OpenAI said its in-house Jalapeno AI chip outperformed Nvidia’s GB300 on a public benchmarking system in power efficiency and response speed.

The ChatGPT maker plans to use Jalapeno for AI inference later this year.

Richard Ho, OpenAI’s chip chief, said the processor was developed with Broadcom, a US chip designer that builds custom semiconductors for clients, to cut data-center power costs.

He said Jalapeno was not tested against Nvidia’s next-generation Vera Rubin chips and is not designed to train AI models.

OpenAI said it used a smaller open-source model, along with models from DeepSeek and Moonshot AI, in the public tests.

The company is developing a second version of the chip and still expects to rely on Nvidia and other providers as its computing demand remains high.

OpenAI unveiled Jalapeno on June 24 as its first AI chip.

🔗 Source: Bloomberg

Recent OpenAI developments