Nvidia announced that it would take a $5.5 billion financial hit after U.S. authorities imposed new export restrictions on its H20 HGX AI GPU designed for the Chinese market.

The U.S. government cited H20's memory and interconnect bandwidth as well as its potential use in supercomputers as reasons for restrictions. Nvidia's H20 HGX is a cut-down version of its H100 processor, specifically engineered to comply with U.S. government restrictions imposed on GPUs bound for China in late 2022.

Compared to H100, the H20 features significantly reduced AI performance and dramatically reduced HPC performance, making it barely suitable for supercomputer workloads that require FP64 compute precision. However, H20 HGX still carries 96GB of HBM3 memory with a peak bandwidth of 4 TB/s and 900 GB/s NVLink interconnections to build 8-way AI servers, thus providing competitive performance in AI workloads.

As a result, H20 HGX has become rather popular in China. Chinese companies like Alibaba, ByteDance, and Tencent have deployed large AI training and inference clusters based on these GPUs. Earlier this month, it was reported that Chinese tech giants spent $16 billion on H20 HGX GPUs in the first quarter ahead of expected export restrictions by the U.S. government starting in May.

In addition to restricting sales of Nvidia's H20 GPUs to Chinese customers, the U.S. Commerce Department also restricted shipments of AMD's Instinct MI308 to China and other D:5 countries. Since AMD is considerably behind Nvidia in sales of AI GPUs, this will barely have a drastic effect on the company's business.