The US government pledged $500 billion for Stargate. China just answered with $295 billion of its own, and the terms reveal something the dollar figure alone does not: this plan is designed specifically to work without Nvidia. On June 9, 2026, Bloomberg reported that China's National Development and Reform Commission is drafting a blueprint to spend 2 trillion yuan, approximately $295 billion, over the next five years on a nationwide AI data center network. The requirement that at least 80% of the chips powering those facilities come from domestic suppliers like Huawei is not a preference or a goal. It is a stated design constraint for the world's second-largest AI infrastructure buildout.
What Actually Happened
According to Bloomberg's reporting, the National Development and Reform Commission, China's top economic planning body, is in the process of drafting a blueprint for spending approximately 2 trillion yuan ($295 billion) over the next five years to build AI data centers across the country. The state-operated telecommunications giants China Mobile and China Telecom are expected to operate the bulk of the facilities and ensure they connect into a coherent national network. The target completion date for the initial interconnected system is 2028. The plan is not yet finalized, but the involvement of the NDRC signals a level of political commitment that typically precedes formal announcement in China's policy process.
The domestic chip requirement is the key operational detail. The plan specifies that at least 80% of the AI chips and supporting technology in the new facilities must come from domestic suppliers, with Huawei Technologies and its Ascend chip family as the primary intended vendor. This effectively excludes Nvidia, AMD, and Intel from the large majority of the buildout. It also excludes the US-designed network and storage hardware that typically accompanies Nvidia's H100 and H200 GPU installations. The financing structure relies primarily on sovereign debt instruments, including ultra-long-term special government bonds, along with state-backed industry funds, with commercial loans and private investment filling in the remainder. One estimate of the total cost, when electricity grid upgrades are included, puts the complete infrastructure price tag at more than 5 trillion yuan, equivalent to roughly $700 billion at current exchange rates.
The broader policy context is important. China's AI infrastructure buildout follows a multi-year escalation. The government's "14th Five-Year Plan" established AI as a strategic priority. The current initiative is an order of magnitude larger in both capital commitment and geographic scope. State firms including China Mobile, which operates more than 1 billion mobile subscribers, and China Telecom are not passive infrastructure providers here: they are designated as the primary operators of a network that the NDRC envisions as a unified national compute fabric. By 2028, the plan envisions disparate data facilities linked into a single coherent system that can allocate compute resources dynamically across regions, a capability the US does not have at the federal level and that would give Beijing direct operational control over a large share of domestic AI training capacity.
Why This Matters More Than People Think
The surface reading of this story is a dollar-for-dollar infrastructure arms race: US Stargate at $500 billion, China at $295 billion, Europe at roughly $120 billion, Japan adding $10 billion. The deeper reading is about chip stack divergence. If China builds $295 billion worth of AI infrastructure on Huawei's Ascend chips, trained on Chinese-developed software stacks and Chinese-compiled model frameworks, the result is not just a physical separation of data centers. It is a technical bifurcation of the entire AI development ecosystem. The models trained on Huawei Ascend hardware will be optimized for Ascend architectures. The frameworks, compilers, and tooling developed for that stack will diverge from CUDA-based tooling over time. Within a decade, comparing the performance of a Chinese-developed model against a US-developed model may require translating between two genuinely incompatible compute paradigms.
The market implication for Nvidia is direct and material. China represented approximately 17% of Nvidia's data center revenue before the US export control restrictions on high-end chips took effect. The $295 billion Chinese buildout, with its 80% domestic chip mandate, effectively forecloses the possibility of Nvidia recapturing that revenue through a future policy change. Even if US-China trade relations normalized tomorrow, the NDRC blueprint would continue to push China toward domestic suppliers as a matter of stated industrial policy. Nvidia's long-term China revenue is structurally impaired, not cyclically depressed. For Nvidia's valuation, which is built in large part on the assumption that global AI infrastructure buildout will use Nvidia hardware as the default, the Chinese domestic chip mandate is a permanent ceiling on its addressable market, not a temporary restriction.
