OpenAI deploys in-house Jalapeño ASIC alongside AMD Turin

On October 2, 2026, Tom's Hardware published a report stating that OpenAI has deployed its in-house Jalapeño ASIC chip internally, paired with AMD EPYC Turin hosts equipped with 1.5TB of memory. IT之家 (ITHome) carried the same account on October 3.

In an interview with Tom's Hardware, Richard Ho, OpenAI's vice president and head of hardware, said the choice of Turin was a "pragmatic" one, and described Nvidia's new Vera CPU as "a little behind" on maturity. Turin, he said, offers strong performance and already meets all of the Jalapeño project's requirements, while partners have extensive experience with the platform, which helps accelerate deployment.

According to the same report, the OpenAI team settled on two principles at the design stage of Jalapeño: "de-risking" and "landing quickly." Ho said the company wants to stay aggressive on performance metrics and cost control, but that platform choices must fit an existing engineering decision framework so the project remains manageable. Jalapeño is currently limited to OpenAI's internal use, with no plans to sell it externally.