Huawei, at the forefront of China's artificial intelligence (AI) rise, is expected to unveil its AI system 'Cloud Matrix 384' that runs generative AI models. Amid increasing U.S. export controls, it has been reported that Huawei has developed the AI system equipped with next-generation AI chips, 'Ascend 910C,' which it designed itself and mass-produced through China's largest foundry, Semiconductor Manufacturing International Corporation (SMIC). Huawei plans to accelerate the establishment of AI infrastructure capable of powering numerous domestic AI models, including DeepSeek.
According to the industry on the 22nd, Huawei is expected to showcase its self-developed AI system Cloud Matrix 384 for the first time at the World Artificial Intelligence Conference (WAIC) to be held in Shanghai, China, on the 26th (local time). Last year, Huawei announced it had developed Cloud Matrix 384 with double the performance of the NVIDIA GB200 NVL72, but this will be the first time it reveals it at an exhibition. Taiwan's Economic Daily noted, 'Currently, the technology is a generation ahead of products released by NVIDIA and AMD,' stating that 'China's AI innovation will significantly impact the global AI industry landscape.'
The computational speed, efficiency, and overall performance metrics of the Cloud Matrix 384 that Huawei will unveil are reported to be twice that of the NVIDIA GB200 NVL72. The GB200 NVL72 is produced based on NVIDIA's latest AI chip, Blackwell. While Cloud Matrix 384 is designed with Huawei's Ascend 910C, which is at one-third the performance level of Blackwell, it is understood to have improved performance by utilizing thousands of AI chips. The Ascend 910C is reported to have recorded 60% performance of the previous generation chip, H100, from the NVIDIA Blackwell series. However, the power efficiency is reportedly about 2.3 times better in NVIDIA's AI systems due to the mass deployment of AI chips.
Huawei is expected to accelerate the establishment of infrastructure that can power Chinese AI models, including DeepSeek, through this AI system. According to analysis released by Huawei in collaboration with Chinese AI infrastructure startup SiliconFlow, it was analyzed that the Cloud Matrix 384 outperformed the performance of AI systems from global big tech corporations, including NVIDIA, in running DeepSeek's AI model 'R1.'
Jensen Huang, CEO of NVIDIA, commented on Huawei's technological capabilities, saying, 'Huawei is a company with much larger and deeper capabilities than NVIDIA in terms of scale and technology,' adding that 'even if there were no NVIDIA in China, Huawei will surely find its own solutions to grow in the AI market.'
There are forecasts that the commercialization of Huawei's AI system and the easing of export controls on NVIDIA's H20 will lead to expanded investments by Chinese AI corporations. Morgan Stanley stated, 'With the lifting of U.S. export restrictions on H20, the speed of China's AI development is expected to accelerate,' predicting that 'the investments related to AI from major Chinese cloud corporations this year will reach approximately $53 billion (about 73 trillion won), an increase of around 60% from last year.'
A semiconductor industry official explained, 'China has been heavily investing to foster the AI industry, but investments had recently stagnated due to NVIDIA's H20 export controls and the slow commercialization of domestic AI chips,' adding that 'with the easing of export regulations on H20 and the acceleration in the establishment of Huawei's next-generation AI chips and associated AI infrastructure, it is projected that large-scale investments will resume.'