The leader in the global AI chip race could change if the Chinese tech giant Huawei makes a bold move. Local media outlets claim that the company’s powerful new computer system, the CloudMatrix 384 Supernode, performs better than similar technology from the massive American chip company Nvidia.
If the performance claims prove to be accurate, the AI hardware breakthrough could transform the technological landscape despite US efforts to limit China’s access to cutting-edge technology.
300 petaflops: Undermining Nvidia's hegemony in hardware
As reported by STAR Market Daily and referenced by the South China Morning Post (SCMP), the CloudMatrix 384 Supernode is a “nuclear-level product”. The hardware surpasses Nvidia’s NVL72 system’s 180 petaflops of computational capability with an astounding 300 petaflops.
In order to solve the computing bottlenecks that have grown more difficult as artificial intelligence models continue to expand in size and complexity, the CloudMatrix 384 Supernode was specially designed.
The system is made to directly compete with Nvidia’s products, which up until now had controlled the global market for hardware for AI accelerators. Originally introduced in September 2024, Huawei’s CloudMatrix infrastructure was created especially to satisfy the growing demand in the Chinese domestic market.
According to reports, the 384 Supernode variant is the most potent AI architecture implementation to date. It can reportedly match the performance of Nvidia’s H100 chips, but with Chinese-made components, and achieve a throughput of 1,920 tokens per second while maintaining high levels of accuracy.
Under sanctions: The technological accomplishment
The fact that the AI hardware breakthrough was made in spite of the harsh technological limitations Huawei has been subjected to since being included to the US Entity List is what makes it so noteworthy.
Due to restrictions on the company’s access to cutting-edge US semiconductor technology and design software, Huawei has been forced to adopt other strategies and depend more on domestic supply chains.
The primary technological innovation behind the CloudMatrix 384’s performance seems to be Huawei’s response to Nvidia’s NVLink, a high-speed interconnect system that facilitates effective communication across several GPUs.
Released in March 2024, Nvidia’s NVL72 system boasts a 72-GPU NVLink domain that works as a single, potent GPU, allowing for real-time inference for trillion-parameter models at speeds that are 30 times quicker than those of earlier generations.
The SCMP reports that Huawei is working with SiliconFlow, a Chinese AI infrastructure startup, to deploy the CloudMatrix 384 Supernode in support of DeepSeek-R1, a reasoning model developed by Hangzhou-based DeepSeek.
Supernodes are AI infrastructure architectures that have more resources than typical systems, such as storage, memory, network bandwidth, improved central processing units, and neural processing units.
Because of the arrangement, they can serve as relay servers, which improves clusters’ total processing performance and speeds up the training of basic AI models considerably.
China’s wider push for AI infrastructure goes beyond Huawei.
Huawei’s AI hardware innovation is not unique; rather, it is a component of a larger movement by Chinese tech firms to develop AI computer infrastructure in their own countries.
Alibaba Group, a big e-commerce corporation, said in February that it would invest 380 billion yuan ($52.4 billion) over three years in AI infrastructure and computer resources. This is the greatest investment in computing that a private Chinese company has ever made.
The development of strong substitutes for Nvidia’s hardware may someday help the global AI community overcome the computational constraints that have impeded AI progress. Competition in this area may expand the amount of processing power that is accessible and provide developers additional choices for how to train and implement their models.
It’s important to remember, though, that Huawei had not yet addressed inquiries about these allegations at the time the report was released.
A major step toward China’s goal of technical self-sufficiency, Huawei’s CloudMatrix 384 Supernode comes as tensions between the US and China continue to rise in the technology industry.
Despite being heavily sanctioned, Huawei has gained compute independence in this area if the performance claims are confirmed by this AI hardware breakthrough.
A wider trend in China’s technology sector is also indicated by this development, as some local companies are increasing their investments in AI infrastructure in an effort to take advantage of the rising demand and encourage the use of domestic chips.