OpenAI is turning its artificial intelligence inward — using its own models to help design the custom silicon that will power its future infrastructure. In the latest episode of The OpenAI Podcast released Monday, OpenAI president Greg Brockman said the company’s AI systems have been applied to the design of its new in-house chip developed in partnership with Broadcom. The AI models, he explained, discovered design optimizations that “would’ve taken human engineers weeks to find.” “We’ve been able to apply our own models to designing this chip,” Brockman said. “You take components that humans have already optimized, pour compute into it, and the model comes up with its own improvements.” According to Brockman, the AI-assisted design process led to “massive area reductions” — meaning smaller, more efficient hardware — and cut weeks off the production timeline. “None of the optimizations are things human designers couldn’t have done,” he added, “but the model found 20 or so improvements that would’ve taken engineers another month to uncover.” Brockman said OpenAI has been developing chip-design expertise internally to better understand the full hardware stack behind its massive AI models. The effort comes alongside OpenAI’s new collaboration with Broadcom to co-develop custom accelerators and expand computing capacity. The companies said they plan to roll out up to 10 gigawatts of custom AI chips between 2026 and 2029, manufactured by Broadcom and deployed across OpenAI’s facilities and partner data centers. “Developing our own accelerators strengthens the broader ecosystem of partners building the capacity needed to push the frontier of AI for the benefit of all humanity,” OpenAI CEO Sam Altman said in a statement Monday.