OpenAI's 'Jalapeño' chip: better perf per watt than Nvidia

openai and broadcom unveiled jalapeño, openai's first custom intelligence processor, designed specifically for llm inference to achieve performance per watt that surpasses current state-of-the-art accelerators. the chip was developed from design to production in nine months—accelerated by openai's own ai models—and is currently running engineering samples including gpt‑5.3‑codex‑spark at target frequency and power. by optimizing for kernels, memory movement, and serving patterns, jalapeño aims to reduce data movement and increase hardware utilization, addressing the specific needs of frontier models rather than adapting general-purpose accelerators. broadcom will handle industrialization and networking integration, with deployment planned at gigawatt scale with data center partners like microsoft beginning in 2026, signaling openai's strategic shift toward a full-stack infrastructure approach.