero targets. Behind the scenes, it was something else: an admission that intermittent renewables alone can’t carry the weight of the AI era.

The Kairos agreement is explicit about the driver. The grid, Google argues, needs new sources of firm power to support AI systems powering scientific advances, business capabilities, and national competitiveness. In other words: large-language models and future AI systems need electricity that doesn’t disappear when the sun sets.

Kairos’s technology uses a molten-salt coolant circulating around ceramic pebble fuel, operating at low pressure and high temperature. That combination is designed to move heat efficiently to a steam turbine while keeping the reactor vessel under less mechanical stress, enabling a simpler, more compact plant layout. The company’s design philosophy leans on inherent and passive safety features.

To get there, Kairos is building through a sequence of hardware demonstrations culminating in a commercial plant. In Tennessee, it has already broken ground on Hermes, a non-power demonstration reactor that became the first non-light-water reactor in the U.S. in over 50 years to receive a construction permit. Google’s wager is that by acting as an anchor buyer for multiple units—an “orderbook”—it can help Kairos shift from prototype to product, reducing costs through repetition.

The U.S. Department of Energy estimates that deploying around 200 gigawatts of advanced nuclear capacity by 2050 could require 375,000 additional workers. Google’s messaging echoes those numbers, framing nuclear as both a clean-power solution and an economic development engine.

By 2025, the story evolved from a single vendor agreement to a broader nuclear strategy.

In August 2025, Google, Kairos, and the Tennessee Valley Authority announced that the first commercial reactor under the orderbook would be built in the Tennessee Valley, supplying power to Google data centres in Tennessee and Alabama.

In May 2025, Google agreed to support Elementl Power, providing development capital to prepare three U.S. sites for advanced reactors, each targeting at least 600 megawatts. The designs are still to be selected, but Google has secured the option to buy power once they are built.

In October 2025, Google and NextEra Energy announced a deal to restart the Duane Arnold Energy Center in Iowa, a 615‑megawatt nuclear plant shut down in 2020. If restored by 2029, it would become one of the first previously closed U.S. reactors to return to operation.

Meanwhile, Google continues pushing geothermal. Its enhanced geothermal system pilot with Fervo Energy scaled to roughly 25 times the original contracted capacity after the creation of a new “clean transition rate” with U.S. utilities.

The company’s energy disclosures show data-centre electricity use rising quickly as AI expands across search, cloud, and consumer products. Even with large efficiency gains, matching every hour of load with carbon-free power has become significantly harder.

From an investor’s perspective, Google’s long-term contracts trade near-term complexity for long-term certainty. Nuclear provides decades-long price stability and carbon-free baseload—aligned with regulatory pressure, customer expectations, and internal climate goals.

The risks are substantial. Advanced reactors have never been deployed commercially in the U.S. Supply chains are thin. Licensing is slow. And no modern-era U.S. nuclear plant that has shut down has yet successfully restarted.

But Google’s position is clear: if AI is the next industrial wave, then electricity is the new foundational infrastructure. Wind and solar remain essential, but they cannot carry continuous compute on their own. Nuclear—new and old—may be required to close the gap.

The commitments Google made in 2024 remain accurate: a world-first SMR deal, a 2030‑2035 deployment window, and up to 500 megawatts of advanced nuclear capacity. What has changed since then is scale. The company is now attaching that vision to real grids, real sites, and real reactors, testing whether Big Tech can do for nuclear what it once did for renewables.