OpenAI Deploys In-House Jalapeño ASIC Chips with AMD Turin Servers

OpenAI has internally deployed its self-developed Jalapeño ASIC chips, paired with AMD EPYC Turin hosts featuring 1.5TB of memory, according to an interview with Richard Ho, OpenAI's vice president and head of hardware, published by Tom's Hardware on October 2.

Ho said the choice of AMD Turin was "pragmatic," noting that Nvidia's upcoming Vera CPU is "a bit behind" in maturity. He added that AMD's Turin chip delivers strong performance and meets all the requirements of the Jalapeño project, while the partner's familiarity with the platform helps accelerate deployment.

OpenAI designed Jalapeño with two key principles in mind: de-risking and rapid implementation. Ho emphasized that while the company aims to stay aggressive on performance and cost, platform choices must follow the existing engineering decision framework to keep the project controllable.

For now, Jalapeño ASICs are for internal use only, with no immediate plans for external sales. However, Ho said that if resources allow, OpenAI does not rule out expanding its use to broader scenarios in the future.