The race to build more AI capacity no longer rides on getting GPUs alone. Electricity, networks, and data centers are becoming just as central, and the consumption forecasts explain why Microsoft, Google, and Meta have started signing long-term nuclear agreements. The International Energy Agency (IEA) estimates data centers used around 485 TWh in 2025 and could approach 950 TWh by 2030.
Nuclear for data centers in 30 seconds
- The IEA expects global data center electricity use to nearly double between 2025 and 2030.
- Microsoft has backed recovering 835 MW of nuclear capacity in Pennsylvania through a 20-year power purchase agreement.
- Google is working with Kairos Power on advanced reactors and eyes up to 500 MW by 2035.
- Meta has announced agreements backing up to 6.6 GW of existing and future nuclear capacity.
- Nuclear offers continuous generation, but new reactors still face uncertainty on cost, timelines, and large-scale buildability.
The leap in generative AI also brings a challenge that wasn’t as sharp during earlier cloud expansion cycles: each new generation of accelerators packs much more electrical power into less space.
Filling a building with servers isn’t enough. You have to deliver hundreds of megawatts to it, put substations and transformers in place, cool the equipment, and keep a stable enough electrical supply for years.
That’s where nuclear becomes relevant again.
Gigawatts start to matter as much as GPUs
AI infrastructure is turning into an extraordinarily power-hungry industry.
The IEA estimates data center energy use grew 17% in 2025 and expects fast growth through the decade. Facilities built specifically for AI are projected to grow even faster.
The concentration of that use is what stands out. A traditional factory can spread activity across several locations. An AI training cluster needs thousands of accelerators connected over very high-speed networks working in sync.
That concentrates the electrical demand too.
The IEA figures that in 2027 an advanced rack could reach a peak demand roughly equal to 65 homes. Multiply by thousands of racks and it’s clear why energy planning increasingly shapes where the next generation of data centers gets built.
The bottleneck may move from semiconductors to energy availability. A company can buy thousands of GPUs, but they’re not much use while they wait years to secure enough grid connection.
Nuclear offers something hard to get from variable sources alone: continuous, predictable generation for decades.
That doesn’t mean replacing renewables. The IEA expects about half of the global increase in data center electricity through 2030 to come from renewables. AI infrastructure is trending toward a mix of renewables, nuclear, storage, gas, and stronger grids.
Microsoft recovers 835 MW and Google invests in new reactors
One of the most notable moves comes from Microsoft.
In 2024, Constellation announced a 20-year electricity purchase agreement backing the recovery of the old Three Mile Island Unit 1, now called Crane Clean Energy Center.
The facility could return about 835 MW of capacity to the grid.
An important distinction: this isn’t the reactor involved in the 1979 Three Mile Island accident. That was Unit 2. Unit 1 ran on its own for decades and closed in 2019 for economic reasons.
The deal shows one effect of data center growth: energy assets that are no longer profitable enough can regain value if a buyer will commit for decades.
Google is taking a different route.
Its agreement with Kairos Power isn’t about recovering large traditional reactors but about deploying advanced, smaller ones.
The first planned project is Hermes 2 in Oak Ridge, Tennessee. The agreement between Kairos Power and the Tennessee Valley Authority proposes putting 50 MW on the grid from 2030. That electricity would feed Google’s data centers in Tennessee and Alabama.
The relationship is more ambitious, aiming for up to 500 MW of capacity by 2035.
The project’s technical goal is to see whether a new generation of reactors can avoid some of the main economic problems historically tied to large nuclear plants.
Small modular reactors (SMRs) aim to use standardized designs and repeatable components, which on paper should cut costs and construction times. But commercial, large-scale proof is still missing.
Meta raises its nuclear commitment to 6.6 GW
Meta has gone further still.
In January 2026 it announced agreements with Vistra, TerraPower, and Oklo that, together with its earlier Constellation deal, back up to 6.6 GW of new and existing nuclear capacity by 2035.
That figure needs context.
It doesn’t mean Meta will suddenly have 6.6 GW from new reactors. Part comes from operating plants, and other parts depend on projects that still need regulatory approval, financing, construction, and startup.
TerraPower initially plans two Natrium units with up to 690 MW, with the option to build six more later.
Oklo aims to develop a nuclear campus in Ohio with up to 1.2 GW.
These projects matter to the data center industry because they suggest a different relationship between generation and large consumers.
Traditionally a plant supplies electricity to a grid serving many customers. AI’s growth is creating individual buyers that may need power on the scale of small towns.
And, crucially, buyers able to sign long-term contracts.
For a nuclear developer, knowing in advance who will buy the electricity for 20 years can heavily shape a project’s financing.
AI could reshape reactor economics
Nuclear has faced a paradox for decades.
It packs enormous energy, reactors can run for long stretches, and their output has low direct carbon emissions. But building new reactors in many Western markets has become costly and slow.
An analysis by Tomas Pueyo, Why Nuclear Is the Best Energy, gathers many of the arguments used to advocate for the technology, from energy density and fuel availability to land use, waste, and safety compared to other sources.
The piece takes a pro-nuclear stance, and some of its readings of radiation, past accidents, waste, and regulation are debatable. Still, it raises a key question that’s especially relevant given AI’s expansion: how much of nuclear’s economic problems come from physical technological limits versus how plants are built and regulated.
The upcoming advanced-reactor projects will help answer that.
If SMRs can repeat designs, manufacture components industrially, and cut construction times, they might find their first significant markets in data centers.
If instead they hit delays and cost overruns like recent large nuclear projects, the tech industry will have to look elsewhere for large capacity.
From GPU shortages to megawatt shortages
In the early years of the generative AI boom, attention centered on NVIDIA and GPU availability.
The next phase blows that problem up much larger.
An AI cluster needs accelerators, HBM memory, high-speed networks, storage, cooling, and electrical infrastructure that can deliver enormous loads. It also needs transformers, substations, and grid connections that weren’t designed for hundreds of megawatts in data centers.
That’s why companies whose core business isn’t power generation are now directly involved in energy supply.
They’re not only building data centers; they’re trying to secure the energy that lets those data centers run.
The IEA forecasts global data center energy use could reach 950 TWh by 2030, roughly double what it was five years earlier. If that holds, energy infrastructure will increasingly decide where new AI capacity can go.
And that may be the biggest shift of all.
For years, data center location depended on connectivity, latency, taxes, land, and proximity to users. The availability of hundreds of megawatts is now climbing that list.
In short, the AI race is no longer confined to chip factories and model labs; it’s moving into power plants, transmission lines, and substations.
Nuclear is back on big tech’s radar precisely because it can supply one of the resources the next AI generation will need at enormous scale: electricity available 24/7 for decades.
Frequently Asked Questions
Why do AI data centers need so much electricity?
Advanced models use thousands of accelerators running at once. Add memory, networks, storage, electrical systems, and cooling, and the total power needs of new data centers rise sharply.
Is Microsoft reopening Three Mile Island?
Microsoft’s agreement with Constellation backs the financial recovery of the old Three Mile Island Unit 1, now called Crane Clean Energy Center. It’s not the Unit 2 reactor involved in the 1979 accident.
Will Google use small nuclear reactors for its data centers?
Google has a partnership with Kairos Power to support deploying advanced reactors. The program aims for up to 500 MW by 2035, though that capacity still has to be built and brought online.
Will nuclear replace renewables for powering AI?
The forecasts point to a complementary mix. The IEA expects renewables to supply a big part of the new electricity for data centers, while nuclear, gas, storage, and upgraded grids serve other needs depending on the market.
