Summary
- OpenAI and Broadcom have introduced Jalapeño, OpenAI's inaugural custom inference chip.
- This chip is tailored specifically for large language model applications.
- The announcement follows earlier speculation about OpenAI's efforts to create custom silicon to lessen its dependence on Nvidia.
On Wednesday, OpenAI revealed Jalapeño, marking its debut in custom artificial intelligence chips, a project that could significantly influence the performance of ChatGPT and upcoming AI offerings.
Created in collaboration with Broadcom, this chip is specifically engineered for the inference tasks associated with large language models—essentially the generation of responses to user queries—and represents OpenAI's most ambitious step towards gaining control over the hardware that supports its AI technologies.
Greg Brockman, President of OpenAI, stated, "Jalapeño is part of our long-term, full-stack infrastructure strategy to make compute more abundant, resulting in AI which is faster, more reliable, more affordable for people and businesses, and can be used to solve more important problems. By designing more of the stack ourselves, we can serve more intelligence with greater efficiency and keep pushing advanced AI toward broader access."
In contrast to many AI chips that are designed for diverse computing tasks, Jalapeño has been constructed with a focus on powering chatbots and other AI systems that utilize large language models. OpenAI has indicated that early prototypes of the chip are currently undergoing tests in their facilities, including with GPT-5.3-Codex-Spark. The company asserts that Jalapeño can provide enhanced computing capabilities while consuming less energy than the leading AI chips available today, although benchmark data has yet to be published.
This announcement validates earlier reports about OpenAI’s initiatives to establish a custom chip program. A report from Reuters earlier this year indicated that OpenAI was aiming to deploy its first in-house AI chip to decrease reliance on Nvidia hardware. The company's chip development ambitions were further emphasized in April, with reports suggesting it was working on a chip intended for smartphones.
Jalapeño is the first in what OpenAI describes as a multi-generation computing platform, expected to start being deployed in data centers later this year, with future iterations designed to support gigawatt-scale AI infrastructures in partnership with Microsoft and others.
Hock Tan, President and CEO of Broadcom, remarked, "Our collaboration with OpenAI represents a fundamental commitment to scaling the physical infrastructure required for the next decade of AI. By co-developing our industry-leading silicon directly with OpenAI, we are enabling the deployment of gigawatt-scale data centers with Microsoft and other partners beginning in 2026."