OpenAI unveils 'Jalapeño' inference chip, advancing its full-stack AI infrastructure strategy
OpenAI has introduced 'Jalapeño,' its first inference chip, alongside a strategy for integrated AI infrastructure development.
On August 25, 2026, news outlets and OpenAI's official blog published key details from 'The full stack behind abundant intelligence.' OpenAI CFO Sarah Friar outlined the company's technology strategy spanning chips, Frontier Models, developer platforms, and consumer products. Each layer reinforces the others to reduce costs, increase scale, and broaden access.
The key highlight was 'Jalapeño,' the company's first inference chip. InferenceX testing showed that Jalapeño delivers higher throughput per kilowatt and lower token latency than typical commercial systems. It also supports multiple model families, including GPT-OSS 120B, DeepSeek R1, and Kimi K2.5, reflecting the company's push for greater control over its supply chain and compute efficiency.
By developing custom chips such as Jalapeño, leading AI companies can reduce long-term computing costs, potentially making AI services more affordable and accessible to everyday users and developers.