US energy company AMPERA has announced a multi-year partnership worth about $9 million, or £7 million, with Imperial College London. The aim is to develop NeuralTwin, a physics-based digital twin platform for advanced energy systems, including nuclear microreactors. The announcement was made on 27 October 2026.
A digital twin is a virtual representation of a plant, updated in real time. NeuralTwin combines high-fidelity physics simulations with AI-based surrogate models. These models are trained to predict the effects of design choices from the laws of physics, without having to recompute everything in full at every step. Traditional digital twins rely on predefined models and computationally intensive simulations. Here, the heavy simulations are paired with fast, trained shortcuts.
According to the two partners, the platform will speed up design and improve performance and safety. It will also support licensing activities and, over time, make plant operation more autonomous. For AMPERA, the expected result is a reduction in time, cost and technical risk in moving from concept to deployment. Founder and chief executive Brian Matthews said the collaboration will boost every aspect of the company’s technology development. From Imperial, co-investigator Jiansheng Xiang described NeuralTwin as a further step in combining scientific machine learning, digital twins and high-fidelity computational methods.
AMPERA is based in Palm Beach Gardens, Florida, and was founded in 2025 by Matthews, a British scientist with a background in nuclear engineering. It is developing thorium-based subcritical microreactors, with a hybrid fusion-fission architecture and in-house production of TRISO fuel. According to BriefGlance, the company has raised a total of $20 million, including $6 million from Super Micro Computer and $10 million from Scorpio Tankers. The microreactors are designed to power individual data centres and industrial facilities.
The context explains the interest in this kind of tool. For advanced reactors, engineering cycles and the licensing process demand a great deal of time and computing resources. A digital twin able to demonstrate a system’s behaviour and safety in real time shortens the work before construction and gives regulators faster data to review. For Imperial College, the project brings established research in scientific machine learning into a concrete industrial case, in the UK as well as in the United States.
If NeuralTwin delivers on its promise, AMPERA’s microreactors could reach their first tests with a simulation dossier already built on a verifiable physical basis. The same approach could then be adopted by other developers of modular reactors aiming for faster licensing.
