OpenAI Unveils Jalapeno AI Chip to Rival Nvidia

OpenAI has developed Jalapeno, a custom in-house AI processor designed specifically for the inference phase of AI models. Created in partnership with Broadcom Inc., the chip aims to reduce operational costs and power consumption in data centers. In tests against Nvidia Corp.'s GB300, Jalapeno demonstrated superior response speeds and higher AI workloads per unit of power, operating at a low voltage of 700 watts.

Further testing against models such as GPT-OSS 120B and DeepSeek R1 showed the chip delivers 1.5 to 1.9 times more work per watt and 1.7 to 3.6 times lower end-to-end latency than existing systems. OpenAI reported that AI played a direct role in the chip's development, shortening the design-to-tapeout timeline to nine months. The company plans to integrate the hardware into its infrastructure by the end of 2026 and is already designing second and third-generation versions.

Despite this move toward a full-stack strategy, OpenAI continues to rely on Nvidia as a key supplier. This includes a data center project in Ohio supported by up to $105 billion in credit from Nvidia. In response to the shifting inference market, Nvidia has announced its own new hardware, including the Groq 3 LPX chip for data centers and the Jetson Orin Nano 2 for edge robotics, the latter of which is expected in early 2027.

The hardware announcement coincided with leadership restructuring at OpenAI, including the departure of its head of data centers.