The strategic logic for China is clear. After US export controls blocked access to Nvidia's A100 and H100 chips in 2022 and 2023, and after subsequent restrictions tightened on alternative chips, Beijing faced a choice: wait for diplomatic resolution or build a domestic supply chain capable of meeting national AI infrastructure demand. The $295 billion plan is the implementation of the second option at a scale that makes it irreversible. Once China Mobile and China Telecom have built and staffed data centers full of Huawei Ascend hardware, the institutional knowledge, maintenance contracts, training data pipelines, and software tooling built around that hardware will create their own lock-in. China is not just buying chips: it is building a generation of engineers and infrastructure operators who know only the domestic stack.
The Competitive Landscape
The US Stargate initiative, announced in January 2026 with a $500 billion commitment from OpenAI, SoftBank, Oracle, and the federal government, is the most direct comparable. Stargate is larger by dollar commitment, but its structure is different: it is primarily a private-sector consortium with government support, not a state-directed buildout. The Stargate facilities will run Nvidia and AMD hardware by default. The US approach bets on private capital markets and commercial competition to drive efficiency. China's approach bets on state coordination and domestic supply chain development to drive independence. These are not the same bet, and the 2028-2030 window will reveal which approach produces more usable compute capacity per dollar invested.
Huawei's Ascend 910C chip is the central variable in the Chinese plan's credibility. The Ascend 910C was developed specifically in response to US export restrictions and represents Huawei's most competitive AI accelerator. Independent benchmarks have placed its performance at roughly 60-70% of Nvidia's H100 on standard training workloads, with higher memory bandwidth but lower raw floating-point throughput. The gap means that $295 billion of Huawei chips will not produce $295 billion worth of Nvidia-equivalent compute. The actual compute output of the Chinese buildout may be closer to $175-200 billion in Nvidia-equivalent terms, which changes the calculus of the arms race. The historical parallel is China's high-speed rail network: built with domestic technology that critics called inferior to Japanese and German alternatives, but scaled to the point where domestic engineering experience and operational refinement narrowed the gap within a decade.
The European AI infrastructure picture is a third track. France, Germany, and allied nations have committed to a collective 110 billion euro AI data center buildout, with the EU AI Act providing regulatory structure that neither the US nor China has matched at the federal level. Europe is not a principal in the US-China chip competition: European facilities will use Nvidia hardware by default and are not subject to the domestic-chip mandates that define China's plan. But the European buildout adds to the global demand for AI infrastructure that creates supply chain pressure on the firms manufacturing advanced chips. TSMC, which manufactures both Nvidia's GPUs and some of Huawei's Ascend chips, is the single point of potential constraint in this multi-front buildout race.
Hidden Insight: The Chip Independence Bet Is a Generational Wager
The $295 billion figure is striking, but the more consequential number in the Bloomberg report is 80%. An 80% domestic chip mandate at this scale is not a procurement preference: it is a declaration that China intends to create a self-sustaining AI compute ecosystem independent of US semiconductor supply chains. To understand why this matters beyond the immediate competitive context, consider what happens to Chinese AI researchers, engineers, and companies over the next five years as this infrastructure gets built. They will develop their models on Ascend hardware. They will write code in frameworks optimized for Ascend. They will build careers in an ecosystem where Nvidia's CUDA is irrelevant, where the dominant hardware vendor is Huawei, and where the performance benchmarks are set by domestically-developed models running on domestically-built chips. That is a generation of human capital formation that cannot be reversed by a trade agreement.
The financing structure reveals Beijing's conviction level. Ultra-long-term special government bonds, which have maturities of 20 to 50 years, are not instruments used for speculative or reversible investments. They are used for infrastructure that governments intend to hold and operate for decades: bridges, power grids, transportation networks. Using them for AI data centers signals that the Chinese government is treating AI compute infrastructure as permanent public infrastructure, in the same category as the Three Gorges Dam or the national highway network. That framing has implications for how Western governments and companies should think about competition: they are not racing against a speculative bubble or a short-term policy priority. They are racing against an institution-building project backed by sovereign debt instruments.
The wildcard is the electricity grid. The Bloomberg report notes that when grid upgrade costs are included, the total could reach 5 trillion yuan or more. AI data centers are extraordinarily power-intensive: a single large-scale facility can consume as much electricity as a small city. China's grid infrastructure in many inland provinces where land costs are lower is not currently capable of supporting the power density required for modern AI training clusters. The grid upgrade component is not a rounding error: it may represent 60% of the total cost. And unlike the data centers themselves, grid upgrades create infrastructure that benefits every sector of the economy, not just AI. China is using the AI buildout as justification for a grid modernization program that it needed regardless, and structuring the financing accordingly.
