Nvidia must go back to the drawing board for Blackwell after chips overheat in server racks. Customers are quietly beginning to worry.
Blackwell is a worry child for Nvidia. After an initial design flaw, Nvidia is now struggling with overheating. The chips can overheat when placed in high-performance server racks with up to 72 processors, The Information learned from well-placed sources. Risk of overheating drastically degrades the chips’ performance.
Concerned customers
Nvidia is now working with its partners to modify the design of server racks so that the Blackwell chips do feel at home in them. The company reassures that this is part of the “normal development process. But The Information hears other noises from customers, where there is still some concern about whether Nvidia will be able to deliver high-performance Blackwell chips in a timely manner.
read also
Hot water, cool servers: Nvidia’s Blackwell chips as pool heaters
Nvidia previously had to recall the first Blackwell prototypes because of a design flaw that caused too low yields. Jensen Huang himself admitted that the problem was “one hundred percent Nvidia’s fault. In late October came the hoped-for news that Blackwell had been given the green light for large-scale production, with the first expected deliveries in January. It is unclear whether that timeline will be compromised.
Waiting with impatience
The big players in the AI and cloud industry are eagerly awaiting Blackwell. Companies such as Google, Microsoft and Meta have already bought up all of Blackwell’s stock before the first chip had even rolled off the assembly line. The tech industry announces even higher investments in AI after each quarter.
Nvidia benefits from this like no other company because of its dominant market position. Huang may even recently call himself the CEO of the world’s most valuable company .