TL;DR

Fuse Energy, a UK supplier serving more than 450,000 households, is taking its AI computing in-house on NVIDIA hardware, according to a company announcement carried by Energy Sustainability Solutions on 20 October. Its customer-facing support AI will move from a closed-source setup to open-weight models on systems Fuse controls. Fuse calls this the first live use in a wider arrangement making NVIDIA its preferred compute provider.

What is changing

Fuse runs its customer service through its app rather than a call centre, and an AI agent handles support queries there. The company says the current closed-source system resolves four in five queries without passing them to a person, and reports 99% customer satisfaction. Both figures are Fuse’s own. It expects the switch to open-weight models and its own compute to keep performance at that level, though the announcement gives no figures for the new setup yet.

Fuse describes the move as an extension of how it runs its energy business. It owns its whole chain, from solar and wind generation through trading to retail supply, whereas traditional rivals usually hand generation, trading and retail to others. The company credits that structure with efficiency gains of about 17% over older suppliers.

Chief executive and co-founder Alan Chang said working with NVIDIA lets Fuse “run AI workloads on our own terms, on infrastructure we control”. Santhosh Gopinath, who runs NVIDIA’s energy business across Europe, the Middle East and Africa, said Fuse plans to use AI beyond support, in decisions on demand, renewable output and volatile markets.

Context

Fuse reported profitability at group level in June, three years after its public launch.

Control over where AI runs also features in Perplexity’s deal with London start-up Valarian which is pitched at public bodies that want AI inside infrastructure they govern. In our view, Fuse is taking a comparable step for commercial reasons, owning the hardware and choosing models it can run itself.

Looking forward

In our view, this is worth watching as a test case rather than a result. Moving a working customer-service agent from a closed model to open-weight ones on in-house compute is a cost and control bet, and the claim that performance will hold is so far untested in public. UK firms weighing the same move should look for Fuse’s post-migration numbers before treating it as proof that open-weight models can replace closed ones for frontline support.