OpenAI launches first in-house AI chip in partnership with Broadcom

OpenAI and Broadcom unveiled "Jalapeno," the ChatGPT maker’s first custom artificial intelligence chip, as the companies move to build a new computing platform aimed at making advanced AI faster, more reliable and more widely available.

The chip, described by the companies as OpenAI’s first “Intelligence Processor,” is optimized for inference -- the process of running AI models to generate responses for users across products such as ChatGPT, Codex and OpenAI’s API.

OpenAI said Jalapeno was designed from scratch around the needs of current and future large language models, rather than adapted from older general-purpose AI workloads.

The chip was developed with Broadcom and Celestica, with Broadcom contributing silicon implementation, networking and connectivity technologies, and Celestica supporting board, rack and system integration.

Engineering samples of Jalapeno are already running machine-learning workloads in the lab at production target frequency and power, including GPT-5.3-Codex-Spark, according to the companies.

OpenAI said early testing shows the first-generation accelerator is expected to deliver substantially better performance per watt than current state-of-the-art systems, though final performance measurements are still underway. A detailed technical report is expected in the coming months.

The companies said the chip’s architecture reduces data movement and balances compute, memory and networking resources to improve utilization. Broadcom’s networking technologies, including Tomahawk networking silicon, will help scale the platform for large data centers.

Jalapeno was developed from initial design to manufacturing tape-out in nine months, a timeline the companies described as one of the fastest ASIC development cycles achieved in advanced high-performance semiconductors.

Greg Brockman, OpenAI’s president and co-founder, said Jalapeño is part of the company’s long-term infrastructure strategy to make compute more abundant and AI more affordable for people and businesses.

Richard Ho, who leads OpenAI’s hardware program, said the chip was optimized around the kernels, memory movement, networking and serving patterns that matter most for frontier AI models.

Broadcom President and CEO Hock Tan said the collaboration marks the beginning of a multi-generation roadmap and will support the deployment of gigawatt-scale data centers with Microsoft and other partners beginning in 2026.

The companies said Jalapeno will be the first accelerator in a multi-generation compute platform, with initial deployment planned by the end of 2026 and expansion expected in the following years.

OpenAI has become one of the largest users of advanced AI chips since the launch of ChatGPT, relying heavily on Nvidia graphics processing units to train and run large AI models.

The Broadcom project marks a major step in OpenAI’s effort to diversify its compute supply and reduce the cost of serving AI products as demand for generative AI continues to rise.