OpenAI Builds First Custom AI Chip Called Jalapeno with Broadcom

OpenAI has recently developed its first custom artificial intelligence chip, which has been officially named 'Jalapeno'. This significant technological hardware development was carried out in a direct partnership with major chip manufacturer Broadcom, marking a new phase in the organization's infrastructure strategy.

In addition to Broadcom, another key industry player, Celestica, worked extensively on the project. Celestica was specifically responsible for developing the essential hardware required for the chip, including the specialized board and the rack systems necessary to support the new technology.

The newly developed Jalapeno chip has already undergone rigorous testing processes. Performance evaluations were conducted using several prominent models, specifically the gpt-oss-120b, DeepSeek R1, and Kimi K2.5 models, to measure its operational capabilities across different frameworks.

Performance statistics released regarding the Jalapeno chip highlight notable efficiency gains. The custom silicon can perform AI tasks at a rate of 1.5 to 1.9 times more per watt of electricity consumed, while response times have successfully decreased by a factor of 1.7 to 3.6 times.

Furthermore, testing demonstrated that for applications requiring instantaneous communication, the chip's performance reached up to 4.1 times higher. Despite the complexity of the hardware, the entire development phase for the Jalapeno chip took a remarkably short period of just nine months.

Looking ahead, the organization has laid out a phased deployment schedule for the new hardware. There is a confirmed plan to begin using the Jalapeno chip in limited quantities by the end of 2026, followed by a broader expansion of its use throughout 2027.