Great Sky and research partners demonstrated a superconducting AI chip processing fusion-plasma data at one million samples a second. The work points towards faster AI hardware for fusion research, though it is a demonstration, not a deployed reactor-control system.
Watch Desk analysis
What happened
The AI Insider reports that the chip handled data from the DIII-D tokamak for anomaly classification and disruption prediction. A third task used simulated data from Google DeepMind’s TORAX simulator to predict sawtooth instabilities, which matter to tokamak operation.
The chip uses superconducting circuitry and optical communication to implement analog neural networks, rather than conventional silicon transistors. Great Sky says it plans to demonstrate a next-generation chip with a neural network ten times larger in December.
Why it matters
Fusion plasmas change quickly, so systems that can process measurements at high speed could help researchers explore real-time monitoring and control. This demonstration shows the chip keeping up with a fast stream of tokamak data on selected tasks. It does not establish that it can control a working fusion reactor, or that the approach is ready for practical deployment.
The useful point is that AI hardware is not only about making familiar chips faster. Great Sky is testing a different computing approach against a demanding scientific problem. Fusion research, it turns out, is a rather unforgiving place to discover whether your chip can keep up.
Our read
A million samples a second is an arresting number; the more meaningful detail is that the chip processed measured tokamak data for two tasks and simulated data for a third. That makes this a tangible research milestone, not a claim that fusion control has been solved. The next-generation demonstration will show whether the approach scales beyond this chip and these selected tasks.
What to watch
- Whether Great Sky demonstrates the planned ten-times-larger neural network in December.
- How the team evaluates accuracy and reliability on measured and simulated data.
- Whether future work tests the chip in a real-time control setting.
Discussion spark: What evidence would convince you that a specialist AI chip is ready to move from processing fusion data to helping control a reactor?
Sources and evidence
- Great Sky’s High-Speed AI Chips Analyze Fusion Plasmas (19 October 2026, 11:50 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.