• Opening the door for nuclear energy
  • Is AI the key to making fusion energy a practical reality?
  • A crucial milestone for efficient energy production

In these pages yesterday my fellow editor Sam Volkering asked whether nuclear might just be the answer to artificial intelligence’s (AI’s) problems.

In the article he was referring to the enormous amount of computing power required by AI, which is already behind stunning growth for the chips and graphics cards developed by companies such as Nvidia.

As Sam explained, the hardware and software needed to power AI applications will require a lot of electricity. According to some estimates, AI data centre racks require seven times more power than traditional types (which themselves already consume a huge amount of electricity).

Certainly, as generative AI tools like ChatGPT improve in functionality and increase in popularity, energy consumption is forecast to increase significantly over the coming years. This means we will likely need a lot more electricity in the future, opening the door for nuclear energy to help fuel the megatrend.

Sam said small modular reactors (SMRs) – i.e. nuclear fission reactors that are a lot smaller than conventional ones – could see big demand as a result, and he may very well be right.

(Saying that, I can’t lie: I’m becoming more and more sceptical about the future of SMRs, at least in their current proposed designs. Although SMRs are thought to be less expensive to build than traditional nuclear power plants because of their smaller size, their size might actually work against them economically, with their smaller power output meaning less revenue for the owning utility but with construction costs that not proportionately smaller. Last year NuScale cancelled its flagship SMR project in Utah amid such worries.)

But another form of nuclear might also be needed to meet the vast energy requirements of future AI: nuclear fusion.

In January this year, Sam Altman, the creator of ChatGPT, said a breakthrough in nuclear fusion would be required to meet the unexpected energy demands of AI.

“There’s no way to get there without a breakthrough,” Altman said at a Bloomberg event at the World Economic Forum’s annual meeting in Davos.

“It motivates us to go invest more in fusion.”

So it’s not without irony that, just one month later, scientists pursuing fusion energy say they have made a key breakthrough in overcoming one of fusion’s biggest challenges to date – by using AI.

As you might know, nuclear fusion is a process that creates energy by mimicking the natural reactions that occur within the Sun. It involves smashing together two atoms with so much force that they fuse into a single, larger atom, releasing huge amounts of energy in the process.

In fact, nuclear fusion has the potential to provide near-limitless energy, leading some scientists to describe it as the “holy grail” of clean energy.

However, there are many obstacles to such claims – including but not limited to generating more energy than it takes to power the reactors, developing reactor-proof building materials, keeping the reactor free from impurities and restraining that fuel within it.

But that last problem, at least, could now be solved.

Engineers from Princeton University and its Princeton Plasma Physics Laboratory have developed an AI model that predicts, and then figures out how to avoid, plasma – the hot, charged state of matter composed of free electrons and atomic nuclei that fuels fusion reactions – becoming unstable and escaping the strong magnetic fields that hold it inside certain donut-shaped reactors called tokamaks.

If the plasma escapes the magnets’ clutches, the reaction ends, so it’s enormously exciting that the engineers’ innovative AI model can predict and prevent these disruptive events happening in real time.

The team found that their AI controller could forecast potential plasma tearing up to 300 milliseconds in advance, plenty of time for it to then change certain operating parameters to make sustained fusion reactions more stable.

The team published its findings last month in the journal Nature.

Be clear: this AI application not only represents a significant technical achievement, paving the way for sustained high-power fusion reactions, a crucial milestone for efficient energy production, but also illustrates the potential for AI to play a pivotal role in making fusion energy a practical reality.

“The experiments provide a foundation for using AI to solve a broad range of plasma instabilities, which have long hindered fusion energy,” a Princeton spokesperson said.

The findings are “definitely” a step forward for nuclear fusion, said Egemen Kolemen, a professor of mechanical and aerospace engineering at Princeton University and an author of the study.

At this stage, all the authors of the study describe their work as proof-of-concept and write in their paper that it’s still very much in the early stages of fine-tuning. They are hopeful, however, that it could be eventually applied to other reactors, also to optimise the reaction or harvest the energy from it.

This is just the latest breakthrough in fusion.

Earlier this year scientists and engineers near Oxford set a new nuclear fusion energy record, sustaining 69 megajoules of fusion energy for five seconds – enough to power roughly 12,000 households for the same amount of time – using just 0.2 milligrams of fuel.

Although that experiment still used more energy as input than it generated, another team in California managed to produce a net amount of fusion energy in December 2022. They have now replicated this process three times since.

Of course, despite the promising progress, fusion energy remains a long way from becoming commercially available. But it’s becoming increasingly clear that AI will play a large role in its development, just as nuclear energy – whatever its form – will ultimately serve AI’s needs.

Until next time,

James Allen
Contributing Editor, Fortune & Freedom