OpenAI Announces New Jalapeño AI Chip is Faster Than Nvidia
According to InferenceX benchmark tests, the Jalapeño chip delivered higher energy efficiency and speed compared to Nvidia's GB200 and GB300 chips across GPT-OSS 120B, DeepSeek R1, and Kimi K2.5 1T models.
OpenAI has announced that its new AI chip, Jalapeño, developed in partnership with Broadcom, offers higher efficiency and lower latency compared to Nvidia's superchips in performance tests.
High Performance and Low Latency
Richard Ho, OpenAI's Vice President of Hardware, stated that the Jalapeño chip combines low latency and high data throughput, two things that are typically compromised in AI systems.
In tests conducted on the InferenceX platform, Jalapeño was compared against Nvidia's GB200 and GB300 superchips. The chip performed 1.5 to 1.9 times more AI work per watt on the GPT-OSS 120B, DeepSeek R1, and Kimi K2.5 1T models.
In tests conducted on the same models, Jalapeño was observed to reduce end-to-end latency by 1.7 to 3.6 times. This will provide users with faster responses and more responsive AI agents.
Manufacturing Partnership and Future Plans
First introduced in June, Jalapeño is an application-specific integrated circuit (ASIC) developed in partnership with Broadcom and specifically designed for AI inference processes.
OpenAI plans to deploy the new chip in small volumes by the end of this year and ramp up production volume in 2027. However, the company did not share a clear number regarding how many chips will be deployed next year.
Richard Ho stated that despite this performance increase, OpenAI does not plan to replace its entire chip infrastructure with Jalapeño, and that they will continue working with strong partners like Nvidia. The company is also continuing to develop second and third generations of the chip.