Home / Technology / China's $295B AI Chip Grid Plan Revealed
China's $295B AI Chip Grid Plan Revealed
17 Jun
Summary
- China plans a 2 trillion yuan computing network by 2028.
- The plan mandates 80% domestic technology for AI infrastructure.
- Domestic chips lag international leaders by 5-10 years.
China is reportedly developing a colossal plan to invest approximately 2 trillion yuan (about $295 billion) into a national AI computing network. This initiative aims to connect data centers across the country into a unified grid, with major state-backed telecommunications firms overseeing its operation. The goal is to have at least 80% of the required technology, including AI chips and infrastructure, sourced from Chinese suppliers.
The ambitious blueprint, spearheaded by the National Development and Reform Commission, intends to link this national computing platform by 2028. Financing is expected to rely on sovereign borrowing and long-term government bonds, with potential additional costs for power grid upgrades that could push total investment to over 5 trillion yuan (about $738 billion).
This development follows Beijing's increasing restrictions on foreign semiconductor products for AI facilities. New regulations mandate that data centers source at least 50% of their chips domestically, with additional exclusions for foreign accelerators in state-funded projects. These policies aim to boost Chinese chip companies like Huawei, reducing reliance on international suppliers.
Despite these efforts, domestic semiconductor manufacturing remains a significant hurdle. China's most advanced manufacturing processes are estimated to be several years behind global leaders, and production capacity is already strained. High-bandwidth memory shortages further limit the assembly of advanced AI accelerators.
Industry experts suggest that even with substantial growth, domestic suppliers may only meet about 76% of China's AI chip demand by 2030. Chinese industry executives acknowledge that domestic AI data center chips are still 5 to 10 years behind international competitors in critical areas. Some AI training tasks have reportedly reverted to using Nvidia hardware due to limitations with Chinese alternatives, highlighting ongoing challenges in supporting the most demanding AI workloads.
