M

uch has been written about the rise of artificial intelligence, not just as a technological revolution, but also as a huge new source of energy demand. Training large language models (LLMs) consumes an eye-popping amount of energy. To train GPT-3 required around 1,270 megawatt hours of electricity, like the power demands of 120 typical American homes for a whole year. The big data centers where AI searches are handled are gobbling up more and more energy. By 2030, data centers are estimated to account for 21 percent of the electricity demand in the United States.

 

An LLM query is 10 times as power-thirsty as a traditional Google search — if every Google search was rerouted through an AI engine, the added energy demand would be equivalent to the power consumption of Ireland.

 

All of this is true, and all of it is hotly debated. But much less attention has been given to the flip side of this equation—AI’s potential to solve some of the most difficult problems in energy generation today.

 

Revolutionary Research for Our Energy Systems

 

From optimizing nuclear fission reactions to aiding the stabilization of plasma in nuclear fusion, AI is already being deployed in cutting-edge research that could revolutionize our energy systems.

 

Nuclear fusion, the process of fusing atomic nuclei to release energy, has long been considered the ‘holy grail’ of energy generation due to its potential for limitless, clean power. While it has been a scientific challenge for decades, recent breakthroughs are bringing fusion energy closer to reality, with AI playing a critical role in accelerating this process.

 

To create a viable commercial fusion reactor, you have to find a way to control plasma —the exotic fourth state of matter— at temperatures exceeding 100 million degrees. That’s hotter than the core of the sun. Managing these extreme conditions has been one of the greatest scientific challenges of our age, and a major reason why progress towards commercial applications has been so slow. The mathematics involved in modeling the behavior of plasma are dauntingly complex — and an ideal candidate for AI applications.

 

Together with high-performance computing, AI techniques are now being used to create predictive models with the potential to stabilize plasma in real time. Researchers at the Princeton Plasma Physics Laboratory, are using machine learning to anticipate plasma instabilities and adjust controls within milliseconds, making fusion reactions far more stable and sustainable. As Steven Cowley, the director of the Princeton lab, put it recently in the Washington Post, “AI is set to revolutionize fusion, reducing development timelines from decades to mere years by enabling faster design iteration and real-time plasma control.”

 

 

 

Others in rival labs are trying to beat them to the punch, creating the kind of scientific race that has often heralded major breakthroughs in the past.

 

Meanwhile, other AI researchers are helping to fast-track the design of new generations of much safer, much more efficient nuclear fusion reactors. Traditional computational models took months to analyze potential reactor designs, but fast AI-driven models can now evaluate competing designs in hours, allowing researchers to explore hundreds of billions of reactor configurations in their search for the best solution. This speed enables more efficient iteration, potentially bringing fusion closer to commercial reality within the next few decades.

 

Fusion is the sexiest area of AI nuclear research, but certainly not the only one.

The AI revolution could prove similarly transformative for traditional fission-based nuclear power. As you’d expect, researchers are using AI models to plan new generations of much safer, much more efficient fission reactors. What you may not know is that AI-technology can improve existing fission reactors as well. AI-powered predictive maintenance systems can analyze vast amounts of sensor data to forecast when critical reactor components will need repair or replacement. By catching issues early, AI minimizes downtime and reduces the risk of catastrophic failures.

 

And AI-applications are also enhancing the efficiency of nuclear reactors by adjusting control settings to optimize fuel use and reactor output. This could extend the lifespan of existing reactors and reduce fuel consumption, making nuclear fission an even more viable low-carbon energy source.

 

Great Strides in Sustainable and Transition Solutions

 

While nuclear energy presents exciting possibilities, AI is also playing a significant role in established renewables. AI models are being used to optimize the placement and operation of solar panels and wind turbines on the basis of weather and climate data. In hydropower, AI is improving water flow management, predicting fluctuations to ensure more consistent power generation. In geothermal energy, AI is analyzing geological data to help identify new underground heat sources, accelerating exploration and increasing the efficiency of geothermal plants.

 

And AI is also advancing the field of carbon capture and storage (CCS). Machine learning algorithms excel at analyzing complex geological data to determine the best locations for carbon sequestration, improving both the efficiency and safety of this critical technology for mitigating carbon emissions.

 

 

 

There are all sorts of ways AI-powered research will transform power generation. Fusion stands out as AI’s potential ‘killer app’—a domain with extraordinary upside potential where AI’s capabilities to manage extreme complexities are indispensable. AI enables real-time plasma management, rapid design iteration, and unprecedented precision in reactor control. If AI succeeds in making fusion commercially viable, all the hand-wringing about how power-hungry Chat GPT is will be looked back on as a quaint side-show.

 

Of course, it’s too early to say whether AI’s additional energy demand will ultimately outstrip the gains it helps create in energy efficiency and generation. But it’s not too early to say that, in focusing almost exclusively on the energy demands of AI and overlooking its potential to revolutionize power generation, our public conversation about energy and AI has become badly lopsided. It’s time to rebalance this debate. Because AI draws a lot of power. But it might end up generating orders of magnitude more.