OpenAI and Broadcom Co-Developed Chip Jalapeño to Deploy by End of 2026

en.Wedoany.com Reported - The inference chip Jalapeño, developed in collaboration between OpenAI and Broadcom, is slated to begin small-scale deployment by the end of 2026. Ho estimates that initial deployment will be "at a very small scale," with larger-scale deployment following in 2027. Its performance benchmark is Nvidia's Blackwell system, but by the time Jalapeño is fully rolled out, competing products may have advanced significantly.

Jalapeño was first publicly announced in October of last year, developed through close collaboration between OpenAI and Broadcom, with OpenAI's own models also participating in the development process. The company plans to build Jalapeño into a multi-generational platform, enabling AI products, models, chips, and memory to be developed in tandem.

This full-stack approach allows OpenAI to optimize for specific stages in the inference process that create friction. Jalapeño's design focuses on minimizing latency during the prefill and communication stages, which OpenAI notes are often the bottlenecks in inference.

"We designed Jalapeño to minimize data movement and communication latency," the company stated in a blog post presenting the results. "This means that model state—including the KV cache used to generate responses—can be explicitly placed and kept localized, while the system activates the right combination of compute, memory, and network for each inference stage."