OpenAI reveals its first AI processor: Jalapeño

OpenAI has introduced its inaugural proprietary server processor, developed in close partnership with semiconductor manufacturer Broadcom.

Dubbed Jalapeño, the new hardware is designed specifically as an application-specific integrated circuit (ASIC) to handle artificial intelligence inference workloads. This means the chip is dedicated to executing live user queries and running specialized agents rather than handling the heavy data ingestion required during the initial training phase of large language models. By creating custom silicon, OpenAI aims to establish a more tailored and efficient physical infrastructure to support its widespread conversational platforms and automated tools.

This strategic move is expected to alleviate the company's reliance on external hardware suppliers, particularly amid ongoing global shortages of high-end graphics processing units. In benchmarking tests, the hardware has demonstrated computational capabilities comparable to major industry-standard processors while exhibiting superior energy efficiency per watt. OpenAI plans to gradually incorporate the new chips into its data centers over the next several years, positioning the launch as the first milestone in a long-term roadmap for its custom server architecture. This step aligns the organization with other technology giants that are increasingly designing proprietary silicon to reduce operation costs and secure their supply chains.