China Drops $295 Billion on a National AI Compute Grid: 80% Domestic Chips, Nvidia Squeezed Out
While American tech giants burn cash building their own separate data centers, China has decided to wire the entire country's compute into one network.
Three key facts
China is drafting a 2 trillion yuan ($295 billion) five-year AI infrastructure plan. According to Bloomberg on June 9, key agencies including the National Development and Reform Commission are drawing up a blueprint to connect data centers scattered across the country into a unified, interconnected compute network by 2028. State-owned enterprises such as China Mobile and China Telecom will operate the bulk of the data centers and ensure connectivity. The entire project is financed through sovereign debt and state-backed funds — essentially the "whole-nation system" approach applied to AI infrastructure.
The core requirement: at least 80% of technology equipment must come from domestic suppliers, with Huawei as the biggest winner. The plan explicitly mandates reliance on homegrown companies for critical technology including AI chips, effectively handing the vast majority of orders to domestic chipmakers like Huawei and squeezing Nvidia out of China's largest single AI procurement. If power grid integration is included, total investment could exceed 5 trillion yuan ($740 billion). This is not just an infrastructure project — it is a supply-chain de-Americanization campaign.
But the number is still modest compared to the U.S. $295 billion sounds staggering, but spread over five years that's $59 billion per year. In 2026 alone, Meta and Microsoft are each spending over $60 billion and $80 billion respectively on AI capex. U.S. tech companies combined will pour more than $725 billion into AI this year. China's advantage lies in concentration — pooling scattered resources under unified coordination. Its disadvantage: domestic chip fabrication still trails Nvidia by at least one to two process generations.
WangDou's Take
$295 billion, five years, one national compute grid — it sounds like a mega-project with mega-swagger, but pull out a calculator and it's less impressive: $59 billion a year, which doesn't even match what Meta alone burns annually. The real story isn't the dollar figure — it's the "80% domestic" mandate. That's a straight-up declaration: Nvidia, your AI chip business in China is done. Huawei's Ascend chips currently run large model training at roughly 60–70% the efficiency of an A100. You can brute-force it by throwing twice as many chips at the problem, but don't ask about the power bills and maintenance costs. China is betting that domestic chips will close the gap within five years. If they don't, this shiny new network will be running compute that's half a beat behind.
Source: Bloomberg — China Preps $295 Billion Plan to Fund Nationwide AI Buildout