Phoenix 24
AI power is becoming a hardware sovereignty race.
London, May 2026. Anthropic is exploring a potential chip supply partnership with Fractile, a British semiconductor startup, as the maker of Claude seeks to reduce its dependence on Nvidia and contain the rising cost of artificial intelligence infrastructure. The talks remain at an early stage and no binding agreement has been signed, but the strategic signal is clear: leading AI firms are no longer treating compute as a neutral input. They are trying to control the hardware layer that determines speed, margins and long-term scalability.
Fractile’s appeal lies in its unconventional chip architecture, which seeks to combine memory and computation more directly on the processor. The company claims this approach could make large language models run significantly faster while reducing operational costs, although its technology remains under development and is not expected to be ready for large-scale data center deployment before 2027. That gap between promise and deployment explains both the opportunity and the risk for Anthropic, which needs cheaper compute but cannot afford infrastructure fragility in a market moving at extreme speed.
The move also reflects a broader shift among major technology companies seeking alternatives to general-purpose chips and tighter control over AI infrastructure. Anthropic currently relies heavily on Nvidia’s high-end processors and on custom chips provided through cloud partners such as Amazon and Google, both of which have invested heavily in the company. A Fractile agreement would not immediately replace that ecosystem, but it could become part of a diversified supply strategy designed to reduce exposure to scarcity, pricing pressure and supplier concentration.
For Britain, the negotiations offer a symbolic opening in the global semiconductor race. If Fractile becomes a meaningful supplier to one of the world’s most visible AI companies, the United Kingdom would gain relevance in a sector dominated by American platforms, Asian manufacturing capacity and Nvidia’s hardware leadership. The deeper lesson is structural: the AI race is no longer only about models, talent or data. It is about who controls the chips that decide how intelligence is produced, priced and deployed.
Información que anticipa futuros. / Information that anticipates futures.