OpenAI and Broadcom unveiled Jalapeño, a custom AI inference chip designed for gigawatt-scale data centers from 2026

OpenAI and Broadcom have jointly unveiled Jalapeño, a custom AI inference chip purpose-built for gigawatt-scale data center deployments. The chip represents a move toward proprietary silicon designed specifically for the inference workloads powering production AI services at scale.

The partnership signals a strategic pivot toward custom silicon solutions for managing the compute and power demands of large AI infrastructure. Rather than relying solely on general-purpose GPUs, OpenAI and Broadcom's collaboration targets the architectural requirements of massive data centers, where inference efficiency and thermal management at gigawatt scale become critical economic and operational factors.

Jalapeño is expected to begin deployment in data centers from 2026 onward, marking a concrete shift in the AI hardware market away from off-the-shelf accelerators toward specialized inference silicon. The move reflects broader industry trends of hyperscalers investing in custom silicon to reduce per-inference costs and optimize energy efficiency as AI workloads become increasingly demanding at scale.