Blog / Tech, New & Old /

Nuclear and AI data centres: why tech wants more power

When Microsoft agreed to buy electricity from a restarted reactor at Three Mile Island, a lot of people did a double take. That is the site of the worst nuclear accident in United States history, from 1979. The reactor involved is the other one on the site, Unit 1, which kept running until it was shut down in 2019 for economic reasons. The company that owns it, Constellation, said in September that it would bring it back, with Microsoft agreeing to buy its power for 20 years.

My view: this is less a story about nuclear power than about what AI is doing to the demand for electricity. And I think the honest position is “understandable, but not free of problems”.

Why AI is hungry

Behind every chatbot answer or generated image is a building full of computers. Training a large model takes a huge number of specialised chips running for weeks. Serving millions of people afterwards takes still more. Those chips use a lot of electricity and make a lot of heat, which then needs cooling, which uses more electricity and often a lot of water.

A normal web server is a bit like a fridge: always on, modest. A rack of AI chips is more like a row of industrial ovens. The companies building them say demand is growing fast, and they’re planning bigger sites than they have built before.

Power companies and grid planners in several countries have started warning that data centres are changing their forecasts. I don’t have a dependable number for you, and I’d distrust anyone quoting one with too much confidence. The direction, though, is clear enough: more demand, concentrated in specific places.

Why nuclear, specifically

Big tech companies have promised for years to run on clean energy. Solar and wind are cheap, but they depend on weather. A data centre needs power at 3 a.m. in a calm, cloudy week, too.

Nuclear has a simple pitch:

  • It runs all the time. A reactor produces steady power day and night, unlike solar or wind.
  • It’s low-carbon. Operating a nuclear plant doesn’t emit carbon dioxide in the way gas or coal plants do.
  • It’s dense. A small footprint produces a large amount of energy.

Microsoft’s deal isn’t the only sign. Amazon bought a data centre campus next to a nuclear plant in Pennsylvania earlier this year, and several companies have talked about small modular reactors, a newer type of smaller reactor that has been promised for a long time but is not yet widely in service.

The reasons for caution

Here’s where I’d slow down.

1. Restarting is not the same as adding. If a reactor that was shut down comes back, that adds supply. But if a data centre simply buys power from a reactor that was already feeding the grid, then everyone else has less clean power to use. Whether a deal adds new clean capacity matters a lot.

2. New nuclear is slow and costly. Projects have often run years late and over budget. Small modular reactors may change that, but so far they remain more promise than product. Tech timelines (AI moves in months) do not match reactor timelines (years, sometimes decades).

3. Waste and risk haven’t gone away. Long-term storage of spent fuel is still unresolved in many countries. Accidents are rare but severe, and local communities carry that worry.

4. Efficiency is the cheaper win. The cleanest kilowatt-hour is the one you never need. Better chips, smaller models, and not using a giant model for tasks a small one can handle all reduce the load. It would be strange if the industry looked only for more supply and not for less demand.

The strongest counter-argument

A fair response to my caution is: “AI demand is real, the grid needs reliable low-carbon power, and the alternative may be more gas plants. If nuclear helps avoid that, it’s a good trade.”

I think that’s a serious argument. If the choice is truly between a nuclear plant and a new gas plant, I’d lean towards nuclear on emissions grounds. My worry is about the framing. The choice isn’t only between those two. It also includes efficiency, building more renewables with storage, and asking whether every AI feature is worth its energy cost. A company that adds an AI summary button to everything is making a decision about power use too.

Why it matters to normal internet users

This might sound far from your day. It isn’t.

  • Cost. If data centres compete with homes and factories for electricity, prices can be affected, in places where supply is tight.
  • Reliability. Anyone who has lived with unreliable grid power, as many people in Nigeria have, knows that electricity is the hidden foundation of everything online. The internet feels weightless and isn’t.
  • Your own use. Each AI request has a small energy cost. It’s tiny for one question and adds up at scale. It’s one more reason to use AI where it helps rather than everywhere by reflex.

If you run a website, it also touches you indirectly. Hosting, caching and a lighter site all use less computing. A fast, lean site is kinder to your visitors’ data bills, and in a small way to the planet.

What to watch

  • Whether more companies sign deals that add new capacity rather than reshuffling the existing supply.
  • Whether small modular reactors actually deliver on schedule.
  • Whether AI companies publish real energy figures for their models. Honest numbers would help everyone have a better argument.
  • Whether efficiency keeps pace with demand.

Key takeaways

  • AI needs a lot of reliable electricity, and nuclear is attractive because it runs constantly with low emissions.
  • Restarting or buying from existing plants is not the same as adding new clean power.
  • New reactors are slow and expensive, so efficiency matters just as much.
  • Electricity is the hidden cost behind every “free” AI feature.

Do you think AI is worth the extra power it needs? I’d like to hear where you draw the line.

/ Written by

Ekom

Ekom is a Website Administrator and Technical SEO Specialist passionate about website management, e-commerce, digital growth, and building digital products. He shares practical insights, experiments, and lessons from working at the intersection of technology, SEO, and the web.

More about me →