AI data centers are consuming electricity at an unprecedented rate, forcing Big Tech to pursue nuclear power as a long-term solution while households face immediate rate shocks. Global data-center electricity use is projected to roughly double to about 945 terawatts per hour (TWh) by 2030, up from approximately 415 TWh in 2024, with artificial intelligence (AI) as the primary driver. Meanwhile, New Jersey residents are experiencing the fastest electricity rate increases in the nation, with average household spending jumping 22 percent year-over-year, or nearly $400 more per home.

Why Is AI Suddenly Demanding So Much Power?

AI data centers run compute-intensive training and inference operations around the clock, 24 hours a day, 7 days a week. Unlike traditional data centers that can flex their power consumption based on demand, AI training clusters cannot pause when renewable energy sources like wind or solar dip. This constant, predictable demand is why nuclear power has become attractive to hyperscalers, the massive cloud computing companies that power the AI boom.

The concentration of this demand makes the problem acute. The United States already accounts for roughly 45 percent of global data-center electricity use, and the International Energy Agency (IEA) estimates US data-center demand could rise approximately 130 percent by 2030. That's a staggering increase that cannot be met by intermittent renewable sources alone.

What Nuclear Deals Have Big Tech Actually Signed?

Three major technology companies have committed to nuclear power in concrete, multi-gigawatt agreements. These deals represent fundamentally different strategies for securing reliable, carbon-free electricity:

• Microsoft and Constellation Energy: Restarting Three Mile Island Unit 1, a retired reactor in Pennsylvania, with a capacity of 835 megawatts (MW) and a targeted restart date of 2028. The company signed a 20-year power purchase agreement (PPA) worth approximately $1.6 billion to restart the facility, which will supply enough power for roughly 800,000 homes, all dedicated to Microsoft's AI operations.
• Amazon and Talen Energy: Securing up to 1,920 MW from an existing nuclear plant in Susquehanna, Pennsylvania, with full capacity expected by approximately 2032 and a contract extending to 2042. Amazon is also purchasing a $650 million data-center campus adjacent to the plant and exploring new small modular reactors (SMRs) for additional capacity.
• Google and Kairos Power: Signing the first US corporate deal for a fleet of new small modular reactors, targeting 500 MW of capacity from approximately half a dozen reactors, with the first site expected to come online by 2030.

These three strategies reveal how hyperscalers now treat power procurement as a strategic bottleneck rather than a simple utility bill. Microsoft is reviving dead infrastructure, Amazon is co-locating compute next to existing plants, and Google is betting on entirely new reactor technology.

When Will Nuclear Actually Power AI Data Centers?

Here lies a critical timing mismatch. AI's electricity demand is rising sharply right now, but most new nuclear capacity will not arrive until 2028 to 2035 and beyond. The nearest-term and most certain projects are the restart of Three Mile Island Unit 1 in 2028 and Amazon's Susquehanna co-location, both of which use existing reactors where the hardest part, building the plant, is already complete.

Medium-term projects like SMRs from Kairos and microreactors from Valar Atomics are genuinely new technology with first sites targeted for 2030 and beyond. New nuclear has a long history of delays and cost overruns, making these timelines uncertain. In the gap between now and when nuclear capacity arrives, data centers will continue to rely heavily on natural gas and the existing electrical grid.

On July 1, 2026, Valar Atomics demonstrated a milestone: its Ward250 microreactor produced live power for Nvidia's Blackwell-architecture AI chips at a small data center in Utah, marking the first known case of a next-generation reactor directly supplying US AI compute. However, this was a small-scale demonstration, not yet an industry-wide solution. Valar, founded in 2023 by 27-year-old Isaiah Taylor, raised $450 million in 2026 and is valued at $2 billion, with plans for a future 30-megawatt, nearly waterless data center in Utah.

Who Is Paying the Price While Tech Waits for Nuclear?

Ordinary households in states like New Jersey are bearing the immediate cost of AI's power surge. Average household electricity spending in New Jersey rose about 22 percent from the prior year, the biggest annual increase of any state, according to data from MIT and Heatmap. For some residents, this translates into impossible choices about whether to run air conditioning during dangerous summer heat.

Catherine Hunt, an 82-year-old retiree in Monroe Township, New Jersey, told reporters that her monthly electricity bill, once around $50, reached $400 in colder months, forcing her to turn off the heat and put on extra layers indoors. Governor Mikie Sherrill issued an executive order directing the state utility regulator to reject proposed utility hikes, but as New Jersey ratepayer counsel Brian Lipman noted, keeping rates from rising further is not the same as making them affordable.

"The fundamental challenge is that we are experiencing tremendous growth and demand for electricity. There are just more people who want to use more electricity, and most of them are AI data centers," explained Abe Silverman, an energy research scholar at Johns Hopkins University.

Abe Silverman, Energy Research Scholar at Johns Hopkins University

PJM, the regional transmission organization that manages the electrical grid for much of the Northeast, confirmed that "substantial demand growth" tied to data centers and "the artificial intelligence race" is pushing prices upward, even though the grid currently has enough resources to meet demand.

How Are States Responding to the AI Power Crisis?

New Jersey lawmakers have begun targeting the largest AI facilities directly. Last week, the state legislature passed a bill requiring data centers drawing 100 megawatts or more to pay for grid upgrades. The legislation also mandates that large power users during peak-demand periods guarantee payment on at least 85 percent of the electricity they request for a decade, helping shield everyday customers if those facilities later reduce their usage.

Beyond regulatory measures, New Jersey is pursuing multiple strategies to ease pressure on the grid:

• Renewable Energy Expansion: Accelerating the development of solar, wind, and nuclear capacity to diversify the energy mix and reduce reliance on fossil fuels.
• Distributed Storage Solutions: Expanding rooftop solar installations and battery storage systems that can store energy during low-demand periods and release it during peak times.
• Virtual Power Plants: The Sherrill administration has directed the state utility authority to study virtual power plants, which connect batteries in homes and other buildings so stored energy can be used when the grid is under the most stress.

Silverman noted that tools like virtual power plants may offer faster relief than waiting for major new infrastructure to be built. "We absolutely need to be thinking about how do we expand the pie," he stated.

Silverman

What Does This Mean for the Future of AI and Energy?

The nuclear-AI partnership represents a structural, long-term solution to a problem that is happening right now. The demand is real and enormous: AI-optimized data centers are projected to see electricity demand more than quadruple by 2030. The marquee deals between Big Tech and nuclear operators are real and concrete. But the vision of AI running primarily on clean nuclear power is a decade-long build-out, with the safest bets being old reactors brought back to life and the exciting-but-unproven bets being small reactors still working toward their first commercial sites.

In the meantime, households will continue to absorb the cost of the transition. The gap between rising AI demand and available nuclear capacity means data centers will lean heavily on natural gas and grid electricity for years to come. For residents already struggling with 22 percent rate increases, the promise of future nuclear power offers little comfort today.