Huawei has officially launched an impressive AI infrastructure system known as the CloudMatrix 384 Supernode, which it asserts competes with and surpasses Nvidia’s leading NVL72 GPU cluster in several critical aspects.
According to reports from the South China Morning Post and confirmation from SemiAnalysis, this new system represents a significant advancement in Huawei’s efforts to position itself as China’s alternative to Nvidia, especially in light of ongoing US export restrictions.
Nuclear Specs
The CloudMatrix 384 system integrates 384 Ascend 910C chips, delivering up to 300 petaflops of dense BF16 computing power. In comparison, Nvidia’s NVL72 system, which connects 72 GPUs through its proprietary NVLink technology, achieves approximately 180 petaflops. Huawei insiders have referred to this system as a “nuclear-level product,” highlighting its strategic significance for China’s AI ambitions.
The Catch
This new supernode features 3.6 times the total memory and 2.1 times the bandwidth of Nvidia’s NVL72. However, it consumes nearly four times the power, a factor that might raise concerns in Western data centers, but is less of an issue in China. With China’s growing and robust energy grid, Huawei can prioritize raw performance over energy efficiency.
Already Deployed
The initial CloudMatrix 384 racks are already operational in Huawei’s data centers in Wuhu, Anhui province. The company’s engineering expertise—particularly in networking, optics, and system integration—has enabled it to navigate export limitations on chips and components by employing intricate global sourcing strategies.
China’s AI Push
Although the Ascend 910C chips still rely on foreign suppliers like Samsung (for HBM) and TSMC (for wafers), Huawei’s achievement in launching a system that outperforms Nvidia in both computing power and memory illustrates the progress China has made in establishing a self-sufficient AI infrastructure. While the CloudMatrix 384 may not be the most energy-efficient option, for Beijing, the benefits may outweigh the costs, particularly as AI becomes increasingly vital to national competitiveness.
For more daily updates, please visit our News Section.