Rather than adapting an existing accelerator design, OpenAI said the processor was created specifically for LLM inference. The architecture was informed by the company’s experience operating ChatGPT, Codex, API services, and future AI products. Broadcom and Celestica contributed to chip implementation, networking, manufacturing, system integration, and production infrastructure.
Engineering samples are already running machine learning workloads in testing environments, including GPT-5.3-Codex-Spark. While final performance measurements are still underway, OpenAI said early results indicate that Jalapeño could provide significantly better performance per watt than current leading systems. The company said a technical report with detailed benchmarks will be released in the coming months.
According to OpenAI, the chip was designed to improve efficiency by reducing data movement and more effectively balancing compute, memory, and networking resources. Broadcom’s technologies, including its Tomahawk networking silicon, are intended to support deployment at large scale.
“The world is moving to a compute-powered economy,” said Greg Brockman, President and Co-Founder of OpenAI. “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.”
Richard Ho, who leads OpenAI’s hardware program, said the processor was developed around the specific requirements of frontier AI systems.
“Jalapeño was designed from the ground up for LLM inference using detailed insights from our close collaboration with OpenAI researchers,” Ho said. “We optimized the architecture around the kernels, memory movement, networking, and serving patterns that matter most for frontier AI models. Based on early testing, Jalapeño will efficiently execute our most important workloads close to the hardware’s theoretical limits.”
Broadcom positioned the project as the start of a longer-term roadmap. The company said it is working with OpenAI on future generations of AI infrastructure that combine custom accelerators, networking technology, and system-level integration.
“Our collaboration with OpenAI represents a fundamental commitment to scaling the physical infrastructure required for the next decade of AI,” said Hock Tan, President and CEO, Broadcom. “This is just the beginning of a multi-generation roadmap. 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.”
OpenAI said the chip reached manufacturing tape-out nine months after the initial design phase began. The company attributed that timeline to close collaboration between OpenAI engineers and Broadcom’s semiconductor teams, as well as the use of OpenAI models during portions of the design and optimization process.
Jalapeño is expected to serve as the foundation of a broader compute platform that combines OpenAI-designed accelerators with Broadcom’s connectivity and networking technologies and Celestica’s expertise in boards, racks, and systems. Initial deployment is planned by the end of 2026, with additional generations expected to follow.
OpenAI said the primary goal is to improve the speed, cost, and reliability of AI inference. The company stated that gains in infrastructure efficiency could translate into faster responses, lower costs, and more dependable access across products including ChatGPT, Codex, and its API offerings.
“Democratizing AI means making advanced models available, dependable, and affordable enough for more people to use every day,” the company said. “Jalapeño helps OpenAI turn more of its infrastructure into useful intelligence for students, developers, small businesses, researchers, enterprises, and anyone trying to learn, create, or solve hard problems.”
This analysis is based on reporting from OpenAI.
Image courtesy of OpenAI.
This article was generated with AI assistance and reviewed for accuracy and quality.
