Openai: Custom chip said to reduce dependence on Nvidia

OpenAai: Custom chip said to reduce dependence on Nvidia 29 comments

Image: Infineon

In order to train and use OpenAAI’s AI models (training), the manufacturer, together with its partner Microsoft, is investing hundreds of billions of US dollars in Rake AI centers equipped with GPUs from NVIDIA. It is not only very expensive, but also risky. Custom chips could reduce the risk.

Summary :

Production of the series from 2026 continued on schedule

This reports Reuters, citing unspecified sources.

As a result, the design of a first custom AI chip in the next few months, and then handed over the design to TSMC’s production, called “Tape Out”. Neither OpenAai nor TSMC wanted to comment on this message to Reuters.

Taiwan’s UDN previously reported that Openai will have its own design at TSMC in the future. At that time it was said: next to Apple Openai would be one of the first customers who then reserved the cutting-edge TSMCS A16 process. The start of mass production can be expected in 2026.

The effects still uncertain

However, it remains to be seen whether a chip of its own development will be able to reduce dependence on Nvidia. Currently, AI industries are struggling after peak performance, and thus competitors such as AMD and Intel are finding it difficult to come up against NVIDIA. The highest performance and software that has already been established in the industries is only available there.