Huawei Technologies has launched a new artificial intelligence infrastructure architecture, which, according to the Asian country’s media, will compete with Nvidia’s offerings in addressing computing power bottlenecks.
The new brainchild of the Chinese technology giant was named the CloudMatrix 384 Supernode and was described by the company as a nuclear-level product. According to information released by the Asian country’s media, the mentioned product matches Nvidia’s NVL72 system in alleviating computing bottlenecks for artificial intelligence data centers. In this case, the journalists refer to insiders who are Huawei employees. Representatives of the company spoke to the media on condition of anonymity. The information obtained during this communication was published by journalists last Monday, April 14.
Nvidia’s NVL72 was launched in March 2024. In this case, the 72-graphics processing unit NVLink domain is used, which functions as a single, powerful GPU, providing real-time inference for large language models with trillions of parameters at speeds 30 times faster than those of previous generations.
NVLink is a high-speed interconnect technology developed by Nvidia. In this case, several graphics processing units can communicate with each other and share data more efficiently.
Huawei’s new supernode is currently deployed in the company’s data centers in Wuhu, a city in central China’s Anhui Province. This product provides computing power of 300 petaflops. At the same time, the corresponding indicator demonstrated by NVL72 is 180 petaflops. This was reported by the media, referring to information received from Huawei representatives.
The company has not yet responded to a request from reporters for official comments on the new supernode. Huawei, which has its headquarters in Shenzhen, a city in the southern Chinese province of Guangdong, is currently one of the largest technology giants in the Asian country. The activity of this company, which demonstrates the ability to make breakthroughs and be a competitive player in the area of advanced developments, is important in the context of the present trade war, which involves a high-intensity confrontation between Beijing and Washington. Against the background of the current geopolitical realities, the issue of technological sovereignty is gradually becoming more relevant and actually critically important for the Asian country. In this context, Huawei’s activities can become a kind of platform or one of the constituent elements of a platform for ensuring China’s technological autonomy, which will largely determine its economic potential and political positions, including in the international arena.
Supernodes are artificial intelligence infrastructure architectures having a large number of resources, including central processing units, neural processing units, network bandwidth, storage, and memory. It is worth clarifying that in this case a larger number of resources is implied compared to the indicators typical for standard counterparts. They can function as relay servers, increasing the overall computing performance of clusters and significantly speeding up the training of foundational models.
If the insider information is confirmed at the official level, it will be a convincing argument in favor of the claims that Huawei has made significant progress. In this context, a particularly noteworthy aspect is that the company is one of the main targets of sanctions from the United States. The CloudMatrix 384 Supernode demonstrates Huawei’s ability to independently provide itself with computing power during a period of high tension between Washington and Beijing, including in the technological dimension.
The CloudMatrix infrastructure was launched in September to meet the growing consumer demand for computing capacity amid the global boom in artificial intelligence, driven by GPT from OpenAI and other generative AI models.
The media published information according to which Huawei is cooperating with the Chinese startup SiliconFlow, which is developing a machine intelligence infrastructure. According to journalists, this collaboration allows using the CloudMatrix 384 Supernode in supporting DeepSeek-R1, reasoning model from the Hangzhou-based startup, which became the focus of global attention in January. The mentioned virtual product has become an example of how it is possible to develop performance-competitive artificial intelligence systems at costs significantly lower than the industry average.
According to media reports, Huawei’s new brainchild, using Chinese-made microcircuits, has matched the performance of Nvidia’s H100 chips, achieving a throughput of 1,920 tokens per second while maintaining a high level of accuracy.
Currently, in the Asian country, other domestic companies are also ramping up their investments in artificial intelligence infrastructure. These firms are looking to capitalize on the growing demand for machine intelligence technologies and promote the adoption of home-grown chips in China.
In February, Alibaba Group Holding committed 380 billion yuan ($52.4 billion) for capital expenditure in computing resources and artificial intelligence infrastructure over the next three years. It is worth noting that this volume of financial injections is the largest investment by a private Chinese company in a computing project in history. It is highly likely that the relevant efforts will be scaled up. This assumption is realistic since the development of artificial intelligence continues at a high-intensity pace and holds great promise.
As we have reported earlier, Huawei Revenue Increases to Near-Record High.
