Key information
- Major cloud service providers are experiencing overheating issues with NVIDIA’s latest “Blackwell” racks.
- Orders worth more than $10 billion are being delayed or canceled due to these issues.
- Customers are considering alternative solutions, such as using older NVIDIA AI chips.
Major cloud service providers are adjusting their orders for NVIDIA’s ‘Blackwell’ racks due to reported overheating issues. Sources indicate that early deployments of these racks experienced connectivity issues between chips and excessive heat generation.
Some of these large customers, including Microsoft, Amazon, Google and Meta, had placed orders for Blackwell arrays worth a total of more than $10 billion. In response to these issues, some customers are choosing to postpone purchases and wait for an updated version of the racks or are considering alternative solutions like using older NVIDIA AI chips.
Impact on commercial partnerships
Microsoft’s original plan called for integrating GB200 arrays with at least 50,000 Blackwell chips into one of its Phoenix facilities. However, reports suggest that OpenAI has asked Microsoft to supply a previous generation of NVIDIA’s “Hopper” chips due to delays in deploying Blackwell racks.
This is not the first time that NVIDIA has encountered problems with its Blackwell GPUs. Previous reports have highlighted overheating issues in servers housing 72 processors, leading to multiple revisions to server rack designs by NVIDIA.
If you want access to all articles, subscribe here!
