NVIDIA Faces Delay in Deliveries as Blackwell Chip Overheating Leads to Billion-Dollar Order Cuts

January 14, 2025 – According to a report published by tech media outlet The Information on January 13th, NVIDIA has once again delayed the delivery of its data center AI chips due to overheating issues in racks equipped with the latest Blackwell chips and interconnection failures within the chips.

Racks are structures used in data centers to house chips, cables, and other essential equipment. The report, citing various sources, revealed that major customers including Microsoft, Amazon Web Services, Google, and Meta have reduced their orders for NVIDIA’s Blackwell GB200 racks, with each company’s order amount reaching or exceeding $10 billion.

The report indicated that some clients are waiting for subsequent rack versions or planning to purchase NVIDIA’s older AI chips. For instance, Microsoft initially intended to install GB200 racks containing at least 50,000 Blackwell chips in its Phoenix facility. However, due to the delay, its key partner OpenAI requested Microsoft to provide the previous generation “Hopper” chips instead.

Currently, it remains unclear how significantly the reduction in orders from major customers will impact NVIDIA’s order book. The media outlet believes that NVIDIA will still find buyers for its chips once the faulty GB200 server racks are repaired. In November, NVIDIA CEO Jen-Hsun Huang stated that the company expects to generate billions of dollars in revenue from Blackwell chips in the fourth fiscal quarter, exceeding previous targets.

Leave a Reply