March 6, 2026

As data center operators strain to find enough reliable electrical power to run soaring AI workloads, it’s increasingly clear that nuclear power generation will need to provide a bigger share of that power in the coming years.

Asked which two non-fossil fuel energy sources are gaining the most traction, 32% of data center operators cited nuclear power, trailing only solar and battery energy storage systems, in a recent AFCOM State of the Data Center report. Beyond the advantages of nuclear as a low carbon emission energy source, the industry simply needs more total baseline power generation capacity. Data center power will make up 6.7-12.0% of US electricity consumption by 2028, up from 4.4% in 2023, Lawrence Berkeley National Laboratory forecasts.

“Let’s assume that even a fraction of what everyone is saying about AI comes to pass. We’ll need a ton more power,” says Romi Mahajan, CEO of ExoFusion, a company focused on fusion IP, simulation, design, and scientific innovation, working in plasma confinement and liquid metals. ExoFusion is focused on accelerating the path to commercially viable fusion nuclear power. (Today’s nuclear power uses fission. More on the distinction later.)

Mahajan offers a useful way to think about power options: the “energy stack,” meaning the mix of power generation sources that supply a data center. That mix could include natural gas, solar, wind, nuclear, hydro, geothermal, and coal, and it can come through the electrical grid or be generated on-site of a data center. A diversified mix can improve cost and reliability.

“You never want to have a monoculture in something this important,” he says.

Focusing on the energy stack and the mix of sources gives businesses, policymakers, and communities a common language for hard conversations about how to meet fast-rising electricity demand driven by data centers, electric vehicles, industrial electrification, and population growth.

“Should we get fossil fuels to 5, 10, 20 percent of the energy stack? These are the questions we need to talk through as a society,” Mahajan says.

How Soon for Nuclear’s Impact—Including Fusion?

Mahajan will explore the role of nuclear power for data centers with FusionX Group CEO Stuart Allen and Type One Energy SVP Matt Miles, in a panel discussion at Data Center World, April 20-23 in Washington, D.C.

“This won’t be Pollyanna, we’ll talk about the tough stuff,” Mahajan says. “Is there enough innovation to make fusion viable in the next decade? Are we overbuilding into an energy bubble? Do our planning models account for externalities such as community concerns?”

One key takeaway from the panel will be an understanding of the potential for fission and fusion nuclear technology to address data center power needs, and how soon we might see the impact from those. Mahajan offers a quick overview of the two nuclear approaches:

Fission: This is the traditional nuclear power generation approach, involving splitting a heavy atomic nucleus, and a new nuclear power plant typically has required a 10+ year planning, permitting, and buildout span. But the technology is advancing quickly, and small modular reactors (SMRs) promise to help capacity get deployed much faster. SMRs can be built in factories as prefabricated units that are shipped to a site – whether a power plant or directly to a data center --for installation. Mahajan expects nuclear power to provide a larger share of data center power within 10 years using both conventional nuclear plants and SMRs.

Fusion: This approach involves combining two light atomic nuclei under extreme heat and pressure, releasing massive energy in the process. Using the same process that powers the sun, fusion promises far more power than fission without the nuclear waste or risk of meltdown. The trick is getting more power out than what’s used to cause the reaction.

Given recent progress across the fusion research community, including numerous startups such as ExoFusion and researchers in academia and national laboratories, Mahajan believes we will have pilot plants running fusion process within 10 years.

“Of all our technological and scientific processes, fusion can deliver the highest baseload power of any source we know of. It’s what powers the sun,” he says. “From a societal standpoint, it doesn’t matter who succeeds, as long as someone does. It’s going to take the fusion village.”