OpenAI recently reportedly acquired TSMC's process capacity through Broadcom and plans to produce AI-powered inference systems in 2026.Homemade chips, relevant news indicates that the internal design of this homemade chip will be completed in the next few months and is expected to be put into production through TSMC.
OpenAI's first self-made chip is expected to be used for running artificial intelligence models. It will be produced using TSMC's 3nm process and equipped with HBM high-bandwidth memory, which is expected to reduce its heavy reliance on NVIDIA.
As for general processors, it takes Dazhihui about half a year from design to production, and the cost may involve tens of millions of dollars. In order to speed up chip production, the industry may need to pay more funds to increase R&D and production pace. Therefore, it means that OpenAI may invest more funds in its first self-made chip.
In addition, this self-made chip is also expected to become a "bargaining chip" for OpenAI to negotiate cooperation with chip manufacturers such as NVIDIA, and it will also be able to widen the technological development gap with more artificial intelligence companies.
Previously, OpenAI had considered building its own fab to ensure chip supply flexibility and reduce holding costs through diversified supply, but it is now clearly more inclined to cooperate with external parties to produce its own chips more efficiently.
However, OpenAI continues to cooperate with NVIDIA and plans to introduce accelerators based on its new "Blackwell" display architecture. It also continues to use AMD EPYC processors for Microsoft Azure cloud services to accelerate the operation of artificial intelligence services.
