Cambridge-based ARM Holdings has announced a revolutionary new chip architecture that promises to deliver AI training performance comparable to high-end GPUs at a fraction of the cost and energy consumption.

The new design, codenamed 'Prometheus,' uses a novel approach to parallel processing that ARM claims can achieve 90 per cent of the performance of Nvidia's H100 GPU while consuming just 40 per cent of the power and costing roughly a third as much to manufacture.

"This is not an incremental improvement — this is a paradigm shift," said ARM's chief technology officer. "We are democratising access to AI compute in a way that the current GPU-centric model simply cannot."

The announcement sent shockwaves through the semiconductor industry, with Nvidia shares falling 6 per cent in pre-market trading. Analysts cautioned, however, that ARM's claims remain unverified in independent benchmarks and that the architecture is still 18 months from commercial availability.

If ARM's performance claims hold, the implications for Britain's AI ambitions could be profound, potentially positioning the country as a key node in the global AI hardware supply chain for the first time.