Real-time AI plasma control advances stability in large tokamak reactors

Category: Diagnostics, Magnets, Tokamak, Vessels

Real-time AI plasma control advances stability in large tokamak reactors
AI breakthrough stabilises plasma in Japan’s JT‑60SA tokamak, marking a major step toward controllable, scalable fusion power
(Image courtesy of Fusion for Energy)

NTT and Japan’s Japan’s National Institutes for Quantum Science and Technology (QST) have cracked one of fusion energy’s toughest problems by controlling plasma in real time inside massive tokamak reactors.

Their AI system uses a Mixture of Experts model that combines multiple specialized neural networks to handle the chaotic, rapidly shifting conditions inside the JT-60SA tokamak. The breakthrough is in what it can predict. While conventional systems only manage the plasma boundary, this AI tracks internal current and pressure distribution across the entire plasma volume with accuracy within one centimetre, about 1% of the plasma’s total size.

That precision matters because it’s the internal dynamics that cause disruptions. Plasma instability can end a fusion reaction in milliseconds, and traditional physics simulations are too slow to respond. This AI processes live sensor data and adjusts magnetic field configurations fast enough to keep everything stable.

The technology was validated on real JT-60SA operations without relying on complex physics models, making it the first system to achieve this level of control at scale. For ITER Organization and future commercial reactors like DEMO, which will have fewer diagnostic sensors but demand even tighter stability margins, this approach could be transformative.

NTT and QST have now expanded their partnership to accelerate deployment of AI-driven plasma management across next generation fusion facilities. The work shows how merging telecommunications computing power with fusion physics can solve problems that have stalled progress for decades.

Fusion reactors need to maintain plasma at over 100 million degrees while keeping it stable for extended periods. Getting that right is the difference between a research experiment and a working power plant. This AI control system brings that goal measurably closer, turning fusion from a physics challenge into an engineering one that can be scaled for real energy production.

Stay ahead in the fusion revolution explore more breakthroughs from leading innovators in clean energy technology.