Silicon Valley finds its silver bullet in a desperate race for energy
Tech giants bet on a nuclear revival to power the supercomputer revolution
Creating superintelligence requires a few simple ingredients: a near-infinite trove of data, an army of tech experts, billions of dollars worth of advanced microchips – and a gigantic amount of electricity.
As Silicon Valley spends enormous sums in a race to build artificial intelligence, tech giants are hitting a bottleneck. Without an abundant supply of energy, they will be unable to build the colossal supercomputing hubs they believe will host the next generation of AI systems.
Mark Zuckerberg, Meta’s chief executive, has assembled what is believed to be the world’s biggest arsenal of Nvidia’s latest AI chips. He told a podcast last month that ultimately using them all in one place could require a gigawatt of power, potentially requiring a dedicated nuclear supply. “A gigawatt [data centre] would be the size of a meaningful nuclear power plant,” he said.
Zuckerberg’s comparison is not simply flippant. The desperate race for energy is triggering a wave of interest in nuclear power among the world’s tech barons. Venture capitalists are becoming nuclear boosters, betting that it will be the power source to keep the AI boom running.
As so often happens in Silicon Valley, support for nuclear power is not couched in terms of the profits it could shower down on California but cloaked in quasi-philosophical utopian language.
Marc Andreessen, the prominent tech investor, has become a leading figure in what has been dubbed the “effective accelerationist” movement, which advocates for rapid advancements in areas such as nuclear power and AI with the promise that it will lead to a state of abundance. In a “techno-optimist manifesto” published last year he called nuclear power a “silver bullet”.
Elon Musk has said shutting down nuclear plants is “anti-human”. Bill Gates and Jeff Bezos, two of the world’s richest men, have invested in nuclear start-ups ranging from small modular reactors to fusion.
Sam Altman, the chief executive of ChatGPT-developer OpenAI, has been a nuclear supporter for years. In 2015, Altman invested in and agreed to chair Oklo, a company building futuristic microreactors.
Altman is also the biggest outside investor in Helion, a company seeking to crack nuclear fusion. Fusion is the process that powers the sun and replicating it sustainably on earth is seen as the holy grail of clean nuclear power. Altman has said the breakthrough will be crucial to meeting the demands of superhuman AI.
Last year, Microsoft agreed the world’s first fusion purchase from Helion, saying supplies could arrive as soon as 2028. In March, Amazon acquired a nuclear-powered data centre in Pennsylvania. Meanwhile, DeepMind, Google’s London-based AI lab, is working on technology that would use AI to better control fusion reactions.
Last Energy, a Washington DC-based company developing small modular reactors, says half of its orders now come from companies setting up data centres, compared to a quarter a couple of years ago. Last Energy has signed multiple deals to provide new data centres in the UK, although it is still waiting for regulatory approval.
“We’ve seen a flood of interest as those pain points become more clear,” says Michael Crabb, Last Energy’s vice president of commercial.
The biggest “pain point” is that AI will require huge amounts of energy from somewhere. The International Energy Agency estimates that the amount of electricity used by data centres and AI could double by 2026, at which point they could use as much energy as Japan. A single ChatGPT query has been estimated as requiring as much electricity as 15 Google searches.
“This tech is an energy hog,” says Ray Rothrock, a Silicon Valley venture capitalist who has invested in a string of nuclear start-ups. “The big issue with nuclear today is lack of demand. But now, the demand is increasing because of AI.”
While tech companies have made large investments in wind and solar, their intermittency may not be well suited to data centres that run 24 hours a day. Nuclear has no such problem.
Last Energy’s management team and investor list highlight the increasingly close ties between the nuclear industry and the tech sector. While Crabb is an energy industry veteran, the company’s founder Bret Kugelmass was previously a drone entrepreneur. Investors include PayPal co-founder Luke Nosek.
Crabb says the Silicon Valley types bring “that innovative, problem solving mindset [and] they see nuclear as the obvious answer” to the question of how to power AI.
This boom in nuclear interest is being felt far beyond the Californian borders, with Canadian uranium miner Cameco saying last week that the tech sector’s interest had triggered a surge in demand for its reactor fuel.
“Gone are the days of rolling out new technology without worrying about future potential runaway environmental impacts… nuclear is the clear winner,” Cameco’s chief executive Timothy Gitzel said.
The almost-limitless piles of cash AI companies have to spend helps. OpenAI and its major investor Microsoft are reportedly developing plans for the world’s biggest supercomputer, a $100bn (£79.66bn) project codenamed Stargate. Analysts at Morgan Stanley speculated last week that this would be powered by several nuclear plants.
Altman, meanwhile, has been telling investors that he needs up to $7 trillion to achieve his AI dreams, a budget that would include setting up his own energy supplies.
He has claimed that AI will be the most important invention in human history, but says that without a nuclear revival, we will not be able to power it.
“There’s a lot of parts that are hard,” he said recently, adding: “Energy is the hardest part.”