China strives for technological independence
China is taking a bold step to cut back on its reliance on foreign technology by directing data centers across the country to utilize more domestically produced semiconductors. Industry insiders say state-owned computer centers are now required to source over half their chips from local suppliers. The move is part of Beijing’s increasing pressure to protect its tech future as the United States places tighter export controls on sophisticated chips central to artificial intelligence.
The policy derives from regulations originally brought forward by Shanghai officials in 2023, under which the city’s intelligent computing centers were asked to use at least 50 percent locally made chips by 2025. What started as a city regulation has been promoted to a national mandate with the support of agencies like the National Development and Reform Commission (NDRC) and the Ministry of Industry and Information Technology (MIIT).
Race to build AI infrastructure
China’s drive occurs as demand for computing capacity is taking off. In order to supply its fast-growing AI sector, Beijing has stepped up the development of “intelligent computing centres” – enormous complexes that aim to cluster resources for information-intensive use. Between 2023 and 2024, over 500 new projects were launched in regions such as Inner Mongolia and Guangdong, KZ Consulting reports.
However, the problem is in hardware. AI generative models need high-end semiconductors like Nvidia’s H100 and H800 chips, which Washington has prohibited exporting to China. Although US-certified H20 chips have returned to the Chinese market in recent times, doubts regarding their possible security threats have raised more suspicion from Beijing.
In turn, Chinese companies are resorting to domestic options. Huawei and other enterprises have come up with chips such as the Ascend line, which are finding applications in local AI projects. Although these chips are still not as strong as Nvidia’s, they are managing for applications like AI inferencing. Nonetheless, for training new large models, Nvidia equipment is still the leader.
ALSO READ: Racist Tech Or Public Safety? Notting Hill Carnival’s Facial Recognition Sparks Bias Concerns
Issues of chip adaption
The transition to domestic chips has not been seamless. AI chips typically rely on proprietary ecosystems Nvidia’s CUDA or Huawei’s CANN making it difficult to run models developed on one system with another company’s hardware. For many Chinese AI developers, this has created significant “adaptation” hurdles.
Expert companies, though, are bridging the gap. Beijing’s SiliconFlow, for example, collaborated with Huawei to fine-tune DeepSeek’s R1 models on Huawei Cloud Matrix 384 architecture. Joint studies released in June indicated this platform even surpassed Nvidia’s H800 chips in certain applications.
Nonetheless, for the majority of AI data centres, the transition is turning out to be expensive and technologically daunting. A source in the industry said small players, lacking top-level engineering teams, are not being able to manage mixed sets of local and foreign chips. This has acted as a brake on adoption and compelled companies to re-strategize.
China’s long-term strategy
Beijing’s chip requirement represents more than near-term imperatives. It represents a long-term plan for achieving self-sufficiency in key technologies, particularly as the international semiconductor environment becomes more politicized. As Washington tightens controls and Chinese businesses hasten to catch up, the stakes have never been greater in the struggle for technological dominance.
