OpenAI unveiled Jalapeño, its first custom AI inference chip developed with Broadcom, designed specifically for large language models. The chip aims to reduce reliance on Nvidia and improve efficiency for ChatGPT and future AI products. Early versions are being tested with GPT-5.3-Codex-Spark, with data center deployment expected later this year and gigawatt-scale infrastructure planned with Microsoft and partners starting 2026.

Key facts

• Jalapeño is OpenAI's first custom AI chip, developed with Broadcom.
• Designed specifically for large language model inference, targeting chatbots.
• Early testing includes GPT-5.3-Codex-Spark; claims higher efficiency and lower power.
• Part of multi-generation compute platform; data center deployment starts late 2025.
• Aims to reduce reliance on Nvidia and enable gigawatt-scale AI infrastructure by 2026.