OpenAI has officially deployed its proprietary Jalapeño ASIC chips within its internal infrastructure. These custom processors are integrated into server systems powered by AMD EPYC Turin CPUs, which are equipped with 1.5TB of memory.
Richard Ho, Vice President and Head of Hardware at OpenAI, stated that the decision to utilize AMD Turin processors was driven by pragmatic engineering requirements. He noted that Nvidia’s Vera CPU currently lags in terms of market maturity.
The development of the Jalapeño project was guided by two core principles: risk mitigation and rapid implementation. OpenAI prioritized aggressive performance targets and cost control while ensuring the hardware platform aligned with existing engineering frameworks.
At present, the Jalapeño ASIC is restricted to internal use at OpenAI, with no immediate plans for commercial distribution. However, the company has not ruled out the possibility of expanding the chip's availability to broader use cases in the future.
