Technology firms are in a hurry to generate sufficient power to drive the AI boom, and they are increasingly turning to nuclear power as the solution to their escalating energy dilemma.
Behind AI’s Insatiable Appetite for Electricity
Meta signed a 20-year deal with Constellation Energy last month to assist in reviving an Illinois nuclear plant, following Amazon, Google, and Microsoft in seeking nuclear deals. The catch is, it will be years before nuclear energy can catch up with the ginormous electricity needs that are being generated by AI today.
The numbers tell a compelling story. According to the U.S. Department of Energy, electricity usage in data centers has increased threefold over the past decade alone and could double or triple again within the next decade by 2028. Data centers could consume as much as 12% of the total electricity used in America at that time.

So what is behind this vast hunger for power? The reason is in the way that real AI functions. To train an AI application like Meta’s Llama or ChatGPT takes prodigious amounts of computing power. The applications must “learn” from huge datasets using specialist computer chips known as GPUs that can perform thousands of calculations at once within networks of units.
Even after training is complete, AI programs continue to use a lot of electricity whenever an individual invites a chatbot to compose something or generate an image. Such a process, referred to as inferencing, causes the AI to reason through new data and generate responses from the data it already has.
The Power Problem of AI
All this processing creates massive amounts of heat, so data centers require strong air conditioning systems to prevent equipment from overheating. Most operators are resorting to water-cooling systems, but that still requires a lot of electricity to power the pumps and cooling equipment.
The environmental implications are enormous. As technology companies have spent years mapping out how they will power their operations with clean energy, the sudden popularity of AI has put a hole in well-laid plans. Today, most of the United States’ data center power is still created by combusting fossil fuels, primarily natural gas and sometimes coal, the International Energy Agency says.
The future energy mix for U.S. data centers illustrates the problem: renewable energy, such as wind and solar power, accounts for approximately 24% of the energy, and nuclear accounts for roughly 15%. The rest is from fossil fuels, and they create the planet-warming emissions that tech firms are attempting to eliminate.
Bridging the Gap Between Immediate Energy Needs and Long-Term Clean Solutions
France is an interesting case. The nation produces approximately 75% of its electricity from nuclear power, a world record. French President Emmanuel Macron has been touting this as a competitive strength for the development of AI, jokingly rephrasing Donald Trump’s “drill baby drill” slogan to “Here, there’s no need to drill, it’s just plug baby plug.”
But even as firms such as Meta are committing to the nuclear long-term, their short-term strategy continues to depend on natural gas. Entergy, a major US utility firm, is pushing to build gas-fired power plants in Louisiana solely to serve a gargantuan Meta data center campus.
This produces a timing issue that the whole tech sector is facing. Nuclear power produces a clean, continuous flow of electricity that potentially may be able to power AI’s enormous power requirements at some point, but the construction of new nuclear capacity or the revival of old plants takes years to achieve. AI demand is increasing in the meantime.
The International Energy Agency predicts that in the next few years, the majority of new power for data centers will be supplied by gas-fired power stations. These are relatively easy to construct and offer stable power, but also contribute to greenhouse gas emissions, which lead to global warming.
This places tech firms in a tricky spot in that they are attempting to hold their AI ambitions and their climate promises together. The industry response seems to be a two-part solution: natural gas for short-term demands while, simultaneously, investing in nuclear and green projects that will deliver cleaner fuel in the future. And the question is whether this plan will be able to produce clean energy quickly enough to satisfy AI’s increasingly growing electricity needs.