OpenAI has launched Jalapeño, its first custom AI chip designed for large language models, in collaboration with Broadcom.
Summary
- OpenAI teams up with Broadcom to launch Jalapeño, its first custom AI inference chip.
- This chip is designed specifically for large language model applications.
- The move aims to reduce OpenAI's reliance on Nvidia's technology.
OpenAI has officially announced the launch of Jalapeño, its first custom-designed artificial intelligence chip, which is expected to enhance the performance of its AI tools, including ChatGPT.
Developed in partnership with Broadcom, Jalapeño is specifically optimized for inference tasks in large language models, focusing on generating responses to user inputs. This initiative marks a significant stride for OpenAI in controlling the hardware that powers its AI innovations.
Greg Brockman, President of OpenAI, emphasized that "Jalapeño is part of our long-term strategy to create a more abundant computing environment, making AI faster, more reliable, and more accessible for everyone. By designing more of our infrastructure, we can deliver intelligence more efficiently and expand access to advanced AI technologies."
Unlike many AI chips designed for a wide range of computing tasks, Jalapeño is tailored for chatbots and systems utilizing large language models. OpenAI is currently testing early prototypes of the chip, including applications with GPT-5.3-Codex-Spark. The company claims that Jalapeño will deliver superior computing performance while being more energy-efficient than current leading AI chips, although performance benchmarks have not yet been released.
This launch confirms previous reports regarding OpenAI's efforts to create a custom chip program aimed at reducing dependence on Nvidia hardware. Earlier this year, reports indicated that OpenAI was focused on developing in-house AI chips, with plans for future chips targeting smartphones.
Jalapeño is the first in a series of chips that OpenAI plans to roll out, with expectations for deployment in data centers later this year. Future versions are designed to support expansive AI infrastructures in collaboration with Microsoft and others.
Hock Tan, President and CEO of Broadcom, noted that "Our partnership with OpenAI signifies a commitment to scaling the infrastructure necessary for the next decade of AI. By co-developing silicon with OpenAI, we aim to enable the deployment of gigawatt-scale data centers with partners like Microsoft starting in 2026."
