China's $295 Billion AI Infrastructure Blitz Is Designed to Lock Out Nvidia Forever

For years, U.S. chip export controls have been described as a ceiling on China’s artificial intelligence ambitions. Beijing’s response, now formally taking shape, is to blow through that ceiling with state money. On June 9, Bloomberg reported that Chinese government agencies have drafted a blueprint to deploy roughly 2 trillion yuan — approximately $295 billion — across five years to erect a nationwide network of interconnected AI computing hubs. The plan, being coordinated by the National Development and Reform Commission, represents the most ambitious AI infrastructure initiative ever conceived by a single government, and it comes with a deliberate design feature: at least 80% of all hardware must be sourced from domestic suppliers.

That single clause effectively evicts Nvidia and AMD from one of the world’s largest potential equipment procurement contracts.

State Capital, State Goals

The funding structure reveals just how seriously Beijing is treating this initiative. Sovereign debt instruments — including ultra-long-term special government bonds that can run 30 years or more — are expected to carry the bulk of the financial load. State-backed industry funds, commercial bank loans, and private investment will fill the remainder. State-owned telecoms China Mobile and China Telecom are designated as the primary operators of the new facilities, a choice that ensures the infrastructure remains under direct government oversight rather than in the hands of commercial cloud providers like Alibaba or Tencent, whose own capital spending falls outside the 2-trillion-yuan estimate.

By 2028, the plan envisions disparate regional computing facilities — many already under construction — linked into a single coherent national network. Think of it as a sovereign cloud, disaggregated across the country but programmable as a unified resource.

When electricity grid upgrades and power transmission infrastructure are folded into the project, the price tag climbs from 2 trillion yuan to something closer to 5 trillion yuan, or roughly $700 billion at current exchange rates. That’s more than twice the annual capital expenditure of all four major U.S. hyperscalers combined.

Huawei’s Coronation

The domestic-sourcing mandate has an obvious beneficiary. Huawei, which spent years navigating U.S. entity-list restrictions, has steadily built a portfolio of AI accelerators, including the Ascend 910 series, to serve exactly this market. Nine Chinese AI chips from Huawei, Alibaba Cloud (through its in-house Hanguang line), and Shanghai-based Biren Technology recently cleared government security reviews, making them eligible for deployment in sensitive national infrastructure. That certification matters: data centers operating on behalf of state telecoms and government agencies require cleared hardware, and the list of approved vendors now has real depth.

Bloomberg Intelligence analysts noted that while the plan benefits the Chinese economy broadly, “Huawei is the main winner.” The chip giant is positioned to capture the largest single slice of hardware contracts. Domestic networking and memory players, including Cambricon and various domestic DRAM producers scaling up through government subsidies, also stand to gain.

For Nvidia, the scale of foreclosed opportunity is staggering. The company has spent years cultivating Chinese cloud and enterprise customers, and while the entity list already restricted many sales, a government-directed 80% domestic mandate goes further — it removes even the possibility of indirect access through intermediaries or foreign subsidiaries. AMD faces similar exclusion, though its China revenues were smaller to begin with.

The Chip Supply Chain Angle

The plan’s ambition runs squarely into a stubborn constraint: whether domestic suppliers can actually deliver at the required scale and quality. Huawei’s Ascend 910B, manufactured by SMIC using a mature 7nm-equivalent process, currently lags Nvidia’s H100 and H200 on raw performance. Bloomberg Intelligence analysts cautioned that “capital alone cannot solve chip constraints” — a point reinforced by independent analysts who noted that throwing money at a fabrication bottleneck does not automatically produce leading-edge silicon.

China’s semiconductor industry has made real progress, but on a timeline measured in years, not months. The AI infrastructure plan appears to be betting that domestic chip performance will improve sufficiently over the five-year deployment window to make the facilities viable for frontier model training — a bet that is far from certain.

Complicating matters further, Taiwan is reportedly considering legislation that would criminalize the smuggling of AI chips to China above certain computing thresholds, aligning Taipei more closely with Washington’s export control framework. If passed, that law would close another avenue that Chinese buyers have used to access restricted hardware through secondary markets.

Context: The Global Infrastructure Race

China’s announcement lands as U.S. hyperscalers are in the middle of their own unprecedented spending cycle. Microsoft, Google, Meta, and Amazon collectively announced more than $725 billion in AI capital expenditure plans for 2025 and 2026 combined — a figure that is itself historically extraordinary. The U.S. government has also begun prioritizing domestic AI infrastructure through initiatives embedded in recent executive orders.

The difference is structural. In the U.S., infrastructure spending is driven by commercial incentives and capital markets discipline, with companies deploying against revenue models and shareholder expectations. In China, the $295 billion plan is explicitly state-directed, funded through sovereign debt, and oriented toward national strategic goals rather than near-term profitability. The two models are not competing on the same terms.

That asymmetry creates a scenario analysts have described as a bifurcated AI compute stack: one built on Nvidia and hyperscaler infrastructure, accessible to most of the world; another built on Huawei and state-directed capital, accessible to China and potentially to partner nations that buy into the Chinese ecosystem.

Taiwan and the Supply Chain Squeeze

Taiwan’s potential anti-smuggling legislation deserves particular attention. For the past two years, gray-market channels — involving transshipment through third countries and foreign subsidiaries — have allowed some volume of restricted chips to flow into China despite export controls. A Taiwanese law targeting these channels, combined with existing U.S. and Dutch restrictions on advanced lithography equipment, would significantly tighten the squeeze on China’s ability to acquire frontier hardware from outside.

That pressure, paradoxically, may accelerate the domestic chip roadmap. The more effectively outside suppliers are locked out, the stronger the commercial and political case for aggressive investment in domestic alternatives — which is precisely what the $295 billion plan is designed to fund.

What It Means

The plan is not a declaration that China has solved its chip problem. It is a declaration that China intends to route around it, spending its way toward a domestic alternative that doesn’t depend on U.S. or allied supply chains. Whether that bet pays off depends on variables — SMIC’s fabrication yields, Huawei’s design pace, power grid buildout timelines — that remain deeply uncertain.

What is clear is that the global AI infrastructure landscape is now explicitly divided along geopolitical lines, with capital allocation on both sides running at a scale that would have seemed fantastical five years ago. The race isn’t just about who has the best models. It’s about who controls the compute layer underneath them — and Beijing has just made its position on that question unmistakably clear.