OpenAI details its Jalapeño inference ASIC

OpenAI unveiled its Jalapeño inference ASIC at Hot Chips in August 2026, designed in a remarkably short window with heavy AI assistance. VP of Hardware Richard Ho said the chip targets efficiency, low latency and throughput, and runs any transformer-based LLM, not just OpenAI models.

• Jalapeño is OpenAI's inference ASIC, revealed at Hot Chips in August 2026
• Main goal is efficiency: lower latency and inference cost
• Chip runs open-source models, adapted in roughly two months
• Built via full-stack co-design inside OpenAI