OpenAI Set to Finalize First Custom Chip Design This Year
OpenAI is advancing toward its goal of reducing its reliance on Nvidia by finalizing the design of its first in-house artificial intelligence (AI) chip, sources familiar with the matter told Reuters. The company plans to send its first custom-designed chip for fabrication at Taiwan Semiconductor Manufacturing Co. (TSMC) in the coming months, marking a significant step toward mass production, which is expected to begin in 2026.
The process, referred to as “taping out,” involves sending the chip design to a factory for production. While the initial tape-out can cost tens of millions of dollars and take six months for completion, there’s no guarantee the first version of the chip will be successful. If issues arise, OpenAI would need to diagnose and repeat the tape-out process, which can delay production further.
OpenAI views this chip development as a strategic move to enhance its negotiating position with other chip suppliers. The company’s engineers plan to build upon this initial design, creating increasingly advanced processors with broader capabilities for future iterations. If the first tape-out is successful, OpenAI aims to test its custom AI chip as a potential alternative to Nvidia’s chips later this year.
OpenAI’s in-house team, led by Richard Ho, who joined from Google’s custom AI chip program, is collaborating with Broadcom to design the chip. Despite being a smaller team compared to those at tech giants like Google and Amazon, OpenAI’s chip development is progressing at a remarkable pace, outpacing the years-long efforts of other companies in the space.
Currently, Nvidia dominates the AI chip market with an 80% share, but the increasing costs and reliance on a single supplier have prompted major companies, including OpenAI, to explore alternatives. OpenAI’s custom chip is designed to train and run AI models and will initially be deployed on a limited scale. The chip will be manufactured using TSMC’s advanced 3-nanometer process technology and will feature systolic array architecture, high-bandwidth memory (HBM), and extensive networking capabilities—similar to Nvidia’s chips.
While the first chip is expected to play a limited role within OpenAI’s infrastructure, the company plans to expand its AI chip program in the future. To match the scale of Google or Amazon’s AI chip programs, OpenAI would need to expand its engineering team significantly.