OpenAI is making significant progress in developing its first proprietary AI chip, which could go into production as early as later this year. The company is partnering with TSMC to manufacture this chip, which should help reduce dependence on Nvidia chips.
So reports Reuters based on sources. OpenAI previously desired to develop its own AI processors because of the global shortage of specialized processors. The project is led by Richard Ho, who previously worked on Google’s custom AI chip program.
Collaboration with TSMC and Broadcom
The chip is being developed by an internal team of 40 people, in collaboration with Broadcom. Progress is now said to be smooth. Accordingly, the company expects the designs to be sent to TSMC within a few months, a process also known as taping out. TSMC will handle the manufacturing.
TSMC will produce the chip using the advanced 3-nanometer technology. Mass production should start in 2026. Normally, the tape-out process costs tens of millions and takes about six months to complete production of a chip. This process could be faster if OpenAI pays substantially more. However, it is unknown how this will play out in this case. There is a chance, however, that turning the design into a working chip will lead to delays, for example, if problems arise that need to be analyzed.
OpenAI’s first chip will be used primarily to run AI models and have a limited role in the company’s infrastructure. This is a first step in what may become a broader chip development program designed to reduce dependence on Nvidia and mitigate rising costs.
Tip: OpenAI wants to raise billions from Apple, Nvidia and Microsoft