According to informed sources, Nvidia's (NVDA.US) new generation Blackwell ai chip has serious overheating issues in high-capacity server racks. These problems have led to design adjustments and project delays, causing major customers such as Google, Meta, and Microsoft to worry about whether they can deploy Blackwell servers as planned.

Informed sources revealed that the Blackwell gpu is designed specifically for ai and high-performance computing (hpc), but has encountered overheating issues in servers with 72 processors, with power consumption of up to 120 kilowatts per rack. The overheating issues have forced Nvidia to make multiple revisions to the rack design, limiting gpu performance and potentially damaging hardware. Customers are concerned that these technical issues will delay the deployment of processors in data centers.

To address this challenge, Nvidia has requested suppliers to adjust the rack design and work with partners to optimize the cooling system. While engineering improvements like these are a standard part of large-scale technology releases, they have further delayed the product delivery timeline.

In response to this matter, an Nvidia spokesperson stated that the company is closely collaborating with cloud computing service providers, and that design adjustments are a normal part of the research and development process. Nvidia hopes that through this collaboration, the final product will meet performance and reliability expectations, while expediting the resolution of technical bottlenecks.

The revised Blackwell gpu only entered mass production at the end of October this year, with the earliest shipment expected by the end of January next year. Google, Meta, Microsoft, and other technology giants rely on Nvidia gpus to train their most powerful ai models, making the delay inevitable and impacting these customers' research and product release plans.

In March of this year, Nvidia showcased the Blackwell chip, stating at the time that shipments would begin in the second quarter.