.png)
April 29, 2025
China is pursuing ambitious goals in the field of Artificial Intelligence (AI) and is relying on a comprehensive strategy that goes far beyond the development of individual high-performance chips. At the center of this strategy is the expansion of large-scale AI infrastructures, where system size and computing power in combination are prioritized. A prominent example of this is the Chinese technology group Huawei, which, despite ongoing US sanctions, has made remarkable progress in the development of AI hardware and software.
Huawei is working intensively on the development of new AI processors in the Ascend series. According to reports, the Ascend 910D is currently in the testing phase and is intended to replace Nvidia's powerful GPUs in the future. While the Ascend 910D is expected to compete with the Nvidia H100 in terms of computing power, it is expected to have significantly higher energy consumption. Huawei is relying on innovative packaging technologies to connect multiple silicon chips and increase performance.
In parallel, Huawei has introduced the CloudMatrix 384 system, a rack-scale system based on the predecessor chip Ascend 910C. This system networks 384 of these chips and thereby achieves impressive computing power, significantly exceeding that of Nvidia's GB200 NVL72 system. The CloudMatrix system is also ahead in terms of memory capacity and bandwidth. However, it requires many times the energy compared to the Nvidia system.
A special feature of the CloudMatrix 384 is the fully optical connection technology. Huawei completely dispenses with copper cables and instead relies on optical transceivers. This architecture enables high bandwidth and low latency, but also places higher demands on the cost and complexity of the system.
Despite the considerable progress, Huawei remains dependent on foreign suppliers for chip manufacturing. The production of the Ascend chips reportedly took place at TSMC in Taiwan, and Huawei is also reliant on foreign sources for High Bandwidth Memory (HBM). This highlights the ongoing challenges arising from the US sanctions.
Although China's largest chip manufacturer SMIC has expanded its production capacity, there is still a technological gap to leading manufacturers. The further development of the Chinese chip industry depends significantly on future export restrictions and investments in research and development.
Given the existing limitations, Huawei is focusing on system-level optimization instead of maximizing the performance of individual chips. The desired computing power is to be achieved by building large, networked systems. The high energy consumption of these systems is accepted in China given the political priority of technological independence.
Huawei's strategy illustrates the Chinese approach to taking a leading role in the field of AI by expanding data centers and developing its own software ecosystems. The focus is on creating a comprehensive AI infrastructure that will form the basis for innovations in various application areas.
Bibliography: - https://cointelegraph.com/news/china-based-huawei-to-test-ai-chip-aiming-to-rival-nvidia-repor - https://semianalysis.com/2025/04/16/huawei-ai-cloudmatrix-384-chinas-answer-to-nvidia-gb200-nvl72/ - https://www.aljazeera.com/news/2025/4/27/chinas-huawei-develops-new-ai-chip-to-compete-with-uss-nvidia-report - https://techstartups.com/2025/04/28/huaweis-ascend-910c-system-reportedly-outperforms-nvidias-h100-in-key-metrics/ - https://timesofindia.indiatimes.com/technology/tech-news/innovation-not-just-chips-huaweis-strategy-for-chinas-ai-dominance/articleshow/111531424.cms - https://ifp.org/the-h20-problem/ - https://the-decoder.com/ - https://www.sify.com/ai-analytics/unbelievable-how-chinas-outsmarting-us-chip-ban-to-dominate-ai/ - https://www.csis.org/analysis/deepseek-huawei-export-controls-and-future-us-china-ai-race