China's $295 Billion AI Data Centre Grid Mandates 80 Percent Domestic Chips, Locks Out Nvidia

[Amber78]

China's National Development and Reform Commission is drafting a 2 trillion yuan plan, approximately 295 billion dollars, to build a unified national AI computing grid over five years. State-owned China Mobile and China Telecom will operate the bulk of the data centres and connect them into a single national network by 2028. The mandate requires that at least 80 percent of all technology used, including AI accelerator chips, must come from domestic suppliers. That requirement effectively terminates Nvidia and AMD's access to the largest new computing procurement in the world.

The practical effect is to guarantee a captive market at unprecedented scale for Huawei's Ascend chips and domestic alternatives from Alibaba, Biren Technology and Moore Threads. Nvidia confirmed in its most recent quarterly filing that no Data Centre Hopper product shipments occurred to China during the period, compared with 4.6 billion dollars in the same period a year earlier. The China market has not contracted in the usual sense. It has been legislated away. Huawei projects around 12 billion dollars in AI processor revenue for 2026, a roughly 60 percent increase from prior year, and this mandate could dwarf even that trajectory if supply constraints can be overcome.

The scale comparison to the US is instructive. Meta, Microsoft, Google and Amazon combined are spending around 725 billion dollars on AI infrastructure in 2026 alone. China's plan is 295 billion over five years, but it is state-directed rather than market-driven. Analysts note that when power grid upgrades are factored in, the total capital requirement could exceed 5 trillion yuan. The plan formalises a two-stack world for AI infrastructure where Western and Chinese compute ecosystems are fundamentally incompatible.

China drafts $295 billion plan to build national AI data center grid running on 80% homemade silicon - projected 2028 timeline could run into limits of local chip productionwww.tomshardware.com

[Mike]

The 80 percent domestic mandate is the key number. At this scale that guarantee of demand is what makes Huawei Ascend chip production economically viable at scale. Export controls have accidentally created a domestic market that would not otherwise exist

[Blake_73]

Five years to build a unified national computing grid is genuinely ambitious. The Eastern Data Western Computing initiative has been running for years and is still not fully interconnected. 2028 seems optimistic

[Anthony]

Nvidia losing 9 percent of revenue from China and not being able to replace it is a real problem. The H20 chip designed specifically to comply with export rules is also now restricted. They have no compliant product for China

[ECWAlfie47]

DeepSeek running V4-Pro on Huawei Ascend hardware is the proof of concept the NDRC needed to make this decision. If the flagship Chinese frontier model runs domestically then the infrastructure plan is credible

[DiamondDallas86]

The power grid requirement is the sleeper issue. Chinese data centre power demand rising by 300 to 500 billion kilowatt-hours is a fifth of the country's total power demand growth. The grid has to scale first

[Taker92]

For any company operating in both the US and Chinese AI markets the bifurcation is now funded policy not theoretical risk. You need two completely different infrastructure architectures and they cannot share much

[Megan34]

I would be very cautious about the 295 billion headline. Chinese infrastructure spending announcements have historically been aspirational rather than committed. The actual deployment may take longer and cost differently

[Neil57]

This is the response to export controls that the US government should have modelled more carefully. Restricting chip access did not slow Chinese AI. It forced China to build an indigenous supply chain that will outlast any export control regime