Has OpenAI really thought through its plans to produce its own chips?

A Reuters exclusive reports that OpenAI is working on its first self-designed chip so as to reduce its heavy dependence on Nvidia.

The idea, apparently, is to finish the design by the end of this year and start mass production in 2026. For a typical production run, the company is looking at spending tens of millions of dollars, and may have to pay substantially more if it wants to speed up the manufacturing process. And even then, there is no guarantee that the chip will function properly; so should any problems arise, OpenAI would have to diagnose the problem, remake the affected parts and repeat the production process.

In short, the design and production of task-specific chips is enormously complicated and expensive. Even companies with the resources such as Microsoft or Meta have tried and failed. What’s more, chips designed by non-specialists like Nvidia for specific tasks are less adaptable to technological changes and market demands.

The main advantages for a company like OpenAI of designing its own hardware are being able to optimize performance, reducing dependence on Nvidia, and creating competitive advantages and long-term savings derived from reduced operating costs and greater efficiency. However, there are also huge risks: the first and most obvious is the high…