Huawei Technologies of China recently debuted new artificial intelligence infrastructure architecture that claims to solve computing power bottlenecks similarly to the United States chip giant Nvidia. The company introduced the CloudMatrix 384 Supernode last week, calling it a "nuclear-level product," and stated it delivered the same Computer Power as Nvidia' NVL72 as a AI Data centre to solve computational Bottlenecks, STAR Market Daily reported Monday citing unnamed ads from the Huawei.

Nvidia's NVL72 was unveiled in March 2022, and has a 72-graphics processing unit (GPU) NVLink domain that operates as one GPU with real time inference to trillion parameter large language models (LLMs) at a speed 30 times than prior generations.

NVLink is an interconnect technology developed by Nvidia that allows multiple GPUs to work together more efficiently by enabling communication and sharing of data between systems at a closer and faster pace.

Huawei's new supernode deployed in Wuhu (a city in central Anhui province where Huawei has data centres), achieved 300 petaflops of computing power versus Nvidia's NVL72 model which achieved 180 petaflops according to Huawei data cited in the report. Huawei did not respond immediately on Tuesday to a request for comment.

Supernodes refer to AI infrastructure architectures that have more resources, i.e., central processing units, neural processing units, network bandwidth, storage and memory, than regular versions. Supernodes can act as relay servers, improving the cluster's overall computing performance and accelerating foundational model training. If verified, Huawei's improvement demonstrates the US-sanctioned company moving ahead in computing power self-sufficiency at a time when US-China tensions grow over technology. The company, based in Shenzhen, launched its CloudMatrix infrastructure in September in response to the growing demand for computing capacity amid the world's AI boom, manifestly produced by OpenAI's GPT and other generative AI models.