Powering AI's future: The case for nuclear energy in data centers

The AI energy crisis

Nearly half of the planned 2026 large data center projects have yet to disclose a power strategy, according to research by Currence, formerly Sightline Climate. Compounding the constraint, critical electrical components, including transformers, switchgear and backup batteries, are increasingly in short supply.

The rack-level power shift is equally dramatic. Traditional data centers were designed for 5 kW to 10 kW racks. However, integrating AI into these legacy facilities requires introducing modern GPU racks, such as Nvidia's Blackwell-generation models, which draw at least 120 kW each.

With the 2026 Vera Rubin NVL72 approaching 200 kW and the future Rubin Ultra NVL576 "Kyber" model expected to reach 600 kW by 2027, power density has increased by 60x to 120× over the past decade.

Former Google CEO Eric Schmidt, in testimony before a U.S. congressional panel, said that AI data centers will need 29 GW of additional power by 2027 and 67 GW more by 2030, according to Tech Policy Press.

Yet the grid is not prepared. The US national grid was built for 1% to 2% annual growth, but AI is demanding tens of gigawatts of power right now. The grid cannot handle that demand.

Connecting new data centers with highly demanding AI workloads to the grid requires submitting a request and joining an interconnection queue with a 4- to 5-year average wait time. Every month of waiting costs these companies their place in the AI race. A 60 MW data center loses roughly $14.2 million per month due to delays, according to research by STL Partners and Foresight. As a result, data center managers are opting to bypass the grid entirely and generate power on-site while exploring alternative energy sources.

Renewable energy cannot reliably meet this demand because it is intermittent. Solar cannot work at night, and wind is unpredictable. Although backup generators remain an option, they are expensive, pollute the environment and are mainly designed for emergencies, not continuous operations.

While the tech giants have made net-zero and carbon-neutral commitments, the AI-driven demand widens the gap between commitment and reality. Data center sustainability metrics reveal that, despite industry-wide efficiency gains, emissions are now rising as AI workloads are outpacing efficiency improvements.

As of today, about 40% of the electricity used by data centers comes from natural gas, and about 15% comes from coal, according to the IEA. Nuclear energy is growing in popularity, accounting for 20% of data center power and emerging as the most promising path to sustainable AI scalability.

Why nuclear energy fits AI workloads

Nuclear energy emerges as a leading approach, offering unparalleled baseload reliability, high power density, and a carbon-free generation pathway that aligns perfectly with the needs of advanced computing and AI workloads.

Baseload reliability

To conduct a single training run for a frontier model, an uninterrupted power supply is needed for weeks and even months. Nuclear is perfect for this as it delivers energy 24/7 year-round, regardless of the weather and time of day.

With nuclear reactors achieving capacity factors above 90%, matching the operational profile of AI workloads, they far exceed solar and wind.

Power density and quality

A typical nuclear reactor generates up to 1,000 MW or more of electricity within a facility spanning a few square miles. This is especially favorable to data center operators facing land acquisition constraints.

A one-gigawatt nuclear plant can power a hyperscale campus equivalent to a small city. Compared with the hundreds of square miles of solar farms that would otherwise be required. Nuclear plants deliver consistent voltage and frequency with minimal fluctuation.

Carbon-free generation

Nuclear energy generates no direct CO₂ emissions, providing a straightforward route for data center decarbonization that supports net-zero goals while increasing computing capacity.

Grid independence and transmission efficiency

Data centers can be co-located with nuclear plants, eliminating the need for long-distance transmission and reducing energy loss and maintenance costs. This on-site or near-site generation will allow operators to bypass the congested interconnection queues that are currently delaying most projects by years.

Industry momentum: Tech Giants embrace nuclear

Major tech companies have made significant investments and entered into agreements in the nuclear energy sector, signaling a strong commitment to sustainable power sources.

Microsoft

In September 2024, Microsoft signed a 20-year deal with Constellation Energy to restart Unit 1 of the Three Mile Island nuclear facility in Pennsylvania. Constellation intends to spend $1.6 billion, with a $1 billion federal loan. The reactor should be operational in 2028, and the tech giant will buy every megawatt of energy it generates for 20 years, totaling about 835 MW.

Google

In October 2024, Google announced a contract with Kairos Power to develop a fleet of small modular reactors (SMRs) targeting 500 MW, with the first reactor expected to come online by 2030. The tech giant also contracted with Elementl Power to prepare sites for advanced nuclear installations of at least 600 MW each.

Amazon

In the spring of 2024, Amazon entered into a nuclear power purchase agreement with Talen Energy in Pennsylvania and bought a 960 MW data center adjacent to Talen Energy's Susquehanna nuclear plant for $650 million.

Amazin also invested over $1 billion in several nuclear projects in partnership with X-energy, Dominion Energy and Energy Northwest, developing SMRs totaling over 5 GW of potential capacity.

The SMR Advantage

Unlike traditional nuclear reactors, which are built to supply power to the grid and can take up to a decade to build and cost billions of dollars upfront, SMRs are factory-built and can be trucked to the site.

Each SMR has a capacity of 50 MW to 300 MW and can be deployed in increments to match actual compute demand rather than relying on projections. This "pay-as-you-go" model is similar to how cloud hyperscalers expand their regions.

This modularity allows operators to bypass interconnection queues by delivering power directly to data center campuses. SMRs don't yet exist at a commercial scale. However, with NuScale's US460 design receiving the standard design approval from the Nuclear Regulatory Commission (NRC) in May 2025, the first one in the US might be ready by 2030.

Challenges and Considerations

Despite the promise of nuclear energy for AI advancement, it faces hurdles that must be addressed in strategic planning.

Regulatory hurdles

The regulatory framework for obtaining NRC licensing for new reactors or designs is adapting to modular technologies. Still, it remains a multi-year process, despite the executive order issued by President Trump in 2025, which set more aggressive permitting deadlines.

The risk behind first-of-its-kind deployments in the case of SMRs is very real, and delays can disrupt project timelines.

Economic factors

There is a huge upfront capital cost associated with going nuclear: $6,417 to $12,681 per kW, compared to approximately $1,290 per kW for natural gas.

However, with hyperscalers having long infrastructure commitments, the economics of running nuclear for 40 to 60 years at low marginal fuel costs make clear what the volatile natural gas market offers.

Technical integration

Connecting nuclear plants to data centers and delivering power to GPU racks requires highly skilled, specialized engineering.

For high-density racks, advanced cooling architectures such as direct-to-chip liquid cooling, rear-door heat exchangers, or immersion cooling need to be implemented to manage thermal loads and AC-to-DC conversion far beyond what traditional HVAC was designed to manage.

Scalability questions

A forecast by Goldman Sachs Research projects that 85 GW to 90 GW of new nuclear capacity will be needed by 2030.

A critical question is whether deployments and executions with energy companies will align with AI's growth trajectory.

Water and community concerns

Data centers and nuclear plants require large amounts of water for cooling. This usually comes from a shared source used by community residents, raising concerns about water availability and quality. These concerns have to be addressed to maintain the social license required to operate.

Final thoughts

Nuclear energy is no longer optional for AI infrastructure; it is a necessary component. The race between energy availability and AI innovation will define the next decade of technological progress.

The exponential growth in demand for AI infrastructure isn't slowing down anytime soon, and meeting this huge demand will require long-term policy commitment and deeper industry collaboration.

Wisdom Ekpotu is a DevOps engineer and technical writer focused on building infrastructure with cloud-native technologies.