OpenAI and Broadcom Unveil Jalapeño Chip

OpenAI has introduced Jalapeño, its first custom-built inference processor designed specifically for large language models. Developed in just nine months, the chip aims to optimize performance per watt for LLM workloads, including GPT-5.3-Codex-Spark.

Engineering samples are currently undergoing lab testing, with initial data center deployments expected by late 2026. This purpose-built ASIC represents a move toward a vertically integrated stack, focusing on memory movement and networking efficiency for future agentic AI.