OpenAI Says Jalapeño AI Chip Beats Nvidia GB300 on Power Efficiency and Speed
OpenAI has published benchmark results for Jalapeño, its AI inference chip developed with Broadcom, showing it outperforms Nvidia's GB200 and GB300 superchips. In tests using the InferenceX platform across three models, Jalapeño delivered 1.5 to 1.9 times more AI work per watt and cut end-to-end latency by 1.7 to 3.6 times. OpenAI plans small-batch deployment by the end of 2026 and production expansion in 2027, while stressing Nvidia remains a key partner.
OpenAI published a blog post on August 25 disclosing performance test results for its self-developed artificial intelligence chip, Jalapeño. The chip, an application-specific integrated circuit designed for AI inference tasks, was developed in collaboration with Broadcom and was first publicly unveiled in June 2026. In comparative tests using the InferenceX benchmark platform, Jalapeño was measured against Nvidia's GB200 and GB300 superchips. Results showed that across three models — GPT-OSS 120B, DeepSeek R1 and Kimi K2.5 1T — Jalapeño completed 1.5 to 1.9 times more AI work per watt than the comparison systems, while reducing end-to-end latency by 1.7 to 3.6 times. Richard Ho, OpenAI's vice president of hardware, said the chip achieves both lower latency and higher throughput, whereas AI systems have typically required a trade-off between the two.
OpenAI plans to deploy Jalapeño in small batches by the end of 2026 and gradually expand production capacity in 2027, though it did not disclose specific deployment volumes. Ho said the chip delivers high performance at 700 watts of power, helping to reduce data center electricity costs. OpenAI does not position Jalapeño as a full replacement for Nvidia chips; given its enormous computing demands, the company will continue to work with multiple suppliers, with Nvidia remaining an important partner. The tests did not include Nvidia's latest-generation chip, Vera Rubin, which has only just begun shipping.
OpenAI is advancing follow-up development of the Jalapeño chip. The company said it accelerated the chip's development process with the help of its own AI models. The collaboration with Broadcom was announced in 2025, and both parties highlighted the chip's development speed in June 2026. Ho said the company has reached a power and cost level that can meaningfully reduce infrastructure costs.