China is preparing to spend around 2 trillion yuan — roughly $295 billion — building a nationwide network of AI data centers over the next five years, according to a Bloomberg report published June 9. The plan, still being drafted by key government agencies including the National Development and Reform Commission, would represent the most coordinated state investment in AI infrastructure anywhere in the world.

The most striking detail: the build-out is designed to rely on domestic suppliers for at least 80% of AI chips — a requirement that would effectively lock out NVIDIA, AMD, and other US semiconductor companies.

What the Plan Actually Involves

State-owned telecoms giants China Mobile and China Telecom would operate the bulk of the data centers, connecting them into an inter-linked national computing grid. The funding would flow primarily through sovereign debt and ultra-long special government bonds — the same instruments Beijing has used for strategic infrastructure in the past.

Chip suppliers positioned to benefit include Huawei, Alibaba, Biren Technology, and Moore Threads, all of which received expanded government clearance in May 2026 when Beijing approved nine categories of domestically developed AI chips for deployment across government and security-sensitive sectors.

The five-year headline figure of $295 billion becomes even larger when power grid upgrades are folded in — analysts estimate the full capital requirement could exceed 5 trillion yuan if grid expansion is included.

For context: US companies like Meta, Microsoft, and Google are projected to spend $725 billion on AI infrastructure in 2026 alone. China’s plan is spread over five years, but it is also state-directed rather than market-driven — which means it will move with a different kind of speed and predictability.

Why the 80% Domestic Chip Mandate Matters

The chip requirement is not incidental to the plan — it is central to it. The ongoing US export controls on advanced semiconductors have pushed Beijing to accelerate domestic alternatives. NVIDIA’s H100 and H200 chips, which power most large-scale AI training globally, have been progressively restricted from Chinese buyers since 2022.

The May 2026 approvals of Huawei’s Ascend series and chips from Alibaba, Biren, and Moore Threads signal that Beijing believes its domestic alternatives have matured enough for serious deployment at national scale. Whether these chips can match the performance of NVIDIA’s top-tier hardware remains debated — but for inference workloads and many enterprise applications, the gap has narrowed significantly.

The practical implication is a splitting AI infrastructure market: one ecosystem anchored in US chip architecture, another anchored in Chinese silicon. Businesses operating across both geographies will increasingly face hardware and software choices that determine which ecosystem they can participate in.

The Broader Tech Cold War Context

This announcement lands against a backdrop of escalating technology competition. The Trump-Xi Beijing summit in late 2025 produced some tactical agreements, but the structural technology decoupling has continued. NVIDIA won a temporary reprieve on H200 exports to China following that summit, but the new buildout plan makes clear Beijing is not waiting on US goodwill — it is building infrastructure that does not need it.

The 5-year timeline is also notable. 2031 would see this infrastructure fully operational around the same time multiple geopolitical flashpoints around Taiwan, trade, and tech standards are expected to peak. Building sovereign AI infrastructure is, for Beijing, an insurance policy against disruption in either direction.

For AI labs and enterprise vendors, the signal is clear: the global AI market is splitting into two hardware substrates. Products, models, and services that want to operate in China will need to run on Chinese silicon. Those built on NVIDIA’s CUDA ecosystem will need alternatives.

What This Means for Business

If you run AI workloads today, your hardware choices have a geopolitical dimension. Most enterprise AI currently runs on NVIDIA infrastructure — either directly or through cloud providers like AWS, Azure, and Google Cloud that run NVIDIA hardware. For the vast majority of businesses outside China, this poses no immediate operational risk. But the long-term fragmentation of the AI chip market does carry implications worth watching.

Data sovereignty is becoming a real question. If you have operations or customers in China, the infrastructure you can run AI on will increasingly be China-specific. This affects everything from the models you can legally deploy to where your inference happens and who has access to the outputs.

The AI infrastructure arms race is accelerating on both sides. The US is investing $725 billion in AI infrastructure through private companies this year alone. China is investing $295 billion over five years through state mechanisms. Both are betting that whoever builds the most compute wins the next decade of AI development. For businesses that depend on AI capabilities, this competitive dynamic is likely to drive continued cost reduction and capability improvement in the near term — even as the underlying infrastructure splits geographically.

For data teams specifically: the China plan includes a strong emphasis on AI training workloads and large-scale model development. This could accelerate the development of Chinese-origin AI models — including open-source variants — that become competitive alternatives to US-origin models. Teams building AI systems today should track whether these Chinese alternatives become viable options for cost or compliance reasons.

The global AI infrastructure picture now includes three overlapping races: raw compute capability, energy availability, and now geopolitical alignment of the hardware itself. China’s $295 billion plan is the clearest signal yet that the third race is fully underway.