For years, a single company has quietly sat at the centre of the artificial intelligence revolution — not as a model-maker or a researcher, but as the supplier of the hardware on which almost everything runs. Nvidia's grip on the AI chip market has been so complete that its name has become almost synonymous with AI infrastructure itself. That dominance, however, is now being tested from multiple directions at once, and the latest challenge has a distinctly spicy name. OpenAI has confirmed plans for a custom inference chip called Jalapeño, developed in partnership with semiconductor firm Broadcom.

The move places OpenAI alongside Google, Apple, and SpaceX in a fast-growing cohort of technology companies that have decided to design their own silicon rather than remain permanently dependent on a single external supplier. The strategic logic is not about staging a dramatic exit from Nvidia; it is about hedging risk, gaining engineering flexibility, and tailoring hardware to the precise demands of specific workloads. Inference — the process by which a trained AI model actually generates responses — places different pressures on a chip than training does, and custom designs can be optimised accordingly. The template that most observers cite is Apple's transition away from Intel processors, completed in 2020 with the introduction of its M1 chip.

That shift delivered performance improvements and power-efficiency gains that third-party silicon had never been able to provide, and it handed Apple a degree of vertical integration that strengthened its competitive position considerably.