Researchers are increasingly using artificial intelligence technologies to help solve some of the biggest challenges facing the energy sector including, ironically, the massive surge in electricity demand caused by large language models themselves. The current and expected rise in energy consumption from AI data centers is driving a wave of investment into advanced energy alternatives capable of delivering huge amounts of reliable electricity without major greenhouse gas emissions.

Among the technologies being viewed as a potential silver bullet is nuclear fusion, which has made major progress in laboratories in recent years, partly thanks to AI tools.

In this context, scientists at Ames National Laboratory in Ames, Iowa, are developing a specialized AI tool designed to model how different materials behave inside nuclear fusion systems, with the goal of improving research methods and making both the scientific process and fusion systems more efficient.

The tool, known as DuctGPT, was developed based on an earlier model from the National Institute of Standards and Technology called AtomGPT. The Duct version combines large language models with physics-based simulations to identify materials capable of withstanding the harsh environment inside a nuclear fusion reactor.

Nuclear fusion the same process that powers the sun relies on extremely high temperatures that most materials cannot tolerate. In addition to resisting temperatures reaching thousands, millions, or even hundreds of millions of degrees, these materials must also remain sufficiently ductile to allow practical manufacturing.

Finding the right material remains one of the biggest obstacles preventing commercial nuclear fusion, while also representing a massive opportunity for the scientific team capable of solving the challenge, potentially unlocking a near-unlimited source of clean energy. Identifying such materials requires exploring and modeling an enormous range of possible alloy combinations.

This type of project is particularly well suited to large language models. In a Financial Times report published last year titled How AI Could Deliver More Energy Than It Consumes, the newspaper noted that discovering new materials, catalysts, or processes capable of producing energy more efficiently is exactly the type of needle in a haystack problem where AI excels.

The new tool is already showing highly promising results in fusion research. The team behind DuctGPT said the time required to discover new alloys for fusion experiments has been reduced from months of research work to just a few hours.

Scientist Prashant Singh from Ames Laboratory said: Now when you ask the system to design a material for nuclear fusion with the critical properties required for reactors, it provides the appropriate elemental compositions along with their expected characteristics.

Although DuctGPT is one of the newest and most promising applications of large language models in nuclear energy research, it is not the only one. Another tool called Diag2Diag is being used to help monitor and control plasma behavior in fusion experiments, specifically to prevent a phenomenon known as Edge Localized Mode or ELM.

This instability rapidly erodes the materials surrounding the plasma, creating major challenges in massive and expensive projects such as Europes ITER reactor and Chinas EAST reactor.

In the United Kingdom, the British government is investing 45 million, or roughly $60 million, to build an AI-powered supercomputer at the UK Atomic Energy Authority campus in Oxfordshire.

The computer, called Sunrise, is expected to begin operations next month. According to a report published by Interesting Engineering in March, officials say the system will help scientists better understand the highly complex physics inside fusion reactors.

The report added that combining advanced computing with AI models could allow researchers to test ideas virtually before building extremely expensive experimental systems.

Together, these tools could dramatically accelerate nuclear fusion research at a time when the need for breakthroughs has become more urgent than ever. While investing in unproven technologies remains a high-risk bet, nuclear fusion now appears closer to reality than at any previous point, as scientific breakthroughs accelerate, competition intensifies, and major technology companies move aggressively into the sector.

The enormous and unprecedented energy demand created by artificial intelligence has become so large that the tools needed to address it may also need to be unprecedented which helps explain why AI solutions themselves may ultimately become the only way to solve the problems AI created in the first place.