OpenAI and Broadcom have unveiled Jalapeño, a custom artificial intelligence (AI) chip designed to improve the speed and efficiency of large language model (LLM) inference.
The companies said Jalapeño was built specifically for running AI models rather than training them. It is the first chip in a planned multi-generation computing platform aimed at supporting OpenAI’s growing infrastructure needs as demand for AI services continues to rise.
In a recent blog OpenAI said, engineering samples of the chip are already running machine-learning workloads in testing laboratories, including workloads related to GPT-5.3-Codex-Spark. Early testing suggests the chip could deliver significantly better performance per watt than current leading AI accelerators, although final performance measurements have not yet been released.
OpenAI said the chip was designed from the ground up using its understanding of how large language models operate. The company worked with Broadcom and hardware manufacturer Celestica on chip production, networking systems and large-scale deployment infrastructure.
Greg Brockman, OpenAI’s president and co-founder, said the chip forms part of a long-term effort to make computing resources more abundant and AI systems faster, more reliable and more affordable.
OpenAI also said the project moved from initial design to manufacturing tape-out in nine months, a development cycle the company described as one of the fastest for a high-performance custom semiconductor. The company said its own AI models were used to assist parts of the chip design and optimisation process.
Broadcom chief executive Hock Tan said the partnership is intended to support the infrastructure required for the next phase of AI growth. The companies plan to deploy the platform in large-scale data centres beginning in 2026, working with partners including Microsoft.