OpenAI and Broadcom unveiled Jalapeño, OpenAI's first custom AI inference chip, targeting ~50% cost reduction for LLM inference. Designed and tested using OpenAI's own models over a nine-month cycle; deployment targeted for end-2026.

Broadcom and OpenAI have announced a collaboration on custom AI accelerators, with the two companies jointly developing Jalapeño, a new chip designed specifically to run inference for large language models. The partnership centers on 10 gigawatts of custom AI accelerators that are scheduled to begin arriving in the second half of 2026.

The timing of initial shipments aligns with the strong ramp Broadcom projected for the back half of the year in its latest earnings call. More significantly, Broadcom plans to deploy racks of AI accelerator and network systems starting in the second half of 2026 and continuing through the end of 2029, providing multi-year visibility into future demand. This roadmap echoes CEO Hock Tan's March statement about having "a line of sight to achieve AI revenue from chips, just chips, in excess of $100 billion in 2027."

OpenAI is one of five custom AI silicon customers at Broadcom, alongside Google, Meta, Anthropic, and ByteDance. Beyond chips, Broadcom is also well positioned to capitalize on growing demand for networking infrastructure supporting these systems.

The announcement carries implications for Microsoft, which has an existing partnership with OpenAI. By co-developing industry-leading silicon with OpenAI, the two companies are enabling the deployment of gigawatt-scale data centers with Microsoft and other partners beginning in 2026. As this capacity comes online, Microsoft should begin to monetize it, potentially helping to ease margin pressures on the company. The deployment represents a significant opportunity for Microsoft to capitalize on the data center and cloud infrastructure demand driven by AI adoption and usage.