Fuse is pre-training an energy foundation model on its own demand data and says it cut forecast error to 1%, about 8x better than the industry

Fuse Energy co-founder and CEO Alan Chang says the UK energy company started pre-training an energy foundation model on its in-house demand data two weeks ago, aiming for a 'superhuman energy trader'. On Fuse's own figures, demand forecast error is 1.00% for the transformer, against 3.46% for its existing classical ML stack and 7.84% for the industry standard: 3.5x better than its own best model and about 8x better than the industry. Chang calls it early, but says every point of forecast error is pure cost when buying power in a volatile market. Last week Fuse said it is moving AI workloads in-house onto NVIDIA hardware and open-weight models. It is hiring AI researchers.