DOE Picks Commonwealth Fusion to Build an AI Digital Twin of SPARC

The U.S. Department of Energy has selected Commonwealth Fusion Systems (CFS) to lead a project that will build an AI-enabled digital twin of SPARC, the company's demonstration tokamak in Devens, Massachusetts. The award is part of DOE's Genesis Mission, the AI-for-science initiative launched in November 2025. Public details are thin, but the structure of the project says a lot about where AI is heading in hardware development.

What Was Announced

• The program: DOE announced 12 new Phase II awards in the Genesis Mission, totaling $159 million through its Office of Science, plus six new Phase I awards. Phase II now covers 14 projects.
• CFS's role: It is the only private company leading a Phase II project, and the only fusion-focused one. The project is titled "Accelerating the path to commercial fusion energy via an AI-enabled digital twin platform for SPARC."
• The work: Software linking SPARC to DOE supercomputers, training datasets and AI surrogate models, reinforcement-learning and agentic AI as an operations "co-pilot," and visualization pipelines.
• The partners: Ten, including Lawrence Livermore, Oak Ridge and Princeton Plasma Physics laboratories, SLAC, Carnegie Mellon, MIT, Google DeepMind, NVIDIA, Synopsys and General Atomics.
• What was not disclosed: The per-project award amount for CFS and the project's end date. We relied on the Securities.io report of the announcement and could not retrieve DOE's own page.

What a Digital Twin Is For

A digital twin is a simulation model kept in sync with a real machine. Physics simulations of plasma are accurate but slow. An AI surrogate model learns from those simulations and from measured data, and can then answer "what happens if" questions far faster, though only as well as its training data. For SPARC, the hope is to help plan experiments, tune operations and spot problems before they cost time on a machine that, as our fusion milestones guide notes, was about three-quarters built as of April 2026 and is targeting net energy in 2027.

CFS says the work builds on earlier digital-twin efforts with NVIDIA and Siemens. A reinforcement-learning "co-pilot" is the most ambitious piece: an AI system that suggests or helps control how the machine is run. How much authority such a system would have over a tokamak is not stated in the announcement.

AI and Fusion, Side by Side

This section is analysis, not reporting. The same week, General Fusion and UKAEA announced a measured temperature milestone for a very different design, which we cover in our piece on the LM26 result. The two stories are a useful contrast. One is a physical measurement with a new diagnostic. The other is an effort to use AI to shorten the loop between experiment and understanding. Fusion has both problems: it needs hardware that works and the ability to learn from each run quickly. A caution is warranted. A digital twin can speed up learning about a machine that exists, but it cannot show that SPARC will reach net energy. That test is a physical one, still ahead.

It also fits a wider pattern of AI companies and power. Our coverage of nuclear power and the AI boom shows demand running in one direction; this award shows AI tools running in the other, toward building the supply.

Why It Matters

• It is federal validation for a private fusion company. CFS is the only private firm leading a Phase II project.
• The project is about speed. AI surrogates aim to cut the time needed to model and tune a machine.
• The key numbers are missing. No per-project dollar amount or end date has been published.
• Net energy is still a physical test. A digital twin does not replace SPARC producing more energy than it consumes.