China is preparing to put about $295 billion behind AI data centers, but the bigger story is control. Beijing wants a national compute network built around domestic suppliers, with Huawei in the center and Nvidia pushed out of the state-backed lane.
China is turning artificial intelligence infrastructure into a national project, not just another spending cycle for cloud companies. A plan now under discussion would direct roughly 2 trillion yuan, or about $295 billion, into a nationwide network of connected AI data centers over the next five years.
The size of the number matters, but the design matters more. According to Reuters, citing a Bloomberg report, the National Development and Reform Commission is among the agencies drafting a blueprint for interconnected computing hubs, with China Mobile and China Telecom expected to operate much of the network. The goal is to pull China’s scattered data center buildout into a more unified system by 2028, rather than leave local projects to compete for chips, power and customers on their own.
The plan is still in early discussions, and details could change. That caveat is important. But the direction is already clear enough for chipmakers and investors to pay attention. Beijing wants at least 80% of the technology used in these AI hubs, including AI chips, to come from domestic suppliers such as Huawei. That would leave little room for Nvidia or Advanced Micro Devices in projects tied to state capital.
Huawei has spent years trying to make its Ascend chips a serious alternative to Nvidia accelerators. The hardest part has not been only raw performance. It has been ecosystem. Developers do not move away from Nvidia lightly because chips come with software tools, optimization work, networking choices and operational habits that are expensive to replace.
A state-backed data center plan changes that calculation. If telecom carriers, cloud providers and public sector users are pushed toward Chinese accelerators at scale, Huawei gets more than a large order book. It gets a protected home market where developers have a reason to make Ascend work, software teams have a reason to optimize for it, and buyers can justify the move because the policy signal is obvious.
China Is Making Decoupling Practical
For years, the US has tried to slow China’s access to advanced chips through export controls, while China has tried to reduce its reliance on imported technology. Those two forces now point in the same direction. Nvidia can still dominate much of the global AI accelerator market, but China is building a separate lane in which CUDA is no longer the default assumption and domestic supply matters more than peak benchmark performance.
That does not mean the transition will be easy. Huawei and other Chinese suppliers still face constraints in advanced manufacturing, high-bandwidth memory, interconnects and software maturity. A system built around domestic chips may need more power, more hardware and more engineering work to reach the same training performance as the best Western stacks. State funding gives Beijing room to absorb some of that inefficiency while local suppliers improve.
The financing model also matters. The reported plan would lean on sovereign debt, state investment funds, bank lending and private capital. It does not include all spending from Alibaba, Tencent and other Chinese technology groups, which means the actual AI infrastructure buildout could be broader than the headline figure suggests.
For Western hyperscalers, the comparison is uncomfortable. The United States has relied heavily on private capital from Microsoft, Amazon, Google, Meta and OpenAI partners to build AI capacity. China is blending private investment with central planning, which gives Beijing more influence over where compute is built, who operates it and which suppliers benefit.
There is a direct market risk here. Nvidia has already said it is not including China data center compute revenue in its outlook, even after US licenses were approved for certain H200 shipments, because it remains uncertain whether imports into China will be allowed. AMD faces the same political pressure. A market that once looked like a major source of future AI chip demand is becoming one where Western suppliers may have little role in state-linked projects.
At the same time, China’s move does not automatically weaken Western AI infrastructure spending. It may intensify it. If Beijing is willing to put nearly $300 billion behind compute, governments and cloud companies elsewhere will face more pressure to treat power, chips, data centers and supply chains as national competitiveness issues.
The next thing to watch is execution. Data centers still need electricity, cooling, networking, skilled operators and chips that can be delivered in volume. If Huawei can meet that demand, it becomes more than a workaround for sanctions. It becomes a hardware foundation for China’s AI economy. If it cannot, Beijing may still build the network, but the gap between spending and usable capability will matter.
For now, the signal is clear. China is not waiting for the global chip market to reopen on better terms. It is using state capital to build its own compute stack, and AI infrastructure is becoming one of the clearest front lines in the wider technology split between China and the United States.
Also read: China is turning AI infrastructure into a state project • ICEYE becomes a €10 billion test for defense tech valuations • Michael Saylor is only partly right about AI hurting Bitcoin
