OpenAI and Broadcom Launch Jalapeño, a Custom AI Inference Chip for LLMs
OpenAI and Broadcom have jointly unveiled Jalapeño, OpenAI’s first custom Intelligence Processor and a significant move in the company’s push to control its own AI infrastructure from top to bottom. The OpenAI Jalapeño inference chip is designed specifically for large language model workloads, and early testing suggests it will deliver performance per watt that exceeds the current state of the art.
Unlike general purpose accelerators adapted from earlier AI tasks, Jalapeño was built from scratch with LLM inference at the center of every design decision. OpenAI says the chip was optimized around the actual serving patterns, memory movement, and networking demands that power ChatGPT, Codex, and its API products today, while also being built to handle future LLMs across the industry.
The chip was co-developed with Broadcom from initial design to manufacturing tape-out in just nine months, a timeline OpenAI describes as the fastest ASIC development cycle ever achieved in high-performance advanced semiconductors. OpenAI’s own models were used to help accelerate parts of the chip design and optimization process, a detail the company was quick to highlight as proof that AI is already improving the infrastructure used to run future AI.
Engineering samples are already running machine learning workloads in the lab at production target frequency and power, including GPT-5.3-Codex-Spark. A full technical performance report is expected in the coming months.
OpenAI President and Co-Founder Greg Brockman framed the announcement as part of a broader infrastructure ambition. “Jalapeño is part of our long-term full-stack infrastructure strategy to make compute more abundant,” he said, adding that designing more of the stack internally allows the company to serve more intelligence with greater efficiency.
Broadcom CEO Hock Tan confirmed that the first deployments are expected by the end of 2026, with gigawatt-scale data center rollouts planned alongside Microsoft and other partners. The chip is the first in what both companies described as a multi-generation compute platform, developed in partnership with Celestica for board, rack, and system integration, and leveraging Broadcom’s Tomahawk networking silicon for large-scale production.
For users, the practical promise of the OpenAI Jalapeño inference chip is simpler: faster responses, more reliable access during peak demand, and cheaper AI products. OpenAI is betting that owning more of the hardware layer gives it the leverage to keep pushing those improvements forward, generation after generation.
