Nvidia's new AI chips face overheating issues: report

Nvidia's highly anticipated Blackwell AI chips are reportedly facing overheating issues, raising concerns about potential delays for customers relying on the technology.

According to a report by global news platform The Information, the overheating occurs when the chips are installed in server racks designed to hold up to 72 units.  

The issue has led Nvidia to repeatedly ask its suppliers to redesign the server racks in an effort to address the problem. These changes are happening late in the production process, according to company employees, customers, and suppliers cited in the report. However, Nvidia has not yet notified customers of any official delay. 

According to the report, a company spokesperson acknowledged the ongoing engineering challenges but described such iterations as "normal and expected", emphasising Nvidia's collaboration with leading cloud service providers to resolve the matter.  

The Blackwell chips, announced in March 2024, represent a significant leap in performance, boasting speeds up to 30 times faster for AI tasks like chatbot responses. Nvidia initially planned to ship the chips in the second quarter of 2024, but earlier delays have already pushed back the timeline.  

Major tech companies like Meta, Google, and Microsoft, which rely on Nvidia's chips for their AI operations, may now face concerns about whether they will have sufficient time to prepare their data centres.  

While Nvidia has not commented on whether the rack designs are finalised, the overheating issue adds to existing delays in the release of the Blackwell chip lineup, impacting its rollout to the broader market.