Within the next five years, companies in the artificial intelligence (AI) game are going to need enough power to light up New York City 17 times over, and Goldman Sachs tells us that less than 10% of the needed power will be available by 2030.
Companies such as Microsoft, NVIDIA, Intel, Amazon.com, and many others the world over – including likely several in China that we know very little about – are investing at a rate of knots in trying to meet this need. I’ve written about the amount of power that AI will need here.
The AI giants are spending billions, if not trillions, of dollars in a bid to get there, although “there” seems to be a moving target because research into power needs is constantly being done.
Exactly how much is being spent is anyone’s guess because, as the Organisation for Economic Co-operation and Development (OECD) points out, there is an absolute paucity of information.
In a September paper, Advancing the measurement of investments in artificial intelligence, the organisation points to the fact that numbers, even when they are made public, lack transparency, comparability, or verification.
Some numbers are actual spend; others are soft commitments, budget allocations, or multi-year commitments.
Not only is AI technology advancing at a massive pace, but nuclear energy is moving as fast as possible to keep up.
AI investment information often conflates different purposes and project scope, sometimes only including the value of research and development, while other disclosures will be for specifics such as a chip factory or a power plant. Not only that, but there are assorted methodologies for arriving at what should be a simple number: capex.
What complicates the picture further is that the OECD’s report doesn’t have a category for “power infrastructure”. Under equipment, where this should rightly belong, there’s computer hardware, software and databases, and telecommunications equipment.
While we may not know exactly how much is being spent on fuelling AI’s power needs, we do know that the choice of source is generally nuclear. Again, this is based on public statements from the big companies.
A quick whip-around and calculation process courtesy of AI indicates that the larger companies that have provided figures are, combined, spending around $3 billion a year on nuclear power.
Nuclear is the most obvious choice as a power source. It’s available around the clock to provide constant power and can be built to serve a particular company’s need instead of having to be plugged into the grid. And it’s fairly clean as far as forms of baseload energy go.
NVIDIA, for example, is expanding its nuclear energy involvement through investments and technology partnerships. It is also backing a project in Japan to build a data centre near existing nuclear generation to secure clean, reliable power.
Google is teaming up with Kairos Power to investigate using small modular reactors (SMRs) to run its AI data centres. Microsoft has struck a deal with Constellation Energy, which runs one of the biggest nuclear reactor fleets, to lock in clean power for its operations.
Meanwhile, start-ups like Oklo, backed by OpenAI CEO Sam Altman, and TerraPower, founded by Bill Gates and partially funded by NVIDIA, are working on next-gen nuclear tech, including SMRs, that could one day keep data centres humming. NuScale Power is also in the game, developing these reactors for clean electricity generation.
The next-gen nuclear reactors, being modular in design, enable them to be plugged into renewable energy, and they are low carbon. Not only is AI technology advancing at a massive pace, but nuclear energy is moving as fast as possible to keep up.
What’s also important to note, however, is that there’s a vast amount of other work being done by the AI giants themselves to help mitigate any potential impact on the environment. They are, in fact, using AI to find more environmentally-friendly ways of using AI.
Nuclear certainly has a chequered past – with Chernobyl being a prime example. As long as it’s used sensibly, nuclear currently looks like the best possible solution to power up a technology that will inevitably and inextricably become more a part of our lives.
If we can safely find a way to power technology that will help us be better stewards of planet Earth, I’m all for it.
Share