Nvidia’s top clients in the AI chip sector, such as Microsoft, Amazon, Google, and Meta, are reportedly postponing their orders for the upcoming Blackwell GPUs due to concerns about overheating. Despite these issues, there is confidence that the newly announced RTX 50 gaming GPUs, which are also based on the same architecture, will not face similar problems.

According to The Information, customers have opted to delay orders or revert to Nvidia’s previous generation, Hopper, amidst these overheating concerns and glitches.

Jensen Huang, Nvidia’s CEO, acknowledged last October that the Blackwell AI GPUs had design flaws which hindered the shipping of AI racks. He noted, “The design flaw caused the yield to be low. It was 100% Nvidia’s fault.”

Yet, he did not specifically mention overheating issues, instead referring to the number of functional chips produced from a single wafer.

Reports of overheating related to the Blackwell series have been around since last November, indicating that Nvidia’s liquid-cooled racks containing 72 Blackwell GPUs have undergone multiple redesigns to resolve these issues.

It is unclear if the overheating problems are linked to the design flaws mentioned by Huang, but the results of these issues remain vital to monitor as Nvidia prepares to unveil the RTX 50 gaming GPUs at CES.