OpenAI today unveiled Jalapeño, its first custom-designed AI chip that it claims is “faster, more reliable, more affordable for people and businesses” as enterprise deployment competition heats up.
Developed with chipmaker Broadcom, Jalapeño is an application-specific integrated circuit, or ASIC, optimized for AI inference – the computationally intensive process of generating responses from models such as ChatGPT. The chip should ship later this year, according to an OpenAI blog post.
The move places OpenAI alongside Google, Amazon, Microsoft and Meta, all of which have developed custom AI chips to reduce dependence on Nvidia, whose graphics processors remain the dominant hardware platform for AI. As AI demand has surged, access to Nvidia chips has become one of the industry’s biggest constraints.
OpenAI said Jalapeño was designed from the ground up for large language model inference rather than adapted from earlier computing architectures. The startup claims that early testing shows the chip delivers “substantially better” performance per watt than current state-of-the-art systems, though it has not yet released benchmark results. A technical report is expected later this year.
The announcement marks the first tangible product from a broader partnership OpenAI and Broadcom announced last year to build 10 gigawatts of custom AI infrastructure, a project intended to support the massive computing demands of future AI models.
For OpenAI, the chip is about more than performance. Inference has become one of the company’s largest operating expenses as ChatGPT usage climbs. OpenAI recently reported more than 800 million weekly active users, creating enormous demand for computing power. Custom silicon could allow the company to lower serving costs while improving speed and reliability.
Broadcom CEO Hock Tan said the collaboration is the beginning of a multi-generation roadmap aimed at supporting gigawatt-scale AI data centers. Tan told Reuters that Jalapeño’s performance is comparable to Nvidia’s Blackwell processors and Google’s TPU chips, though OpenAI will initially use the hardware internally rather than sell it to customers.
OpenAI said Jalapeño was designed in just nine months, a timetable it described as one of the fastest advanced semiconductor development cycles on record. OpenAI executives said AI models themselves helped accelerate portions of the chip design process.