OpenAI Details Custom Jalapeño Chip Designed to Speed Up AI Inference

Industry Pulse News Desk · 2026-08-25

The custom ASIC processor developed with Broadcom delivers lower latency and higher throughput for artificial intelligence models.

OpenAI announced Tuesday that its custom-built artificial intelligence chip, code-named Jalapeño, delivers faster processing speeds and improved task completion efficiency compared to competing processors. The chip is specifically designed to optimize AI inference, the computational process of running trained models to execute tasks and deploy digital agents.

First introduced in June, Jalapeño is an Application-Specific Integrated Circuit developed in partnership with semiconductor manufacturer Broadcom. The hardware aims to address infrastructure bottlenecks as demand for generative AI tools and enterprise applications continues to scale globally.

During a press briefing, OpenAI Hardware Vice President Richard Ho said the processor achieves lower latency alongside higher throughput. Ho noted that traditional AI chip architectures typically force systems to trade off between response speed and overall computational volume, a limitation Jalapeño is engineered to overcome.

Technical performance benchmarks released by the company focus on reducing time between tokens, a key metric measuring how rapidly a model generates successive pieces of text or data during a response. Lower token delivery times allow interactive AI applications to operate with reduced lag for end users.

The deployment of proprietary silicon reflects a broader industry shift among major AI developers seeking to reduce reliance on third-party hardware suppliers while optimizing operational costs and system performance for specialized workloads.