NVIDIA is probably one of the biggest suppliers of hardware used by AI companies to process and run their models. However, OpenAI might not want to be overly reliant on NVIDIA. According to a report out of Taiwan, OpenAI CEO Sam Altman has reportedly paid Taiwan’s TSMC a “secret visit” to create an in-house chip.
OpenAI wants TSMC to build their own AI chip
According to the report, the OpenAI CEO visited both TSMC and Foxconn to discuss its “Stargate” project as well as the potential of creating their own in-house chip. It was previously rumored that OpenAI’s chip project was to be developed in collaboration with Broadcom.
However, ever since OpenAI snagged some key players from Google’s TPU project, the company has set its sights on a more ambitious plan, which is to create their own chip designs from scratch. Unfortunately, apart from that, we don’t really have the specifics of what kind of chip OpenAI has in mind.
It is expected that the company will most likely use TSMC’s 3nm process with integration expected as soon as 2026. Whether or not it will be able to successfully replace NVIDIA’s chips remains to be seen. However, it’s not hard to see why OpenAI wants to go down this path.
Why switch from NVIDIA?
The problem with relying on other companies is that you’re more or less at their mercy when it comes to the advancement of technology. This was an issue that companies like Apple faced when it used Intel’s CPU for their Mac computers. It prevented Apple from creating the computers they wanted when they wanted.
It is why the Cupertino company ultimately decided to create its own chips, resulting in Apple Silicon. This could be one of the reasons why OpenAI wants to create its own chips. Designing its own chips could allow OpenAI to create AI models specific to what they want. That way they control both the hardware and software, similar to Apple’s iPhones and Google’s Pixel phones.