A team of international partners, including NVIDIA and General Atomics, has built an AI-enabled digital twin for a fusion reactor to tackle the challenge of making fusion energy work on Earth. By using NVIDIA technology and AI surrogate models, researchers can interact with the reactor virtually and accelerate the path to practical fusion energy.

Traditionally, simulating plasma behavior on supercomputers takes weeks, but General Atomics is now using AI surrogate models to predict plasma behavior in seconds, reducing the risk of damage and speeding up research. These models, running on NVIDIA GPUs, deliver accurate predictions faster than physics-based simulations, helping control fusion reactors.

NVIDIA and General Atomics are developing a fully interactive digital twin of the DIII-D reactor inside NVIDIA Omniverse, powered by NVIDIA RTX PRO Servers and NVIDIA DGX Spark. This virtual reactor fuses sensor data, simulations, engineering models, and AI surrogate models, allowing scientists to test ideas and run scenarios without touching the real machine.

This innovative approach to fusion research shifts from a pure physics challenge to one powered by computing and smart algorithms. By providing near-real-time interactive answers, the digital twin serves as a “fusion accelerator” to test new ideas, optimize reactor designs, and advance commercial fusion energy faster.

Read more at NVIDIA: NVIDIA, General Atomics Advance Commercial Fusion Energy