China Plans to Spend 2 Trillion Yuan on Data Center Network Within Five Years
Claire Weston
China is drafting a ¥2 trillion (~$295 billion) five-year plan to build an interconnected national data-center network for AI; this is Beijing's largest-ever compute-infrastructure push, aimed at a domestic power base independent of U.S. technology.
Where does the money come from?
The primary funding vehicle is sovereign bonds — ultra-long special treasury bonds with maturities exceeding 10 years, plus national strategic-industry investment funds. Bank loans and private capital fill in the rest.
Self-built data-center spending by Alibaba, Tencent, and other private firms is not included in the ¥2 trillion figure. This means → the ¥2 trillion is purely government-led incremental investment; private capital sits on top.
If grid integration is factored in, total investment could rise to at least ¥5 trillion.
Who builds it, who runs it?
The NDRC (国家发改委) and other core agencies are drafting the plan. China Mobile and China Telecom — both state-owned — will operate and interconnect the data centers.
The plan requires at least 80% of AI chips and other core technology to come from domestic suppliers, with Huawei (华为) as the anchor; Nvidia and AMD are effectively shut out.
In plain terms = this is not ordinary infrastructure — it is a nationwide compute grid running on domestic chips and state-owned operators, designed to remove U.S. technology from every link in the chain.
Are domestic chips ready?
In May this year, nine domestically made AI chips — from Huawei, Alibaba, Biren Technology (壁仞科技), and Moore Threads (摩尔线程) among others — passed a government security review, clearing them for high-security applications.
The U.S. has agreed to let Nvidia sell its previous-generation H200 chips to Chinese buyers, but shipments have not yet begun. This means → even a partial relaxation of export controls comes with real delivery lag, which widens the window for domestic substitutes.
The H200 trails Nvidia's latest Blackwell generation by roughly one technology node. Put simply = what America is willing to sell is already not the best — China can afford to wait and also cannot, so it is running both tracks at once.
What will the network actually do?
The overarching goal is to consolidate scattered data facilities into a unified national compute network by 2028.
Forrester Research chief analyst Charlie Dai says the unified network will pool fragmented regional resources, give businesses broader access to high-performance computing, and accelerate AI-model iteration and cross-industry agent services.
This reflects Beijing's deeper intent: the plan is not just about building server rooms — it is about driving large-scale AI deployment in healthcare, transport, and urban management.
Is the money enough compared with the U.S.?
¥2 trillion is roughly $295 billion, spread over five years. By contrast, Meta, Microsoft, and other U.S. tech giants have committed $725 billion in AI capital expenditure for this year alone.
But China's data-center construction costs are structurally lower — cheaper labor, components, and building costs, plus local-government subsidies. This means → the same dollar buys more compute infrastructure in China.
Whether the plan lands on schedule ultimately depends on whether the domestic chip supply chain can keep pace in both scale and performance. The proposal is still in early-stage discussions; details may change.
Content is for reference only, not financial advice.