AI and the odd alignment of capitalists and net-zero climate activists

On 20 September, Microsoft announced that it was collaborating with a utility named Constellation Energy to refurbish and restart one of the dormant units at the Three Mile Island nuclear plant — the facility that came close to melting down in 1979. Microsoft will provide the funds and buy all the energy produced for the first 20 years.

Put simply, this is a computer company getting into the nuclear power business. That’s a long way from Basic, DOS, Windows and Office.

Why? Because … well, AI.

Climate activists (and particularly the net-zero crowd) are presumably happy. Or at least not unhappy, considering the alternative sources of industrial-scale power that might otherwise be used. No one is pretending that a profit-oriented, publicly traded monolith like Microsoft is particularly interested in net zero, other than in a general sort of I-care-about-the-planet way.

But there is a coincidental alignment of interests at play here. Microsoft gets long-term, reliable and predictably priced power for its AI data centres and also gets to talk (a little disingenuously) about its commitment to clean energy. The net-zero crowd gets to, well, flatten their carbon projection curves a bit and feel a little more hopeful about beginning a non-fossil baseload future.

Read more: Tech titans tout breakthroughs in AI-driven climate models while their own emissions rise

Microsoft’s announcement did not come as a complete shock. After the tectonic disruption of OpenAI’s ChatGPT3 release in November 2022 and (as importantly) the arms race it set off, it was pretty clear that the compute needed to train and operate the foundation models was going to be demanding.

“Insatiable” is probably a more appropriate adjective. AI requires training, and even as training techniques become smarter and lighter, the rule is quite simple — the more compute available, the more muscular the AI. It is not only the training, it is also what happens after training — AI doing useful stuff requires it to “think”. Thinking needs compute and compute needs energy.

Of course, Microsoft is not the only player for whom AI is central to its future strategy. Bill Gates is separately funding a nuclear startup called Terrapower. Meta, Amazon, Anthropic, Google, Oracle, Nvidia, OpenAI and a gaggle of very serious non-US players from France to China are all considering, planning, investing in or building nuclear power plants.

What about Elon Musk?

Eagle-eyed readers will notice that I have left out Elon Musk’s xAI. It also needs power. However, he is in a spot of trouble with the US Environmental Protection Agency for having deployed massive gas turbines in Memphis, without permission, to run his data centres. So he may run into resistance.

xAI is also not flush with cash like some of the others and moving funds from one Musk company to another sails very close to the wind legally. Mind you, as I have said previously, don’t underestimate Musk; I suspect he too will want his own nuclear plant and will find a way.

Read more: Elon Musk and the entanglement of money, politics and power

In any event, while it may have made sense a couple of years ago for a tech giant to locate a data centre near a nuclear facility and to draw off some of its output, now that AI has changed everything, it makes more sense to own a nuclear facility, invest in one or at least to have usage rights secured.

Leaving aside the labyrinthine compliance requirements, nuclear plants have traditionally taken a long time to build (about five years) and have eye-watering construction costs ($10-billion or more). This meant that state institutions with healthy coffers have often been involved, with both the common good and national interest at stake. What has changed is that there is a new generation of more efficient reactor designs (Small Modular Reactors — SMRs) available, which are far more attractive across multiple axes.

They are cheaper (resulting in a 60% reduction of per-unit electricity costs compared to older reactors) and are faster to build (1.5 years), not to mention smaller and simpler to operate and maintain. This entirely changes the financial calculus. The big tech companies have done their spreadsheets and now, not only does it make more sense, but oil, gas and coal price volatility can be ignored — which amounts to a massive reduction in uncertainty.

One can imagine the scene — Netflix-style. A young nerdy engineer stares at a screen on which there is a busy Excel spreadsheet. He presses a key, the spreadsheet changes numbers. His eyes widen. He stands up, sprints across an open-plan office, down the stairs, across a parking lot, into another building and bursts into a meeting of senior executives. Everyone stares at him. He is panting. Finally, he stammers, “We need to build a nuclear plant.” He is greeted with a stunned silence.

Only two SMRs have been deployed globally (one in China, one in Russia), but there are others in various stages of planning or construction, and there are about 18 different types of SMR designs on offer from a wide spectrum of manufacturers, from giants like Hitachi and Westinghouse to startups like X-Energy, funded by Amazon, and Kairos, from which Google is buying seven reactors.

New technology obviously brings new risks, and, in the case of nuclear power, failure is not an option — nobody wants another Chernobyl. So there is caution bordering on nervousness, to be sure, but, because of the potentially huge competitive rewards, hesitation is also risky.

While many of the new SMRs will be deployed in sectors other than AI, it is largely AI that is unlocking a river of investment. Without affordable power there is no future for AI. Without first securing that affordable power, no investor can dream of historically unprecedented dividends — and perhaps even a better world.

There is somewhat of a virtuous circle here. The SMR industry cannot thrive without customers, and now they are lining up, with AI at the front of the line. This has led to more investment into affordable nuclear energy. Better reactors will follow. Lower prices from economies of scale. More sales, more investment, even better reactors. Until they are everywhere.

Of course, this poses another question — is a new era of affordable, clean nuclear power at last within humanity’s grasp? And, if so, who gets the credit? Is it the grandees of AI? Or the efforts of the climate change lobby? Or both?

Expedience makes for strange bedfellows. DM

Steven Boykey Sidley is a professor of practice at JBS, University of Johannesburg. His new book It’s Mine: How the Crypto Industry is Redefining Ownership is published by Maverick451 in SA and Legend Times Group in UK/EU. It’s available now.