ANALYSIS: Beijing's $295B AI Plan — The Chip War Enters a New Era

ANALYSIS: Beijing's $295B AI Plan — The Chip War Enters a New Era

Introduction: China goes all-in technologically

Two trillion yuan and one objective: dominate global AI

On June 9, 2026, Bloomberg revealed information that would reverberate through trading floors and defense ministries from Washington to Tokyo: China is preparing a spending plan of 2 trillion yuan — approximately $295 billion — over five years, to build a national network of data centers dedicated to artificial intelligence. Commissioned by the National Development and Reform Commission (NDRC), Beijing's most powerful economic planning body, this project aims to transform China into a world-leading AI infrastructure power — without Nvidia, without AMD, without any American supplier. The domestic chip coverage target is set at a minimum of 80%.

The figure of $295 billion is staggering, but to put it in perspective: the United States invests through its private companies alone (Meta, Microsoft, Amazon, Google) approximately $725 billion in AI for the year 2026 alone. But there is a fundamental difference: whereas American investment is fragmented among competing firms, China's plan is a coordinated state project, planned over five years, financed by sovereign debt, operated by state telecom companies, and supplied by identified domestic vendors. It is a model of state capitalism applied to the digital revolution — and it deserves serious analysis, not condescending dismissal.

The date and the actors: who does what in this plan

The plan, still in preliminary drafting stages when Bloomberg revealed it, involves specific actors. The NDRC writes the master plan. The two major state telecom operators — China Mobile and China Telecom — will handle the construction and operation of most data centers and ensure their interconnection into a coherent national network by 2028. Financing comes primarily from sovereign debt and very long-term government bonds exceeding ten years. The private sector (Alibaba, Tencent, ByteDance) continues its own investments in parallel — but the state portion alone represents this $295 billion plan. If power grid upgrades are included, the projected total reaches 5 trillion yuan, approximately $740 billion.

The technical architecture: an intelligent network, not just servers

From "east data west computing" to the national grid

To understand the ambition of this plan, one must know its context. Since 2021, China had been developing an initiative called "East Data West Computing" — moving computing workloads from overcrowded coastal areas to western regions (Inner Mongolia, Guizhou, Ningxia) where electricity and land are cheaper. That project had created relatively disconnected regional computing hubs. The 2026 plan goes much further: it aims to interconnect these hubs into a single national network, comparable to an intelligent electricity grid but for computing power. Like an AI smart grid: if a data center in Beijing is saturated, demand is routed toward available resources in Guizhou or Inner Mongolia.

The full interconnection target is set for 2028. This means deploying ultra-high-capacity fiber optic infrastructure between sites, harmonizing workload management systems, and creating an internal computing power market where users (companies, government agencies, researchers) access resources available anywhere in the country, transparently. It is technically ambitious — but China has already demonstrated its ability to deploy network infrastructure at this scale.

The 80% domestic chip mandate: the real crux of the issue

The most important clause in the plan — and the most geopolitically explosive — is the domestic procurement mandate of at least 80% for AI chips and key components. This mandate is not a preference or an aspirational goal: it is a procurement rule that structurally excludes Nvidia and AMD from the majority of contracts. And this does not contradict American export controls — it doubles down from below. Even if Washington were to ease its restrictions tomorrow, the Chinese plan would still limit foreign chip purchases to a maximum of 20%.

The timing is significant. In November 2025, Beijing had already banned state-funded projects from using foreign accelerators, even ordering some ongoing sites to remove Nvidia, AMD and Intel equipment already installed. In May 2026, China officially approved nine categories of domestic AI chips for public procurement and sensitive sectors. These decisions foreshadowed the June 2026 plan — they laid its regulatory foundations.

Huawei crowned: how the Shenzhen firm prevailed

The rise of Ascend chips in the context of sanctions

The great beneficiary of the $295 billion AI plan has a name: Huawei. The Shenzhen company, under pressure from American sanctions since 2020, has methodically developed its AI chip ecosystem under the Ascend brand. The Ascend 910 series — notably the 910C and the new 950PR — is now the leading alternative to Nvidia in the Chinese AI accelerator market. According to market data cited by Bloomberg, Huawei projects approximately $12 billion in revenues from its chips in 2026, a year-on-year increase of 60%.

