OpenAI to launch AI chip in 2025, challenging Nvidia

OpenAI is developing its own AI chip to reduce reliance on Nvidia, with production set to begin at TSMC in 2025 and mass production expected by 2026. The project, led by Richard Ho (ex-Google) in collaboration with Broadcom, aims to strengthen OpenAI’s position in the AI hardware market.

If successful, the chip could offer an alternative to Nvidia, which dominates 80% of AI accelerators. OpenAI’s chip program, part of the $500B U.S. Stargate initiative, will require significant scaling, as its 40-person team is far smaller than those at Google or Amazon. Estimated costs could exceed $500M, while Meta and Microsoft plan to invest $60B and $80B in AI infrastructure next year.