- Google is ready to go nuclear in the name of AI.
- The search giant has struck a deal to bring new nuclear plants online to power its AI data centers.
The energy demands of the data centers behind the generative AI boom are proving so great that they're pushing Big Tech firms into an unfamiliar position: becoming long-term customers to the nuclear industry.
Google became the first tech giant to broker a deal for entirely new nuclear power plants after it unveiled a partnership with industry firm Kairos Power on Monday. The deal will see the startup build "multiple" small modular nuclear reactors for the search giant.
The first of these reactors — which includes the one Kairos began constructing in East Tennessee in July — is set to be online by 2030. The others should be ready by 2035 and will enable up to 500 megawatts of "new 24/7 carbon-free power" to US electricity grids. Assuming continuous operation and average consumption levels, that's enough to power about 438,000 US households for one year.
Google CEO Sundar Pichai hinted earlier this month that this was coming after he said in an interview that his company was "evaluating technologies like small modular nuclear reactors."
While deal terms have not been disclosed, the significance of the arrangement is hard to overstate — and not just because the US has brought only three nuclear reactors online over the past two decades.
AI is pivoting Google and its rivals toward nuclear power as they signal the need for a new solution to the tech world's growing energy problem.
Generative AI strains the energy sector
Big Tech companies are racing to build smarter AI, but this requires a lot of energy.
Announcing the Kairos deal, Michael Terrell, senior director of energy and climate operations at Google, noted that the grid "needs new electricity sources" to support the energy needs of AI technologies the search giant is now heavily invested in.
Google revealed just how energy-intensive its AI push has been earlier this year. Its annual environmental report, published in July, showed its emissions had risen 13% year-on-year in 2023, driven by "increased data center energy consumption and supply chain emissions."
That 86-page report didn't mention nuclear power, bar a single endnote. It shows that Google's Kairos deal is a sign that it is now ready to consider all options in its search for a supply of clean electricity to its data centers.
That's particularly important as it looks to build smarter versions of its AI model, Gemini. Ethan Mollick, a professor at Wharton, noted on X that the Google-Kairos deal "increases the odds that we will see AI models scale through at least 3 more generations/orders of magnitude (post GPT-5) to 2030."
The Google deal to acquire small nuclear reactors to power data centers increases the odds that we will see AI models scale through at least 3 more generations/orders of magnitude (post GPT-5) to 2030.
— Ethan Mollick (@emollick) October 15, 2024
According to Epoch, power was the binding constraint. https://t.co/FQtUsE0L9d pic.twitter.com/lChspwoZjC
Others have recognized this, too. Since OpenAI launched ChatGPT in late 2022, the world's biggest tech companies have been forced to confront the fact that staying competitive in generative AI will require vast supplies of clean energy.
By one estimate, a search request on OpenAI's buzzy chatbot demands around 10 times as much electricity as a general Google search. The large language models behind it make the computing power in data centers buzz and whirr intensively to produce cogent responses.
While solar and wind power offer clean means of electricity generation, some in the tech industry have suggested that they are not enough to meet the demands of AI.
For instance, Meta's chief AI scientist, Yann LeCun, said last month that AI data centers would have to be built next to nuclear power plants as, in his view, they can produce the continuous "gigawatt-scale, low-cost, low-emission electricity" needed.
"The advantage is that there is no need for expensive and wasteful long-distance distribution infrastructure," he wrote on X while suggesting solar and wind are "nice" but neither simple nor cheap.
Microsoft's nuclear moves
Google rival Microsoft signaled interest in nuclear power in September after it struck a 20-year power supply agreement with Constellation — an agreement that would see the Baltimore-based energy company reopen the nuclear plant on Three Mile Island.
The Island's second unit was shut down in 1979 after it became the site of the most serious nuclear meltdown on US soil, so the move from Microsoft to turn to the site is significant — even if it is reviving the island's first unit as opposed to the first one. It is expected to come online by 2028 and provide over 800 megawatts of power, about enough to keep a small-to-mid-sized city running.
In March, Amazon bought a data center in Pennsylvania from Talen Energy for $650 million, which is powered by the nuclear Susquehanna Steam Electric Station.
Meanwhile, OpenAI boss Sam Altman said at Davos earlier this year that an energy "breakthrough" driven by greater investment in nuclear fusion would be needed to accelerate future AI developments. Altman invested $375 million in a personal capacity into fusion power company Helion Energy in 2021.
The prospect of the energy source taking on a greater role in Big Tech's operations becomes even more significant when considering the nuclear industry's decades of stagnation due to delayed projects.
That doesn't seem like a deterrent, however. Big Tech firms look all in on AI. Expect them to be all in on energy sources that keep the technology ticking.