These figures are spectacular. ByteDance alone placed orders for $5.6 billion in Huawei chips for 2026. Alibaba and Tencent are also customers. The Huawei Ascend 950PR chip reportedly outperforms the only Nvidia chip still authorized for sale in China — the H20, a deliberately throttled chip designed to comply with American export controls — by 2.8 times. In other words, even on raw performance metrics, Huawei is now competitive. This was not true eighteen months ago.

The software ecosystem: the alternative to Nvidia's CUDA

Chips are one thing. The software ecosystem is another — and that is where Nvidia long held an almost insurmountable lead. Nvidia's CUDA platform, developed since 2006, is the de facto standard for GPU programming. Millions of developers, thousands of tools, decades of optimization. Huawei has been developing its own software stack for several years — CANN (Compute Architecture for Neural Networks) — as an alternative to CUDA. It is also expanding its Atlas cluster range to enable large-scale deployments.

The stakes are enormous: if an entire generation of Chinese engineers learns to develop on the CANN/Huawei stack rather than on CUDA/Nvidia, technological dependence on the American ecosystem erodes permanently. Nvidia CEO Jensen Huang himself acknowledged in May 2026 that his company had "largely conceded" the Chinese market. A rare statement in its candor — and one that illustrates how much Beijing's plan is perceived as an irreversible reality even by its primary losers.

Nvidia excluded: the numbers of an announced rout

Jensen Huang and the admission of defeat

The statement by Jensen Huang, CEO of Nvidia, is worth its weight in strategic gold. In May 2026, speaking to investors, he stated that Nvidia had "largely conceded" the Chinese market for advanced AI accelerators. This is not modesty — it is an accounting assessment. China had represented up to 20% of Nvidia's total data center revenues. That market is now structurally closed.

Nvidia's situation in China combines two obstacles. First, American export controls prohibit the sale of the best chips (H100, H200, B100) to the Chinese AI market. Only the H20, a deliberately throttled version, remains authorized. Second, the 2026 plan mandates 80% domestic chips, which mechanically limits the market share accessible to any foreign supplier to a maximum of 20%. These two combined locks create a structural ceiling that Jensen Huang deemed insurmountable — hence his public admission.

The paradox of the lost "continent-market"

Citi analysts calculated the implications of the Chinese plan. Assuming a unit cost of approximately 200 million yuan per megawatt, the plan implies approximately 10 gigawatts of new AI capacity over five years, or approximately 2 gigawatts per year. This would represent a 50% expansion of installed AI capacity in China compared to the current estimated installed base of 20 gigawatts. Every megawatt that escapes Nvidia represents hundreds of millions of dollars in lost revenues. Over five years, the total impact could amount to tens of billions of dollars in missed revenue.

The paradox for Nvidia is that in the short term, the loss is limited: global demand for its chips far exceeds supply, and every chip not sold in China finds a buyer elsewhere within minutes. But in the long term, losing China means losing an entire "continent-market" of developers who will learn on alternative technology stacks. In twenty years, the impact on the global AI ecosystem could be far deeper than the immediate revenue losses alone.

Domestic beneficiaries: Huawei, Biren, Moore Threads

The ecosystem of approved Chinese chips

In May 2026, the Chinese government officially approved nine categories of domestic AI chips for public procurement and sensitive sectors. This list includes Huawei's Ascend series (910C, 950PR), Alibaba's Hanguang 800, Biren Technology's BR100 and BR104, and Moore Threads Technology's MTT S80. This is an official certification act that opens public procurement to these suppliers and positions them as the preferred vendors for the $295 billion plan.

Biren Technology and Moore Threads are less internationally known companies, but they have raised significant capital and are developing proprietary GPU architectures. Moore Threads developed chips based on the MUSA architecture (Moore Threads Unified Shader Architecture), an alternative to CUDA. These companies have benefited from massive investments by semi-public Chinese investors and a unique market window: the one created by Nvidia's exclusion. The $295 billion plan is their generational opportunity.

