en.Wedoany.com Reported - China is drafting a plan to invest approximately 2 trillion yuan (about $295 billion) in building a nationwide AI computing network, connecting data centers across the country into a unified computing platform, primarily operated by state-owned telecom companies such as China Mobile and China Telecom. The National Development and Reform Commission is responsible for formulating the blueprint, aiming to complete large-scale infrastructure deployment by 2028.
Financing sources are mainly sovereign borrowing and ultra-long-term government bonds, and energy infrastructure upgrades may significantly increase total costs. If related grid upgrade expenditures are included, the total funding requirement could exceed 5 trillion yuan (about $738 billion).
Policy requirements call for a substantial increase in the localization ratio, with officials hoping that at least 80% of foundational technologies (including AI chips and supporting infrastructure) come from domestic suppliers. Starting in 2025, data centers must procure at least 50% of their chips from domestic manufacturers; in November of the same year, state-funded projects were required not to use foreign accelerators in new facilities. Projects with less than 30% completion must remove components from Nvidia, AMD, and Intel.
These measures are creating market opportunities for domestic chip companies like Huawei, while reducing reliance on overseas suppliers such as Nvidia, AMD, and Intel. The policy aims to ensure that key AI tools and large language models run on domestic hardware, but replacing imported processors still faces significant challenges.
China's semiconductor manufacturing capacity mainly relies on SMIC and a few state-approved foundries. SMIC's co-CEO Zhao Haijun has warned that excessive infrastructure expansion could lead to underutilization of production capacity. The company's most advanced stable manufacturing process is roughly equivalent to the 7-nanometer node, with current capacity utilization exceeding 93%. Due to multiple chip design companies competing for the same limited production resources, rapid capacity expansion under existing wafer quotas remains difficult.
High-bandwidth memory (HBM) constitutes a major bottleneck, limiting the number of advanced accelerators available for AI workloads and tool deployment. Industry estimates suggest that by 2030, domestic suppliers may only meet about 76% of China's AI chip demand, while the market size will expand to $67 billion. Huawei shipped an estimated 812,000 chips last year, but supply chain constraints continue to impact production scale.
Chinese industry executives say that domestic AI data center chips still lag behind international competitors by 5 to 10 years in some categories. Reports indicate that DeepSeek, after attempting to use Huawei alternatives for heavy workloads, has returned to using Nvidia hardware for certain training tasks, showing that Chinese processors still face difficulties in the most demanding AI training environments.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
