US export restrictions on Nvidia chips have forced China to shift towards developing its own GPU technology, but the transition has been slow due to compatibility and cost issues. The high costs and complexity involved in switching to domestic alternatives make it challenging for data centers to adopt new hardware.
Chinese start-ups are making progress in GPU hardware and software development, but the challenges posed by incompatible systems and technology gaps remain significant. A government-backed think tank in Beijing has suggested that Chinese data centers continue to use Nvidia chips due to their relatively low costs compared to domestic alternatives.
The US export ban on Nvidia GPUs was triggered by concerns over China’s military capabilities. However, China’s AI computing power continues to grow rapidly despite the sanctions. As of 2023, China’s computing capacity increased by 27% year-on-year, reaching 230 Eflops. GPU-based computing power grew even faster, with a 70% increase over the same period.
The rapid build-out of internet data centers in China has raised concerns about overcapacity and under-utilization. While some data centers can choose Nvidia’s A100 and H100 high-performance computing units when needed, many struggle to manage hardware resources due to discrepancies in GPUs, AI accelerators, and network structures.
Source: https://www.techradar.com/pro/Chinese-data-centers-told-to-stick-to-Nvidia-chips-domestic-chips-not-compatible