The real limits of domestic production

The plan's main bottleneck is not planning or financing — it is chip production. SMIC, China's leading integrated circuit foundry, was already running at over 93% capacity utilization according to some reports. The 7nm-equivalent technology that SMIC can produce remains inferior to the 3nm and 2nm of TSMC in Taiwan — the manufacturer that produces chips for Nvidia, Apple and other giants. Huawei delivered approximately 812,000 AI chips in 2025 — an impressive figure but still far below what Nvidia can produce in a few months.

China's plan must therefore simultaneously solve two challenges: massively increase the production capacity of SMIC and other local foundries, and improve the performance of domestic chips to compete with the world's best architectures. This second challenge is particularly difficult because it depends on access to advanced lithography equipment — specifically EUV (Extreme Ultraviolet Lithography) machines from ASML, whose sale to China is blocked by Dutch and American export controls.

The Western response: where are the counter-strategies?

The American Stargate program and its limits

America's response to Chinese AI ambition is called the Stargate project of OpenAI, complemented by programs from xAI (Colossus), Meta (Hyperion) and other private actors. These investments are massive — the $725 billion mentioned above for the year 2026 alone — but they are fragmented, competitive, and oriented toward short-term commercial objectives rather than long-term strategic goals. Silicon Valley invests for returns on capital, not for national security.

The real challenge for Washington is not the amount invested — on that count, the United States vastly outspends China. It is strategic coordination. The Chinese plan is a unified state project. The American plan, such as it exists, is a mosaic of competing private initiatives that sometimes contradict each other. The question is which of the two models — planned centralization or market competition — will produce better results in the AI race. History offers arguments for both camps.

The European strategy: structural lag

Europe is clearly falling behind in this race. The EU's chip program (European Chips Act), adopted in 2023 with a target of 43 billion euros in investments, has fallen behind in execution. The ambitions of 20% of global advanced chip production by 2030 appear optimistic given delays from TSMC in Germany and the collapse of the Intel project in Saxony. Europe has assets — ASML for lithography equipment, world-class research institutions — but it lacks manufacturing champions comparable to Nvidia, TSMC or Huawei. In the chip war, Europe is more a playing field than a player.

Military implications: AI in the service of the war machine

Dual use of data centers

An aspect rarely mentioned in economic analyses of the Chinese plan is its military dimension. The same data centers that will train AI models for health, transportation and urban management can also train facial recognition systems, intelligence analysis, autonomous targeting and cyber warfare. The bifurcation between civilian and military AI is a convenient fiction — in reality, the same chips, the same software architectures, the same data serve both uses.

Beijing's plan fits within the Chinese doctrine of Military-Civil Fusion, which explicitly provides that civilian technology companies make their capabilities available to the armed forces. Huawei, Alibaba, ByteDance — all have formal or informal ties with the state security apparatus. The AI computing power built in this plan will therefore potentially be available for the military needs of the PLA (People's Liberation Army). This is a dimension that Western analysts cannot afford to ignore.

North Korea, Russia and the semiconductor axis

China's AI plan also has implications for Beijing's authoritarian allies. Putin's Russia, subject to severe Western sanctions, benefits from indirect access to certain technologies via China. The domestic chip ecosystem that Beijing is developing to free itself from American restrictions could, in time, offer a parallel supply chain to sanctioned states. Not necessarily through direct technology transfer, but through the creation of an alternative technological ecosystem that structurally bypasses the order controlled by Washington.

Taiwan in the equation: the silicon nerve center

TSMC, the determining factor

Taiwan is at the heart of this technological war in a way that Western decision-makers must fully internalize. TSMC (Taiwan Semiconductor Manufacturing Company) manufactures the world's most advanced chips — those of Nvidia, Apple, Qualcomm and dozens of others. Without TSMC, the Western technological ecosystem collapses within a few years. That is why the Taiwan question is fundamentally a technological question as much as a geopolitical and democratic one.

If China succeeds in making its domestic chip ecosystem sufficiently competitive through the $295 billion plan, this reduces its need for TSMC — and therefore, potentially, part of its strategic constraint against invading Taiwan. Conversely, if this plan fails or proves too slow, China remains dependent on TSMC for its most advanced chips — which creates a brake on any military action against the island. The success or failure of China's AI plan will therefore have direct consequences on Taiwan's security and, by extension, on the stability of the entire Indo-Pacific.

