NVIDIA AI Chip Issues Trigger Order Cuts from Microsoft and Others

On January 13, according to foreign media reports, NVIDIA's latest-generation AI chip, Blackwell, encountered technical issues during deployment in data centers, including server rack overheating and abnormal chip connections.

These challenges have disrupted deployment progress, leading several major clients of NVIDIA, such as Microsoft, Amazon's AWS, Google, and Meta, to cut portions of their orders for Blackwell GB200 racks.

The Blackwell chip, NVIDIA's next-generation graphics processing unit (GPU), has been highly anticipated for its exceptional performance and energy efficiency. Compared to the previous Hopper generation, Blackwell offers a fourfold improvement in energy efficiency, attracting orders exceeding $10 billion each from tech giants like Microsoft, Amazon, Google, and Meta.

Reports indicate that such defects are not uncommon for new chip technologies but have delayed data center plans for clients like Microsoft. Sources familiar with the matter revealed that Microsoft, a key server provider for OpenAI, had initially planned to install at least 50,000 GB200 racks featuring Blackwell chips at its Phoenix, Arizona facility. However, due to delays in Blackwell's delivery since last year, OpenAI requested Microsoft to provide the previous-generation NVIDIA H200 chips instead. As a result, the Phoenix data center, originally slated for GB200 installations, is now fully equipped with H200 chips.

Insiders warn that if NVIDIA fails to address these issues, the performance of the Blackwell chips may fall short of the company's promises. It remains unclear whether the reduction in orders from clients like Microsoft will impact NVIDIA's overall sales, as other buyers may potentially acquire the affected GB200 server racks.

NVIDIA CEO Jensen Huang stated in November 2024 that Blackwell chips had entered full-scale production and were expected to remain in high demand for several quarters. Huang also projected that sales of Blackwell chips in the latest fiscal quarter would surpass the company's earlier targets. Notably, prior media reports mentioned overheating issues during the initial testing of a flagship liquid-cooled server equipped with 72 new chips, but Huang denied these claims.

Meanwhile, on January 13, the U.S. government announced new export controls on artificial intelligence (AI) technologies, which analysts believe could impact major U.S. chipmakers, including NVIDIA.

According to Nikkei Asia, NVIDIA issued a statement on January 13 criticizing the Biden administration's new export restrictions on advanced AI semiconductors. The company argued that the measures would "hinder technological innovation and economic growth." The restrictions limit U.S. companies' exports of semiconductors to regions such as Southeast Asia and the Middle East, with NVIDIA's advanced AI chips expected to fall under the scope of these controls. NVIDIA emphasized that such measures would not enhance U.S. national security but instead weaken the country's competitiveness.