The Three Mile Island nuclear power plant has two reactors. The second reactor suffered a partial meltdown in 1979 and has remained out of service since. But the first reactor continued to operate without incident until 2019, when it was decommissioned for financial reasons, mainly due to competition from electricity generated by gas and wind power. Kotek says there are relatively few idle reactors that could also be brought back into service relatively quickly, but many power plant owners want to extend their operating licenses for their existing plants to try to ride the energy wave of AI.

Part of the enthusiasm among power plant operators is due to government incentives to keep low-carbon energy online. The Inflation Reduction Act provides tax credits tied to electricity production at existing nuclear power plants, but Kotek says the industry will also have to work to build new reactors if it wants to meet projected energy demand. The number of operating nuclear reactors in the United States peaked at 112 in 1990 and fell to 92 in 2022, and the most recently built reactors in the United States – at the Vogtle Power Plant in Georgia – took more than 14 years in the making and cost over a year. more than double the planned budget.

“The United States showed Vogtle that we’re not very good at building nuclear power plants,” says Todd Allen, chair of nuclear engineering and radiological sciences at the University of Michigan. But Allen points out that China appears to be building nuclear power plants much faster than the United States, so an acceleration is possible, and that if data center energy demand continues to grow, the construction of entirely new plants will increasingly appear. more like an attractive option.

These potentially long lead times partly explain why Microsoft is interested in small modular reactors, which should be faster and cheaper to build. But tech companies tend to focus on finding new sources of energy rather than improving the efficiency of their artificial intelligence operations, says Sasha Luccioni, head of AI and climate at Hugging Face, a company that develops tools to create applications using machine learning. “Regulation could be a way to encourage [great efficiency]starting with mandatory reporting and transparency for companies providing AI tools and services,” she says.

At the Carnegie Mellon University event, Pichai said work to improve the power consumption aspect of AI was still in its “early phases.” “We’re pre-training all of these models inefficiently, absolutely,” he said, but added that inference – actually asking an AI model to perform a task – could become “dramatically more efficient over time.” Google’s emissions in 2023 were 48% higher than their 2019 baseline, mainly due to increased data center energy consumption and supply chain emissions, putting increasing pressure on at risk Google’s goal of reaching net zero emissions by 2030. “The energy needs of AI are increasing now,” says Luccioni, but renewable or low-carbon energy used to power AI not keeping pace quickly enough.

For some, the prospect of the site of America’s most notorious nuclear disaster being used to fuel the AI ​​revolution could be worrying. But Allen emphasizes that the first reactor did not shut down because of operational problems. According to him, restarting the reactor will mainly be about ensuring that it is still in good working order and that there are enough trained personnel to operate it without problems.