The role of ASML in the dependency chain

ASML, the Dutch company that manufactures the world's only EUV lithography machines, remains the most critical link in the semiconductor chain. Without access to EUV machines, China cannot manufacture chips below approximately 7nm — which limits the performance of its AI accelerators compared to TSMC's 3nm chips. The EU and the Netherlands, under American pressure, have maintained the block on EUV machine exports to China. This technological lock is the main structural limitation of China's plan. The key question is: how long will it take China to either obtain these machines or develop an alternative technology?

Impact on financial markets and investment strategies

Chinese winners: a precise list

Citi identified the most direct potential beneficiaries of the plan. On the data center infrastructure side: GDS and VNET, which had already recorded record order books in the first quarter of 2026 (200MW for GDS, over 500MW for VNET). On the local supply chain side: IEIT Systems, Lenovo and ZTE as localized AI server manufacturers. Accelink, optical transceiver manufacturer. Chinasoft, Huawei's IT partner. These are Chinese companies listed on A-shares (Shanghai market) or in Hong Kong, to which Western institutional investors have limited but growing access.

The total investment estimated by Citi, at 200 million yuan per megawatt, implies approximately 10 gigawatts of new AI capacity, representing a roughly 50% expansion of China's installed AI computing power over five years. This figure gives a measure of the demand that will flow through the local supplier ecosystem over the coming years.

Structural losers: Nvidia, AMD, and poorly positioned investors

For Nvidia, the logic is clear: the revenue ceiling in China has been structurally redefined downward. China until recently represented between 15% and 20% of Nvidia's total data center revenues. That figure will structurally converge toward 5% or less in the years ahead. Those who bought Nvidia shares betting on a Chinese comeback as a major market are pursuing the wrong thesis. The correct thesis is that of a Nvidia that grows strongly in the rest of the world — but with a Chinese ceiling now set in stone by Beijing's legislation.

Beijing's "AI Plus" doctrine: beyond data centers

AI as a tool for controlled social transformation

The $295 billion plan is not isolated — it fits within China's policy called "AI Plus", which aims to integrate artificial intelligence into every economic and social sector: health, transportation, urban management, agriculture, education. The stated objective is an improvement in overall economic productivity. The unstated objective, identified by many analysts, is the reinforcement of the Chinese Communist Party's surveillance and social control capacity. The same algorithms that optimize traffic management in Shanghai can identify and track dissidents in Xinjiang.

This reality must be integrated into any analysis of China's AI plan. It is not a neutral technological project. It is a political, economic and military project aimed at consolidating the CCP's power domestically and projecting technological influence globally. Countries that adopt Chinese technological standards — in Africa, Southeast Asia, the Middle East — will inadvertently contribute to extending the reach of the Chinese governance model.

The Digital Silk Roads

This project articulates with the "Digital Silk Road" strategy, which foresees the export of Chinese technological infrastructure and standards to partner countries. If the data centers from the 2026 plan become global references for developing countries, this facilitates international adoption of Huawei equipment, Alibaba platforms and ByteDance applications. It is a form of technological dependency that reverses the power dynamics that Washington is trying to preserve. For African and Asian countries building their digital infrastructure, the choice between Huawei and Cisco, between Alibaba Cloud and AWS, is not merely commercial — it is geopolitical.

Scenarios for the West: possible responses

Option 1: intensify export controls

The first option for Washington is to further tighten export controls on advanced technologies to China — not only chips, but also manufacturing equipment, specialized software, precursor materials. This strategy, already pursued for several years, has shown its limits: it stimulates the development of Chinese alternative solutions (the Huawei paradox), undermines American companies (market losses for Nvidia, AMD, Intel), and requires difficult international coordination with partners (Netherlands, Japan, South Korea) whose companies have their own commercial interests in China.

The second option is frontal competition: massively invest in American semiconductor manufacturing capacity (via the CHIPS and Science Act), accelerate technology partnerships with Taiwan, Japan and South Korea, and develop next-generation chip architectures that maintain a lasting technological lead. This approach is more constructive than restriction alone, but it requires considerable public investment and policy continuity across multiple administrations — a very real challenge in the current American political context.

