Microsoft goes nuclear — AI demands spark unlikely alliance

Generative AI is a power-hungry beast. As its demands for more, larger data centers continue to grow, so does the strain placed on existing power grids and power plants. The answer to the growing AI energy problem? Microsoft believes it's nuclear, and the company has joined the World Nuclear Association (WNA) in what the Director General of WNA, Dr. Sama Bilbao y León, calls "a game-changing moment for our industry."

Within WNA's press release announcing the new membership (via TechRadar), it is expressed that Microsoft is interested in collaborating on a few key topics.

• Advanced Nuclear Technologies — Microsoft will support the WNR and its other members in the "development and deployment of small modular reactors (SMRs) and next-generation reactors." This included fusion energy.
• Regulatory Efficiency — Microsoft will advocate for "streamlined licensing processes" that reduce startup times while maintaining proper security protocols.
• Supply Chain Resilience — Microsoft will aid in bolstering the global supply chains required for nuclear energy.

The company's Energy Technology team, led by Dr. Melissa Lott, is overseeing the new alliance with WNR. Lott says that this partnership "reflects the strategic moment that our industry is working in as we work to meet our carbon-free energy goals."

Microsoft joining the WNA seems like a natural step. Many of the biggest AI and AI-related firms, the Redmond giant included, are quickly realizing that AI's demand for energy is skewing climate targets and decimating power grids.

This isn't the first time that Microsoft has expressed interest in nuclear energy to solve its problems, either. In 2023, when Microsoft's AI venture was really only getting started, the company partnered with Helion, a tech firm focusing on generating energy through nuclear fusion. Helion hopes to begin generating power by 2028.

In 2024, Microsoft signed a 20-year deal with Constellation Energy to restore the Three Mile Island Unit 1 reactor to produce "approximately 835 megawatts of carbon-free energy." Yes, that's the same nuclear plant where the Three Mile Island incident occurred in 1979, leading to a partial reactor meltdown. Affected was Unit 2, separate from Unit 1; Constellation says Unit 2 is expected to be back online by 2028.

Should Constellation and Helion's plans pan out, that still leaves only a couple of years between theoretical nuclear fusion operation and Microsoft's lofty 2030 carbon-neutral pledge.

Microsoft's power consumption has never been higher

How is Microsoft faring in terms of carbon-neutral goals? Not well, at least from a more narrow viewpoint.

A report from Michael Thomas discovered that Microsoft consumed 24TWh of electricity in 2023, which is more than 100+ countries consume. Google was right there alongside Microsoft with the same consumption numbers.

In May, Microsoft published its 2025 Environmental Sustainability Report. Within the report were some massive numbers that threw a snag into Microsoft's plans to go carbon negative by 2030.

Not only did Microsoft see a 23.4% increase in carbon emissions (mostly on AI's tab), but it also saw an incredible 168% increase in overall energy demands compared to the company's 2020 baseline. Those numbers from the Thomas report certainly don't look unrealistic.

Microsoft President Brad Smith noted, following the Microsoft sustainability report, that the 2030 goals are a "marathon, not a sprint."

OpenAI, creator of ChatGPT and the AI firm with which Microsoft is most heavily invested, has a similar energy problem. OpenAI CEO Sam Altman stated in 2024 that the only solution to AI's ever-increasing energy demand is nuclear fusion.

OpenAI launched its latest GPT-5 model in August 2025, and it arrived with higher power consumption than ever before. According to the University of Rhode Island's AI lab, GPT-5 consumes up to 18Wh of energy per query.

With up to 2.5 billion requests on an average day, GPT-5 needs roughly 45GWh to function, or enough energy to power 1.5 million homes in the US. To put that into perspective, a modern nuclear power plant produces somewhere between 1 and 1.6GW of power per reactor per hour.

There's been plenty of argument over exactly how much energy is needed to run generative AI, but the bottom line is that current energy solutions can no longer keep up.

Why use nuclear energy to power AI data centers?

Nuclear energy is often misunderstood. Whether it's nuclear fusion (the combining of atoms) or nuclear fission (the splitting of atoms), both types produce heat, which produces steam from water, which is then used to spin turbines. The spinning of those turbines is what creates power.

Considering the primary by-product of nuclear energy is water vapor, it seems like the magic answer for companies that need a lot of power and not a lot of carbon emissions. You can see why a company like Microsoft is interested.

Of course, that's ignoring the nuclear waste that's produced by the fission process, which is a whole other issue when it comes to safe storage and disposal. There's also the matter of meltdown, which is rare but incredibly deadly and damaging.

Nuclear fusion doesn't produce the same amount of radioactive waste, but it is, unfortunately, still in the experimental stage. As Alex de Vries, data scientist and researcher at Vrije Universiteit Amsterdam, has previously stated, fusion energy to address AI demands is "wishful thinking."