The real race to Artificial Intelligence is about cheap electricity. Here's how that's going.
This is a transcript, for the video found here:
Bullets:
Affordable electric power is the most important constraint facing companies in the AI space.
The cost of AI is converging to the cost of electricity, and the BRICS countries produce new sources of electric power far faster, and far cheaper, than Western countries.
This has profound implications for the coming decades, as China and Russia expand their fleets of nuclear plants.
Globally, nations seeking to develop their own economies will realize the deep advantages in Chinese power plant construction, and Russian uranium supply chains, and build their partnerships accordingly.
Report:
Good morning.
Much of the focus on the AI race is on semiconductor chips, and companies like Nvidia compared to Chinese companies like Huawei. But what is even more important is electricity. Chips can be packaged differently, or more of them can be used, for example. But the chokepoint for this entire industry is electric power, and the cost of that power. This is the head of Chat GPT, in his testimony to Congress. The cost of AI will converge to the cost of energy. Chips, automation, network optimization—all important, but none of it works without lots of electricity, and the energy constraints dictate everything else.
China produces more electric power than the United States, the European Union, and India combined. And China’s cost of that electric power is lower than almost anyone else in the world. What’s more, China brings new power sources online much faster, and much less expensively, than anyone else.
A viewer forwarded me this X post from Mr. Bertrand here, which blows apart the notion that our AI industry is going to be ahead of China’s, assuming that’s true even now. Because electricity is everything. In nuclear power, China is building nuclear plants at least 3 times cheaper—and even up to 7 times—compared to other countries. “Insane advantage”, he calls, and it’s hard to argue. In the UK, the Hinkley Point project will cost $43 billion for 3.2 gigawatts, and comparable plants in China will produce 12 GW for $27 billion, which makes them 6 times less expensive in China. And Hinkley isn’t even done, and engineers there are concerned that further delays will result in even higher costs – up to another billion pounds, that’s 1.3 billion USD higher.
The nuclear industry in the United States is even uglier from a cost perspective. The Vogtle nuclear plant is finally online, after 15 years. The original cost estimate was $14 billion, but the final price tag was over $30 billion. No other reactors are under construction in the United States. But for these two, the cost was $16.6 billion per gigawatt, which is 7 times more than the costs of Chinese nuclear power. China will build 10 reactors and 12 Gigawatts for $27 billion, while the United States spent $37 billion for 2 gigawatts.
China is building nuclear power faster—2 to 4 times faster—and up to 7 times cheaper. “Industry is energy transformed”, and these lower energy costs ripple through economies, compounding along the way, and Chinese companies have enormous competitive advantages over everyone else, likely to last generations.
What does that mean for countries outside China, and outside the US? For the governments across the world, looking to modernize and industrialize their own economies, who will they look to? Who will they buy from? The suppliers for this Vogtle plant in Georgia? Or one of the contractors who still can’t get Hinkley Point done on time, let alone on budget? Question answers itself. Chinese companies will get those construction jobs, and build electric utilities all across the developing world, and enjoy all the advantages that go with that—hundreds of millions of new customers.
Back to the AI race then. Nobody in the US or Europe seems much interested in building new nuclear power plants for household consumers, or to lower energy costs for other manufacturing. But now it’s an AI race with China, and “America’s AI future depends on nuclear energy.” Winning the AI race means we need a lot more of it. “The United States will win the global race”, he says. But first, we need to “unleash our energy dominance.”
The MIT Technology Review gets deeper into it: it’s the tech giants who need lots more electricity, but new power plants take a long time. In the case of nuclear plants, China builds them much faster, much more affordably—and—let’s not forget, China has a much better supply chain for uranium than any of these nuclear plants in the United States or the UK are going to have.
Here is where the US gets its reactor fuel from now, and we can see the problem right away: Russia is a big piece of the pie, and is much friendlier with China than with us, and Russian uranium is also powering a lot of the power plants here in China. But our big tech companies and their data centers are in the US, so that’s where they need new power sources, so they’re signing long-term deals with utility companies.
This is a good example of the problem, though. Amazon needs more power right now, immediately. And it takes a long time to build new plants. So Amazon signed a contract with this utility in Pennsylvania, for nuclear energy for Amazon Web and AI data centers, through 2042. There are “additional plans to explore small modular reactor” technology—so eventually, maybe, they will build a new reactor in Pennsylvania to meet this new enormous demand on the local system.
But they say that it’s not a problem for existing customers. “With a large load customer” —Amazon in this case—"now on the grid, it will result in lower energy bills for everyone else” already on the grid. I’m very curious how that’s going to work, and I can’t wait to come back and look carefully at electric bills in Susquehanna, Pennsylvania, to see just how much households there are saving on their electric bills, now that Amazon is sucking down massive volumes of electricity to power their new data centers. Color me deeply skeptical about any of that, but let’s keep an open mind for now.
So here is the problem. The United States is producing just slightly more electricity today, than we were 20 years ago, in 2005. In the European Union and Japan, it’s slightly less. The lines are declining, slightly, over time. China is parabolic. If it’s true that the key technologies for the rest of this century will be driven by intelligent networks and high-speed computing, and those are driven in turn by access to affordable energy, it looks like this race is already over.
Resources and links:
Where the U.S. Gets Its Enriched Uranium
https://elements.visualcapitalist.com/where-the-u-s-gets-its-enriched-uranium/
China Generated More Electricity in 2024 Than the U.S., EU, & India Combined
https://www.voronoiapp.com/energy/-China-Generated-More-Electricity-in-2024-Than-the-US-EU--India-Combined-5260
X, Arnaud Bertrand
https://x.com/rnaudbertrand/status/1924283058519515485
Reuters, EDF's UK Hinkley Point nuclear plant start date delayed again, costs mount
FORM 10-Q, The Southern Company
EIA, Plant Vogtle Unit 4 begins commercial operation
https://www.eia.gov/todayinenergy/detail.php?id=61963
X, Sam Altman says "eventually, the cost of AI will converge to the cost of energy."
https://x.com/vitrupo/status/1920883714927558872
OpenAI CEO Sam Altman testifies on AI competition before Senate committee — 5/8/2025
Amazon signs nuclear energy deal to power AI data centers
https://cointelegraph.com/news/amazon-nuclear-energy-deal-power-ai-centers
MIT Technology Review, Can nuclear power really fuel the rise of AI?
https://www.technologyreview.com/2025/05/20/1116339/ai-nuclear-power-energy-reactors/
SMR stocks skyrocket as Trump backs nuclear tech and AI demand soars
https://www.capacitymedia.com/article/smr-stocks-trump-nuclear-tech-ai
America’s AI future hinges on nuclear energy
https://www.utilitydive.com/news/americas-ai-future-hinges-on-nuclear-energy/747395/
Cost of Electricity by Country 2025
https://worldpopulationreview.com/country-rankings/cost-of-electricity-by-country