OpenAI and Broadcom Unveil Jalapeño, an LLM-Optimized Inference Chip

OpenAI and Broadcom have unveiled Jalapeño, OpenAI's first "Intelligence Processor" and a custom AI accelerator designed from the ground up for large language model (LLM) inference. The chip represents the first generation of a multi-generation compute platform developed with Broadcom and Celestica, aimed at making advanced AI faster, more reliable, and more accessible.

Unlike general-purpose accelerators adapted from earlier AI workloads, Jalapeño is a blank-slate architecture optimized around the kernels, memory movement, networking, and serving patterns critical to frontier AI models. Engineering samples are currently running ML workloads in the lab at production target frequency and power, including GPT-5.3-Codex-Spark. While final benchmarks are pending, early testing indicates the chip delivers performance per watt substantially better than current state-of-the-art hardware. A detailed technical report is expected in the coming months.

The project moved from initial design to manufacturing tape-out in nine months, which OpenAI describes as the fastest ASIC development cycle achieved in high-performance advanced semiconductors. The rapid timeline was enabled by deep software-hardware co-development and the use of OpenAI's own models to accelerate parts of the design process.

Broadcom contributed silicon implementation and networking technologies, including Tomahawk networking silicon, while Celestica provided board, rack, and system integration expertise. The platform is planned for deployment at gigawatt scale with data center partners including Microsoft, beginning by the end of 2026.

OpenAI executives framed the chip as part of a long-term, full-stack infrastructure strategy. "The world is moving to a compute-powered economy," said President and Co-Founder Greg Brockman. "By designing more of the stack ourselves, we can serve more intelligence with greater efficiency." Broadcom President and CEO Hock Tan described the collaboration as a "fundamental commitment to scaling the physical infrastructure required for the next decade of AI."

Source: OpenAI Blog