The bear case, however, is straightforward. Critics argue that the $295 billion planning target should be evaluated against China's track record on semiconductor self-sufficiency goals. The Made in China 2025 program set ambitious domestic chip targets that were publicly announced and widely analyzed. By 2025, China's domestic chip production met roughly 20-25% of its own demand, well below the 70% target. Skeptics point out that building data centers with domestic chips is much easier than building competitive domestic chips in the first place, and that Huawei's Ascend supply chain is itself dependent on TSMC-manufactured components that could be restricted through additional export controls. The 80% domestic chip mandate assumes Huawei can scale Ascend production to meet the demand of a $295 billion buildout, and that assumption has not been tested at anything close to this volume.
What to Watch Next
The most critical 30-day indicator is the formal publication of the NDRC blueprint. Bloomberg's report describes a document in drafting, not yet officially released. Once the blueprint is published, the specific procurement targets, timeline milestones, and domestic chip volume requirements will become available for independent analysis. Analysts at Goldman Sachs and HSBC have both flagged the Chinese infrastructure buildout as a key variable in their Nvidia revenue forecasts: the formal publication will trigger a revision cycle that will move Nvidia's stock price and the valuations of every US semiconductor company in the supply chain.
The 90-day indicator is Huawei's Ascend production capacity announcement. For the 80% domestic chip mandate to be credible, Huawei needs to demonstrate that it can manufacture Ascend chips at a volume consistent with $295 billion in facility buildout. Current estimates put Huawei's Ascend 910C production at roughly 200,000 to 500,000 units per year. A $295 billion buildout at the scale described would require several million chip-equivalent compute units over five years. Huawei has not publicly committed to a production ramp timeline that matches the NDRC's stated ambition. Watch for any Huawei manufacturing announcements, factory expansion disclosures, or supply chain partner statements over the next quarter.
The 180-day indicator is US export control response. The Biden administration introduced multiple rounds of AI chip export restrictions aimed at China. The Trump administration has maintained and in some cases expanded those controls. The Chinese $295 billion domestic buildout is a direct consequence of those restrictions and, paradoxically, a signal that they are working: China is investing at this scale precisely because it cannot rely on US chip supply. The policy question for the US government over the next six months is whether the domestic chip mandate changes the calculus for further export control tightening. If China is already committed to building without Nvidia, tightening restrictions further may cost US companies revenue without achieving additional strategic containment. The $295 billion plan changes the strategic logic of the export control debate in ways that will play out in Washington policy circles well into 2027.
China's $295 billion AI buildout is not a response to Stargate: it is a bet that chip independence, at sovereign scale, is more durable than any trade relationship.
Key Takeaways
- 2 trillion yuan ($295 billion) over 5 years: China's NDRC is drafting the world's second-largest AI infrastructure commitment, operated by China Mobile and China Telecom
- 80% domestic chip mandate: Huawei's Ascend family must supply the large majority of AI chips, effectively excluding Nvidia and AMD from the buildout by design
- Nationwide interconnected network by 2028: the plan envisions a unified compute fabric linking data centers across Chinese provinces, giving Beijing direct operational control over domestic AI training capacity
- Total cost could reach 5 trillion yuan with grid upgrades: power infrastructure expansion pushes the true program cost to an estimated $700 billion equivalent, financed by ultra-long-term sovereign bonds
- Nvidia's China revenue is structurally impaired: with 80% domestic chip mandates in a $295 billion program, the prospect of Nvidia recapturing Chinese data center market share through policy normalization is now structurally foreclosed
Questions Worth Asking
- If China successfully builds 80% of a $295 billion AI infrastructure on Huawei chips by 2028, and a generation of Chinese engineers develops their skills on that stack, how long before the Ascend ecosystem produces models that compete with US-developed models trained on Nvidia hardware?
- The Made in China 2025 semiconductor targets fell well short of stated goals: is the $295 billion AI data center plan more credible because it requires building facilities, which China has proven it can do at scale, rather than fabricating chips, which it has not?
- If China's AI infrastructure becomes genuinely independent and competitive, and the US maintains export controls regardless, has the export control policy achieved its strategic goal or inadvertently accelerated the outcome it was designed to prevent?
