OpenAI presents Jalapeño, its first processor for AI inference

OpenAI has presented his first processor for artificial intelligence (AI) which he has named Jalapeño and which has been designed for current and future large language model (LLM) workloads.

Jalapeño is a processor designed for work with all LLMswhich is optimized for inference, that is, the process in which the AI ​​collects the user’s indication and applies the knowledge acquired in its training to provide a result.

Its development, which has been completed from scratch in nine months together with Broadcom and Celestica, It is part of OpenAI’s strategy to work not only on AI but also on the infrastructure that supports and drives it, allowing “its models to be faster, more reliable and more affordable for users,” as indicated in a statement shared on its blog.

“We optimize the architecture around the cores, memory movement, network and service patterns that are most important for cutting-edge AI models,” detailed OpenAI hardware program director Richard Ho.

The goal of Jalapeño is optimize computational efficiency, to deliver better performance and lower latency, and drive long-term, scaled LLM usage. This chip is currently running AI workloads in the OpenAI lab at production target frequency and power, including GPT-5.3-Codex-Spark.

The technology firm intends to create a “multigenerational computing platform” that it plans to implement at the end of this year and expand it over time.