Is Jalapeño the end of Nvidia's dominance?
openai and broadcom revealed 'jalapeño', a custom inference chip designed in 9 months, to slash costs and secure supply. this threatens nvidia's monopoly and signals a shift toward vertically integrated ai infrastructure.
openai and broadcom have unveiled 'jalapeño', openai's first custom silicon for large language model inference, designed to optimize data movement and maximize utilization. engineering samples are already running models like gpt-5.3-codex-spark in labs, and the company claims 'substantially better' performance per watt than state-of-the-art hardware, though these figures are unverified. the nine-month development cycle is notably fast for an asic of this class, and openai credits its internal models for accelerating the design process. this move signals openai's shift to full-stack vertical integration to ensure supply and control costs, with microsoft backing the venture by agreeing to purchase 40% of the volume for a late 2026 rollout.