The U.S. government also said earlier in the day it would further restrict AI chip and technology exports, potentially hurting Nvidia’s sales.
The first shipments of racks with Blackwell chips have been overheating and exhibiting glitches in the way chips connect to one another, the Information reported.
A rack, used in data centers, is a structure that houses chips, cables and other essential equipment.
Major customers Microsoft, Amazon.com’s cloud unit, Alphabet’s Google, and Meta Platforms have cut some orders of Nvidia’s Blackwell GB200 racks, according to the report.
Discover the stories of your interest
Nvidia, Microsoft, Google, Meta and Amazon did not immediately respond to Reuters’ requests for comment. The so-called hyperscalers had each placed Blackwell rack orders worth $10 billion or more, the report said.
Some of the customers are waiting to snag a later version of the racks or plan to purchase the company’s older AI chips, according to the report.
Microsoft was initially planning to install GB200 racks with at least 50,000 Blackwell chips in one of its Phoenix facilities, the report added.
However, key partner OpenAI asked Microsoft to provide it with an older generation of Nvidia’s ‘Hopper’ chips as delays popped up, the report said.
It is unclear how the order cuts would impact Nvidia’s sales as there may be other buyers for the “glitchy” GB200 server racks, the report said.
The company is on track to exceed an earlier target of recording several billion dollars in revenue from Blackwell chips in its fourth fiscal quarter, CEO Jensen Huang said in November.
Huang had also denied earlier media reports of a flagship liquid-cooled server containing 72 of the new chips experiencing overheating issues during initial testing.
