OpenAI and Broadcom on Wednesday introduced Jalapeño, a custom-designed AI accelerator described as the company’s first “Intelligence Processor,” marking a step into in-house chip development aimed at powering large language model (LLM) inference at scale.
The companies said early testing indicates the first-generation accelerator delivers “substantially better” performance per watt compared with current state-of-the-art AI accelerators, though final performance metrics have not yet been released. A detailed technical report is expected in the coming months.
Jalapeño is designed specifically for LLM inference workloads, rather than as a general-purpose accelerator adapted from earlier AI applications. OpenAI said the chip was developed to optimize compute, memory, and networking balance in order to reduce data movement and improve utilization efficiency for modern AI workloads.
Engineering samples of the chip are already running machine learning workloads in lab settings at production target frequency and power levels, including workloads based on GPT-5.3-Codex-Spark, according to the companies.
The project represents a collaboration between OpenAI, Broadcom, and Celestica (TSX:CLS), with Broadcom contributing silicon implementation and networking technologies, including its Tomahawk networking chips, and Celestica supporting system integration and production scaling. The companies said the platform is intended to support deployment at gigawatt scale through data center partners over multiple future generations.
“This is part of our long-term full-stack infrastructure strategy to make compute more abundant,” OpenAI President Greg Brockman said in a statement.
“By designing more of the stack ourselves, we can serve more intelligence with greater efficiency.”
OpenAI hardware lead Richard Ho said Jalapeño was built using insights from internal model development and inference systems, with design choices informed by “kernels, memory movement, networking, and serving patterns” used in frontier AI workloads.
Broadcom CEO Hock Tan said the partnership reflects a “fundamental commitment” to scaling infrastructure for the next decade of AI development.
“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,” Tan said.
Broadcom shares were 1% higher following the announcement.