Option 2: Europe as an indispensable actor

Despite its lag in chip manufacturing, Europe holds unique assets. ASML is irreplaceable in the global semiconductor chain. European equipment makers (Zeiss for optics, Aixtron for materials) are also critical. The EU can leverage these positions of strength to actively participate in negotiations on export controls and global technology standards — rather than being a passive object of American-Chinese decisions. The von der Leyen Commission has begun developing a "de-risking" doctrine toward China — reducing critical dependencies without triggering a full trade war. That is a reasonable approach, but it still lacks industrial ambition.

Beijing's strategic partnerships: financing the global AI race

Exporting the Chinese AI model

The $295 billion plan is not only a domestic matter. Beijing has engaged technology partnerships with dozens of developing countries — in Southeast Asia, Africa, Latin America — to deploy its AI solutions in critical infrastructure: power grids, urban surveillance, health systems. In doing so, China exports not only its technology but also its norms, protocols and dependencies. When an African country builds its surveillance network with Hikvision cameras and Alibaba Cloud algorithms, it is not merely choosing a technical supplier — it is choosing a geopolitical ecosystem.

This technology diffusion strategy is deliberately designed to create lasting dependencies. Software updates, equipment maintenance, training of local technicians — all create anchor points that Beijing can use as levers of influence. The West, preoccupied with its own technological battles, systematically underestimates this dimension of China's strategy. While we debate TikTok, China is building the digital infrastructure of the next generation of nations.

The Digital Silk Roads in action

The Digital Silk Road — the technological component of the Belt and Road Initiative — has already invested over $79 billion in telecom and digital projects in 165 countries. With the new AI plan, this infrastructure becomes a deployment platform for artificial intelligence services that Western providers cannot reach, either for commercial reasons or because local governments prefer partners less "demanding" on governance and human rights. That is an uncomfortable reality, but it is the reality: China offers a technological alternative without democratic conditionalities, and for many governments in the Global South, it is an attractive offer.

The West must understand that the battle for global influence is now being fought in the server rooms of Lagos, Nairobi, Jakarta and Buenos Aires, not only in the research laboratories of Silicon Valley or Shenzhen. An effective response strategy must include competitive digital infrastructure offers for the Global South — a domain where Europe in particular is dramatically absent.

Data sovereignty: the invisible but decisive battle

Whoever controls data controls the world

At the heart of the AI competition lies a fundamental reality: artificial intelligence models are only as valuable as the quality and quantity of data on which they are trained. And data — health data, financial data, consumer behavior data, urban traffic data — is the natural resource of the 21st century. China holds a massive structural advantage here: with 1.4 billion inhabitants and privacy regulations far less restrictive than the European GDPR, it can collect volumes of data that Western companies cannot legally allow themselves.

This regulatory differential is one of the least discussed but most determining factors in the AI race. Chinese companies train their models on behavioral datasets of unmatched richness. Their facial recognition systems, for example, were trained on hundreds of millions of faces — a database that no Western actor could legally compile. When these models are then exported to third-party countries, they bring with them a technical superiority directly linked to this data abundance.

The European response: GDPR as brake or shield?

Europe faces a dilemma. The General Data Protection Regulation is a remarkable legislative achievement that protects the fundamental rights of European citizens. But in the context of global technological competition, it also acts as a brake on European companies' ability to develop large-scale AI models. American actors benefit from a more permissive data regime in their home market, and Chinese actors from total permissiveness. Europe must find a balance between legitimate privacy protection and the need to allow its technological ecosystem to compete on equal terms.

Solutions are emerging — federated data spaces, secure enclaves, federated learning approaches that allow model training without centralizing raw data — but their deployment is still too limited and too fragmented to close the gap. A large-scale European data sovereignty project, comparable in ambition to the $295 billion Chinese plan, presents itself as a strategic urgency.

The role of universities and fundamental research in China's AI doctrine

The silent army of researchers

Behind the spectacular announcements of $295 billion, there is a more discreet but equally impressive reality: China now produces more STEM graduates (science, technology, engineering, mathematics) than all Western countries combined. Its universities — Tsinghua, Peking University, USTC — now regularly appear in global rankings for AI publications. The number of Chinese researchers publishing in major international conferences (NeurIPS, ICML, ICLR) has quintupled over ten years. This human capital is the true long-term engine of Beijing's AI hegemony.

