OpenAI and Broadcom unveiled Jalapeño, a custom ASIC inference processor developed in a nine-month sprint to handle large-scale LLM inference workloads.

OpenAI and Broadcom have unveiled Jalapeño, a custom ASIC inference processor developed through a nine-month engineering sprint. The partnership reflects OpenAI's move into proprietary silicon to support its infrastructure scaling.

Jalapeño is designed to handle large-scale LLM inference workloads, addressing a core bottleneck in the AI infrastructure buildout. As a purpose-built ASIC, it targets the inference phase where demand continues accelerating across the industry.

The development signals the strategic shift among AI leaders toward custom silicon to manage growth and control critical infrastructure components. OpenAI's in-house chip work follows similar initiatives by other hyperscalers seeking to optimize performance and efficiency for inference at scale.