OpenAI’s Jalapeño chip promises quicker AI responses than rival technologies.

OpenAI has unveiled its latest AI chip, Jalapeño, which promises to enhance task efficiency and speed up response times compared to existing AI systems, as detailed in a recent blog post. During a press briefing, Richard Ho, OpenAI’s hardware vice president, highlighted that Jalapeño strikes a balance between lower latency and higher throughput, addressing a common challenge where AI systems often have to compromise between the two.

First revealed in June, Jalapeño is an Application-Specific Integrated Circuit (ASIC) developed in collaboration with Broadcom, specifically tailored for AI inference—the process of executing a trained AI model to perform tasks or deploy agents.

To evaluate Jalapeño’s capabilities, OpenAI utilized InferenceX, a benchmarking platform that assesses how effectively AI systems manage inference tasks. The tests measured Jalapeño against the top-performing Nvidia GB200 and GB300 superchips. According to OpenAI, Jalapeño achieved 1.5 to 1.9 times more AI work per watt across models like GPT-OSS 120B, DeepSeek R1, and Kimi K2.5 1T, while also delivering 1.7 to 3.6 times lower end-to-end latency across these models. This means users can expect quicker responses, more agile agents, and improved reliability as demand increases, as noted by Ho.

OpenAI intends to roll out Jalapeño in limited quantities by the end of this year, with plans to gradually increase production into 2027. However, the company has not disclosed the exact number of chips it aims to deploy next year.

Despite these advancements, Ho mentioned that OpenAI does not plan to replace its entire chip lineup with Jalapeño, emphasizing that their overall computing strategy involves strong partnerships, including with Nvidia. The development of the second and third generations of this new chip will continue as well.