How AI and High Performance Computing are Revolutionising Fusion Energy
By Jack Moore, Analyst at Fusion Advisory Services
Fusion Advisory Services and the Clean Air Task Force have released a report at COP29 that explores the transformative role of high-performance computing (HPC) and artificial intelligence (AI) tools in advancing fusion energy. The report was done in conjunction with EPRI, the Electric Power Research Institute.
Here are three key insights from the report, which was released on 15th November 2024.
1. High performance computing and AI are accelerating the iteration process for fusion design
Designing a fusion energy machine is a time-consuming, iterative process. Each design cycle tests and refines parameters, which has in the past required extensive physical experiments to validate proposed alterations. However, recent advances in HPC and AI are reshaping this iterative cycle, making it significantly faster and more efficient.
Researchers can now perform highly detailed simulations of a machine’s parameters “in-silico” (or virtually) eliminating the need for physical prototypes at many stages. These simulations model the intricate interactions within a machine’s design with remarkable precision and speed, allowing researchers to adjust variables and observe potential outcomes almost instantaneously.
AI-driven “surrogate models” take this efficiency even further. These models act as simplified versions of complex simulations, providing highly accurate results with a fraction of the computational resources and time. By streamlining the computational load required for each test cycle, surrogate models enable significantly reduced run times, allowing what may have once taken weeks to be simulated to be accomplished in hours. This dramatically reduces development timelines.
Additionally, machine learning techniques like Physics-Informed Neural Networks (PINNs) allow researchers to efficiently build physical constraints into their model. This allows for the extrapolation of experimental data to future parameter spaces that have yet to be tested, helping to ensure the validity of the simulations.
Over the course of years, the cumulative effect of time saved through accelerated design cycles adds up. This could play a critical role in the global transition to clean energy by making fusion power commercially viable sooner than anticipated, helping address the urgent need for emissions-free, firm energy sources.
2. High performance computing and AI are optimising fusion plant operations
HPC and AI are also improving how fusion plants will operate in real-time. Across various systems in fusion machines, these techniques are being applied both in optimising machine operations.
For example, conventional approaches have typically relied on “disruption mitigation”, using techniques such as shattered pellet injections to minimise damage when a plasma disruption happened, as the complexity of the conditions that trigger them were beyond the capabilities of traditional modeling tools. Today, AI-driven models using deep learning and reinforcement learning techniques can identify markers of instability and track plasma conditions. Tested on several major tokamaks globally, these prediction models have demonstrated the ability to forecast instabilities before disruptions occur, providing sufficient time to adjust plasma parameters and maintain stability.
Additionally, AI is playing a critical role in predictive maintenance by identifying patterns that indicate when components are likely to need replacement before they actually fail. This proactive approach minimises downtime and enhances operational efficiency, which is essential for keeping fusion plants online and cost-effective. By analysing real-time sensor data, AI models can detect early signs of component degradation, preventing small faults from escalating into larger system failures.
AI also can help monitor for “quenches” in superconducting magnets. These are localised heating events that cause sections of superconducting magnets to lose their superconducting properties and become resistive. Quenches lead to a sudden temperature increase and the release of stored magnetic energy as heat, which can result in severe damage to the superconducting coils, potential system failures, or even structural damage to the fusion device. AI-driven quench detection systems significantly enhance monitoring capabilities by using data from Fiber Bragg Grating (FBG) sensors, which measure subtle shifts in wavelengths to track localised temperature changes. Advanced hot spot algorithms, built on this FBG data, increase the sensitivity of quench detection and provide early warnings, This increased detection capability safeguards the integrity of the superconducting magnets in a device and helps prevent costly repairs.
3. Fusion and AI create a symbiotic relationship
The report also highlights a growing synergy between fusion energy, AI, and high performance computing. Data centers—many built to support the proliferation of AI models—are increasing total energy demand. In an era increasingly focused on cutting carbon emissions from traditional sources, it’s clear that new energy demands, such as those from data centers, can’t rely on fossil fuels. Instead, they need zero-carbon energy solutions to sustainably support their growth.
Fusion energy could offer a compelling solution for these demands. With the potential to deliver firm, emissions-free power, fusion aligns well with the constant, large-scale power needs of hyperscalers operating data centers. Fusion’s safe operational profile could translate to fewer regulatory hurdles and greater flexibility in siting facilities closer to urban areas where many data centers are located. As fusion plants begin to come online, data centers could become natural consumers of this reliable, clean energy source.
The use of AI and high performance computing to fast-track fusion commercialisation therefore represents a strategic investment in sustainable energy. A symbiotic relationship is emerging: while AI and HPC are driving fusion development today, fusion energy may fuel the continued growth of AI tomorrow, providing the emissions-free, always-on power that modern computing infrastructure increasingly demands.
In releasing this report at COP29, Fusion Advisory Services and the Clean Air Task Force hope to highlight how these advancements driven by high performance computing and AI are accelerating the commercialisation of fusion energy to be a clean energy solution on a climate-relevant timeline.
Please give the report a read!