Unlock Editor’s Digest Lock for Free
Huawei has begun delivering advanced artificial intelligence chips “Cluster” to Chinese clients, which are increasing orders after being cut off from Nvidia’s semiconductors due to Washington’s export restrictions.
The deep Shenzhen-based technology conglomerate sells over 10 sets of CloudMatrix 384.
Those who received their first delivery include data centers that serve high-tech Chinese companies.
Industry analysts say they were impressed by the speed at which Huawei began shipping CloudMatrix, a system that connects 384 AI processors to provide the computing power needed to develop AI models and services.
“The development of Huawei’s CloudMatrix 384 means China has an AI system that can beat Nvidia’s,” says Dylan Patel, founder of Chip Consultancy Semianalysis.
Huawei’s new products are due to Chinese customers being further restricted from Nvidia’s AI chips. The US group said this month that it will see revenues increase by $5.5 billion after President Donald Trump created an H20 chip that was tailored to comply with previous export restrictions.
Huawei is accelerating chip development to help domestic tech companies compete with global peers despite limited access to the world’s best technologies in the region.
According to a company presentation reviewed by people with knowledge of Financial Times and Matled, CloudMatrix performs significantly better than NVIDIA’s NVL72, a popular AI cluster made up of the US Tech Giants, with 72 of its GB200 chips in terms of computing power and memory.
The CloudMatrix 384 uses Huawei’s Ascend 910C chip.
However, Huawei uses a large amount of chips connected by “supernodes”, with all the processors in the set connected using optical technology, increasing the overall performance of the cluster.
In a company presentation, Huawei claimed that CloudMatrix outperformed NVL72 in key metrics, saying that the calculations were 67% higher and the total memory capacity was more than three times higher.
“To improve the chip performance of the cluster, we compensate for weaker individual chip performance with advanced networking,” says Patel at Semianalysis.
Industry analysts said Huawei has leveraged its telecommunications background to improve Chip Systems’ performance.
Despite its strong metrics, the CloudMatrix 384 has several drawbacks compared to Nvidia’s rival products, industry experts said.
With the use of CloudMatrix 384, more chips are used, which means much higher energy consumption and higher electricity bills. Huawei software systems require more maintenance from experienced engineers compared to Nvidia’s CUDA, which means operating personnel costs are 3-5 times greater.
Recommended
Still, given China’s abundant power and engineers, CloudMatrix was a viable solution for clients. People with sales knowledge said, especially when they are blocked from Nvidia’s most advanced technology.
According to industry insiders, the CloudMatrix 384 sells for approximately RMB 60mn ($8.2 million) per set, with the exact price dependent on the individual contract. Analyst estimates that this is compared to about $3 million in NVIDIA’s NVL72.
Nvidia said there is a range of prices depending on the original equipment manufacturer and customer specifications.
Huawei declined to comment.
