Nuclear Solutions Years Away as Data Centers Drive Up Energy Costs

The data center boom, spurred on by the rise of services and technologies related to artificial intelligence (AI), is anticipated to significantly raise the cost of electricity for many Americans.

Global electricity consumption by data centers is expected to increase by 15 percent per year, four times the rate of consumption by all other sectors, according to an April report by the International Energy Agency, an intergovernmental organization based in Paris.

China and the United States account for roughly 80 percent of that global increase.

“AI will be the most significant driver of this increase, with electricity demand from AI-optimized data centers projected to more than quadruple by 2030,” the report states.

American consumers are already feeling the strain on their household finances as data centers consume an increasing portion of the available power wherever they sprout up. This extra demand encourages utilities to raise prices for all customers.

Energy providers may also need to invest in new generation infrastructure to meet the additional power needs, adding to the costs for consumers.

In the 12 months leading up to August, the average price of electricity for urban consumers in the United States rose by more than 6 percent, according to the Bureau of Labor Statistics, far higher than the historical average increase of a little more than 2 percent annually.

Americans’ electrical costs are expected to rise by as much as 18 percent on average in the next five years, according to a fact sheet by the U.S. Energy Information Administration, outpacing the cost of all other forms of consumer energy.

What is to be done remains an open question, but some companies are positioning themselves to sell modular nuclear reactors as a possible solution.

Nuclear Power Key to Emerging Energy Crisis

One such company is the California-based Oklo, which designs, builds, and operates small modular reactors (SMRs) for data centers.

Brian Gitt, a senior vice president for Oklo, said that SMRs are the only feasible way to generate enough power to satiate the massive energy consumption demands of AI-related data centers.

“Most of the utilities are out of power around the country,” Gitt said last month at an AI forum in Las Vegas.

“So it’s a huge wake-up call for the [energy] industry to say the whole practice of just looking out for a couple of years … is not going to work. We need to look out further.”

To illustrate how quickly energy demands could increase, Gitt pointed to the proposed development of an 11-gigawatt data center in Texas.

“New York City at total peak load is less than 11 gigawatts,” Gitt said. “And this is one data center campus … that is the trajectory of where we’re going.”

The Texas data center is behind-the-meter, meaning it will not have to run its energy through the grid. But the project points to just how quickly and sharply energy consumption rates are anticipated to rise to accommodate the demands of AI-related technologies.

The key problem is that the nuclear facilities required to effectively power such facilities take years to permit and build, to say nothing of their up-front costs.

Gitt said that traditional nuclear power plants can cost as much as $10 billion up front, take as long as a decade to construct, and require hundreds of acres of land and access to enormous water resources for cooling.

That’s where SMRs come in.

Oklo’s key offering is a 75-megawatt SMR that requires about three acres and minimal water and can be built in about three years, if permitting goes according to plan.

The idea is that tech companies can use the SMRs to generate power for their campuses in a modular fashion, adding small reactors as the facilities expand instead of investing in a giant facility all at once.

“By building out your power infrastructure in similar-sized chunks, you are synchronizing the buildout so you don’t overbuild and have power infrastructure that’s sitting there uneconomical for too long, which is really critical to making sure you’re delivering a really cost-effective solution,” he said.

US Nuclear Goals Years Away

Hardships remain. Oklo’s first SMR has been in development since 2019 and is still not expected to be brought online until 2028.

That means it will be years yet before data centers stop pulling power from the grid by any noticeable measure, and all the while, the costs for consumers will continue to rise.

But Gitt said he saw an opportunity in that hardship for tech companies to ingratiate themselves in the communities where they built data centers.

By slightly overbuilding their nuclear power capacity, he said, data centers could run power to the grid, effectively subsidizing local energy prices.

There are many ways, Gitt said, in which “data centers actually can contribute value to the community in terms of reliability and resilience without impacting costs.”

The reality of AI’s energy consumption already appears to be driving relations between U.S. communities and the tech companies building out data centers.

In a company statement on Aug. 4, Google agreed that it would begin curbing electricity usage to free up grid space when requested by utilities in Tennessee.

In exchange, however, the Tennessee Valley Authority will begin purchasing energy from Google’s Hermes 2 reactor when it comes online in 2030.

“We believe this is a promising tool for managing large new energy loads and facilitating investment and growth,” Google stated.

The Trump administration is moving to accelerate nuclear development to support AI development, and President Donald Trump has signed several executive orders aimed at quadrupling the nation’s nuclear energy capacity by 2050, including licensing 10 new reactors by 2030.

Until those facilities are available, however, electricity prices are expected to continue to rise for millions of Americans.