Login to Continue Learning
[Ad_1]
Chinese AI data centers are facing challenges in transitioning from NVIDIA’s AI GPUs to Huawei’s products due to software constraints, according to the South China Morning Post. The Chinese government has mandated that publicly funded AI data centers use at least 50% domestic chips to reduce reliance on foreign chips. This requirement originated from Shanghai municipality guidelines last year, which initially required city computing centers to use 50% domestic chips. These quotas were made mandatory nationwide in 2024.
Chinese Data Centers Struggle with Huawei’s Chips
Following the Trump administration’s approval for NVIDIA to sell its H20 GPUs to China, these chips were involved in controversies suggesting possible backdoors or vulnerabilities. While NVIDIA denied such claims, reports suggested that the Chinese government was cautious about foreign hardware. Soon, it became apparent that China was wary of relying too much on foreign chips for AI computing needs—a point raised by NVIDIA CEO Jensen Huang when advocating for export control restrictions.
The latest report from the South China Morning Post indicates that state-run or owned computing infrastructures are now required to use at least 50% domestically procured chips. This mandate stems from Shanghai’s rules introduced in 2024.
The latest Chinese chips designed by Huawei and manufactured by SMIC can serve as substitutes for NVIDIA’s hardware. However, US sanctions prevent Huawei from relying on TSMC, the only alternative. Consequently, these domestic chips are limited to a 7-nanometer process due to restrictions on EUV equipment sales to SMIC.
NVIDIA’s GPUs run on the CUDA platform, while Huawei’s rely on CANN. This compatibility issue is causing headaches for cluster operators whose AI applications were developed using NVIDIA’s chips but now need to be compatible with Huawei’s chips.
[Ad_2]
