The Chinese technology industry received the first blow with embargoes to be part of any treatment with the United States. That mainly affected Huawei, who could no longer count on Google (read Android) as an operating system. And then the famous tariffs would arrive. But China has begun to answer. Or rather, local companies.
In the last week, Huawei has launched A new architecture for artificial intelligence (AI) that will rival the offers of the American chips giantNvidia, when addressing bottlenecks of the processing capacity.
Presented last week, The Cloudmatrix 384 supernod was described as a “high -level product” that matched the NVL72 Nvidia system In the mitigation of the processing bottlenecks in the AI data centers, according to the South Morning China Post websiteciting anonymous sources from Huawei.
The NVL72 of Nvidia, launched in March last year, has A NVLink domain of 72 graphic processing units (GPU) It works as a single and powerful GPU, which allows real -time inference for large language models (LLM) of billion parameters at speeds 30 times higher than those of previous generations.
NVLINK is a high speed interconnection technology developed by NVIDIA that allows several GPUs to communicate with each other and share data more efficiently. For its part, The new Huawei supernodcompared to the 180 Petaflops offered by the NVL72 of NVIDIA, according to the report, which cites Huawei data.
The supernodos are IA infrastructure architectures equipped with more resources —As central processing units, neuronal processing units, network bandwidth, storage and memory – that its standard counterparts.
This allows them to function as retransmission servers, improving the general computer performance of the clusters and significantly accelerating the training of the fundamental models used in AI.
If confirmed, the advance of Huawei highlights the progress of the company, sanctioned by the United States, in its objective of achieving self -sufficiency in processing power in medium of the growing technological tensions between the United States and China.
According to reports, Huawei is collaborating with the Chinese Siliconflow Infrastructure Startup for Use the Cloudmatrix 384 supernod as a Deepseek-R1 support, the chatgpt competition.
This initiative suggests that the Huawei system, which uses Chinese manufacturing chips, It has matched the performance of NVIDIA H100 chips, reaching a performance of 1920 tokens per second and maintaining a high level of precision.
What does all this mean? That Huawei is already much more independent from the United States in terms of vetoes and that it shows that It is able to match and overcome not only the restrictions, but also the technology that challenges it.