US AI Data Centers Have Completely Lost It: They Want to Power AI With Retired Navy Nuclear Reactors

US AI Data Centers Have Completely Lost It: They Want to Power AI With Retired Navy Nuclear Reactors

AI’s hunger for electricity is getting so extreme that tech infrastructure planners are now looking to the U.S. Navy’s past for answers.

Yes—retired nuclear reactors from aircraft carriers and submarines could soon be repurposed to power massive AI data centers.

When AI Needs So Much Power, Even Nuclear Submarines Start Looking Useful

A Texas-based company called HGP Intelligent Energy has submitted a proposal to the U.S. Department of Energy asking for a federal loan guarantee to kick-start an unusual plan.

The idea: reuse two decommissioned U.S. Navy nuclear reactors to supply power for an AI data center located at Oak Ridge National Laboratory in Tennessee, tied to President  Genesis Mission.

If approved, the project could generate 450 to 520 megawatts of electricity—enough to support some of the most energy-hungry AI systems in the world.

These Are Not Experimental Reactors — They’ve Already Spent Decades at Sea

The U.S. Navy has relied on nuclear power for over half a century, using:

Westinghouse A4W reactors on Nimitz-class aircraft carriers

General Electric S8G reactors on Los Angeles-class attack submarines

Many of these ships are now being retired. The USS Nimitz, commissioned in 1975, is approaching the end of its service life, and a significant portion of Los Angeles-class submarines have already been decommissioned.

Instead of scrapping the reactors, HGP wants to reuse them.

Military Nuclear Reactors Have an Unmatched Safety Record

According to the World Nuclear Association, the U.S. Navy has operated more than 100 nuclear reactors over 50+ years with zero radiological accidents.

That record is now being used to argue that these reactors are not only powerful—but safe enough for civilian AI infrastructure.

If approved, this would be the first-ever case of U.S. military nuclear reactors being repurposed for civilian power generation.

Why This Is Faster and Cheaper Than Building New Nuclear Plants

Building a new nuclear power plant can take 10–20 years and cost far more than AI companies can wait for.

HGP estimates the reused-reactor approach would cost $1–$4 million per megawatt, which is:

Much cheaper than traditional nuclear plants

Faster than small modular reactors (SMRs) being explored by Google, Microsoft, Amazon, Meta, Oracle, and Nvidia

The total project cost is expected to land between $1.8 and $2.1 billion.

What Happens When These Reactors Finally Shut Down for Good?

Decommissioning nuclear reactors is extremely expensive. The company says it plans to:

Create a dedicated decommissioning fund

Share revenue with the U.S. government

Avoid sending reactors to long-term storage sites like Hanford

For context, dismantling the U.S.’s first nuclear-powered aircraft carrier reportedly cost over ten times more than retiring its last conventional supercarrier.

AI’s Energy Crisis Is Forcing Once-Unthinkable Ideas

AI data centers are consuming power faster than grids can expand. As models get bigger and more complex, the energy problem is becoming unavoidable.

“We already know how to do this safely and at scale,” said HGP CEO Gregory Forero, pointing to decades of military nuclear experience.

The Future of AI Might Be Powered by the Cold War

Using nuclear submarine reactors to power AI might sound absurd—but it highlights how serious the AI energy crisis has become.

What was once Cold War technology may soon be fueling the next generation of artificial intelligence.

And if this plan moves forward, it could completely change how AI data centers are powered.