Nvidia's Blackwell AI Chips Overheating Issues

(Reuters) – Nvidia's new Blackwell AI chips, which have already faced delays, are experiencing overheating problems with their accompanying servers. This has raised concerns among customers about their ability to set up new data centers in time, according to a report by the Information on Sunday.

The overheating issue arises when Blackwell graphics processing units (GPUs) are connected in server racks that can accommodate up to 72 chips. Sources familiar with the situation indicate that Nvidia has repeatedly requested its suppliers to modify the rack designs to address these overheating problems. This information comes from Nvidia employees, including customers and suppliers who are aware of the issue, although the specific suppliers were not named.

A spokesperson for Nvidia stated, "Nvidia is working with leading cloud service providers as an integral part of our engineering team and process. The engineering iterations are normal and expected."

In March, Nvidia introduced the Blackwell chips and initially planned to start shipping them in the second quarter. However, delays may impact customers such as Meta Platforms, Alphabet's Google, and Microsoft.

The Blackwell chip architecture combines two squares of silicon into a single unit, significantly enhancing performance with speeds up to 30 times faster than previous models for tasks such as chatbot responses.