It is particularly striking that many of these researchers were trained in American universities — MIT, Stanford, Carnegie Mellon — before returning to China, attracted by competitive salary packages, massive research funding and, sometimes, a sense of patriotic mission. This "brain return" dynamic is deliberately encouraged by government programs like the Thousand Talents Plan. The West has, for decades, trained the engineers now building the technological competition threatening it.

Investing in people, not only in hardware

Faced with this reality, the Western response cannot be limited to semiconductor export restrictions or investments in chip fabrication plants. It must include a massive strategy for attracting and retaining AI talent — facilitated immigration policies for foreign researchers, significant increases in academic salaries, fundamental research funding commensurate with the stakes. The United States has taken some steps in this direction with the CHIPS and Science Act, but Europe remains dramatically behind on the question of technological human capital.

The AI competition is, fundamentally, a competition of minds. Data centers, chips, infrastructure investments — all are necessary but not sufficient. What makes the difference, in the end, is a society's capacity to produce, attract and retain the world's best minds in its research laboratories. On that front, Beijing has a clear and coherent strategy. The West is still improvising.

Conclusion: a historic turning point in the global AI race

What Beijing's $295 billion plan actually changes

Beijing's $295 billion AI plan represents a turning point in the global technology war. It is not simply a large investment in data centers. It is the formalization of a technological emancipation strategy at the scale of a continent-state. It demonstrates that the state-directed economic model can mobilize colossal resources with a speed and coherence that market democracies can only rival if they consent to their own coordinated efforts.

The message delivered by Jensen Huang — "we have largely conceded" China — will not remain an anomaly. Other American technology companies will progressively realize that the Chinese market is structurally closed to them, not by accident but by political design. The global technological bifurcation is underway. The question is no longer whether it will happen — it is already here. The question is how the West will organize its response, and at what speed.

An invitation to take the threat seriously

The risk would be to underestimate this plan — to analyze it as a Chinese bureaucratic effort doomed to inefficiency and waste. That would be to repeat the mistakes made with 5G deployment, with the rise of the electric battery industry, with the ascent of Chinese electric vehicle manufacturers. China does not win every time. But when it commits to a national objective, it mobilizes resources that Western analysts have systematically underestimated. The $295 billion plan deserves to be taken seriously — not with panic, but with urgent strategic clarity.

By Maxime Marquette, columnist

Columnist's transparency note

Biases and positioning

This analysis is written from a pro-democratic Western perspective. I am convinced that China represents a systemic threat to the international liberal order — not out of xenophobia, but from observing the behavior of the Chinese Communist Party toward its own citizens, Hong Kong, Tibet, the Uyghurs and Taiwan. I acknowledge that this perspective influences my analytical framing. Analysts with a different lens might reach different conclusions about the implications of China's AI plan.

What I don't know — and my method

I am not an expert in semiconductors nor in large-scale AI system architecture. The technical data I use comes from journalistic and financial sources, not from specialized academic publications. The assessment of China's actual chip production capacity is the subject of expert debate, and estimates vary significantly. What I present as facts are pieces of information drawn from verifiable sources cited in the Sources section — they reflect the state of available knowledge in June 2026, not an absolute truth.

Sources

Primary sources

Bloomberg — China prepares $295 billion plan to build nationwide AI data center network — June 9, 2026

ABHS — China plans $295B AI data center grid, Nvidia locked out, domestic chips — June 23, 2026

TheUnum.io — China plans to invest $295 billion in national data center network without Nvidia and AMD — June 13, 2026

AShare Insights — The $295 billion bet: how China's national AI compute plan reshapes global capex — June 10, 2026

Secondary sources

Tom's Hardware — China drafts $295 billion plan for national AI data center grid running on 80% domestic chips — June 10, 2026

Investing.com — These stocks will benefit from China's AI ecosystem localisation — June 9, 2026

Bloomberg Newsletter — China's AI infrastructure plan: $295 billion project to challenge the US — June 12, 2026

TeamAI — Guide to DeepSeek models 2026, including the Chinese domestic AI ecosystem — June 25, 2026