OpenAI unveils Jalapeño, its first custom AI chip designed to slash inference costs by 50%

OpenAI has unveiled Jalapeño, its first custom AI chip designed specifically for large language model inference, marking a strategic shift into proprietary hardware developed in partnership with Broadcom to significantly lower deployment costs and improve energy efficiency. The chip, which was engineered from scratch in just nine months using OpenAI's own AI models, targets a 50% cost reduction compared to standard GPUs and offers substantially better performance per watt than current state-of-the-art hardware. This move represents a major expansion of OpenAI's full-stack infrastructure, aiming to support gigawatt-scale data center operations with partners like Microsoft by late 2026 to power global AI services including ChatGPT and Codex more efficiently.