OpenAI announced on Wednesday its first in-house computer chip, named Jalapeno, created to operate ChatGPT and other AI systems with greater speed and lower cost. Developed together with Broadcom, the chip targets AI inference, which involves running trained models to produce user responses rather than the resource-heavy training phase. Early tests indicate Jalapeno offers substantially improved performance per watt compared with current leading options, according to an OpenAI blog post. The firm noted that its own AI systems assisted in the chip’s design, shortening development time. Jalapeno supports various AI models beyond OpenAI’s and will enter use at Microsoft and partner data centers starting in 2026. By designing its own chips, OpenAI seeks to lessen reliance on external vendors like Nvidia. Broadcom’s chief executive described the project as an initial step, with further product generations planned. The partnership between the companies began last year to create specialized AI processors and reduce dependence on Nvidia technology. This announcement forms part of OpenAI’s efforts to maintain leadership in generative AI amid growing competition. High-performance chips remain essential for AI development, with most firms relying on Nvidia and similar suppliers. Other large technology companies have also developed custom chips in recent years to improve efficiency and reduce expenses.
