[News] NVIDIA GB200 Racks Reportedly Overheat, Major Clients Cut Orders

---

---

According to Reuters, citing The Information, NVIDIA’s major customers, cloud service providers (CSPs) such as Microsoft, Amazon, Alphabet’s Google, and Meta Platforms are reportedly scaling back orders of the company’s latest “Blackwell” racks due to overheating problems.

Initial shipments of racks with Blackwell chips have reportedly encountered overheating issues and connection glitches between chips, the report indicates, citing The Information.

As noted by the report, citing The Information, some of these customers had ordered Blackwell racks worth USD 10 billion or more. The report also notes that some customers are opting to wait for a revised version of the racks or are considering purchasing older AI chips from NVIDIA instead.

For instance, Microsoft initially planned to install GB200 racks equipped with at least 50,000 Blackwell chips in one of its Phoenix facilities. However, the report, citing The Information, mentions that OpenAI has reportedly asked Microsoft to provide an older generation of NVIDIA’s “Hopper” chips due to delays associated with the Blackwell racks.

This is not the first instance of NVIDIA encountering issues with its Blackwell GPUs. Another report from Reuters from November 2024, citing The Information, noted that Blackwell GPUs had experienced overheating problems in servers containing 72 processors. As a result, NVIDIA has reportedly made multiple revisions to its server rack designs.

Meanwhile, the U.S. government recently announced potential new restrictions on AI chip and technology exports, which, according to Reuters, could impact NVIDIA’s sales.

Read more

• [News] NVIDIA’s Blackwell GPUs Reportedly Overheat in Server Racks, Sparking Delay Concerns
• [News] NVIDIA’s GB10 Superchip Powering Project DIGITS is Reportedly Built with TSMC’s 3nm Node

(Photo credit: NVIDIA)