ANALYSIS: Trump, Nvidia, and China — how technology sanctions are accelerating the enemy

ANALYSIS: Trump, Nvidia, and China — how technology sanctions are accelerating the enemy

Introduction: the $295 billion boomerang effect

The plan that changed everything

In June 2026, China's National Development and Reform Commission (NDRC) finalized an AI infrastructure plan of historic scale: 2 trillion yuan — $295 billion over five years — to build a nationwide network of interconnected AI data centers. State-run operators China Mobile and China Telecom will handle the bulk of operations. And the plan's central condition: at least 80% of the AI chips used across this network must be domestically sourced.

This decision, first reported by Bloomberg on June 9, 2026 and analyzed by Tom's Hardware, carries an immediate consequence: Nvidia and AMD are effectively shut out of the largest public AI infrastructure procurement in history. Not because their products lack performance — American chips remain technologically superior for most applications. But because U.S. sanctions forced them out, and China seized the opening to build its own alternative.

The sanctions paradox: creating what you meant to prevent

The irony is stinging. Washington banned exports of Nvidia's H100 and H200 chips to China precisely to prevent Beijing from building competitive AI infrastructure for military and surveillance use. The ban on Blackwell chips followed in November 2025. The result: China poured massive investment into its own chips — Huawei Ascend, Biren BR100, Moore Threads MTT S80, Alibaba Hanguang 800 — and now plans to spend $295 billion deploying them at scale, with an 80% domestic content requirement.

Nvidia's CEO Jensen Huang said it himself: the company has "largely ceded this market" to Huawei. China once represented at least 20% of Nvidia's data center revenue. That market is now gone. The sanctions did not stop China from developing AI — they accelerated its technological self-sufficiency.

American sanctions: a chronology of hesitation

From 2022 to 2026: escalation and contortions

The timeline of U.S. restrictions on AI chips is a textbook in strategic incoherence. In August 2022, Nvidia's A100 and H100 were banned from export to China. Nvidia responded by engineering modified versions — the A800 and H800 — calibrated to stay below export control thresholds. Those versions were banned in turn in October 2023. Then, in January 2025, the Biden administration introduced the "Framework for AI Diffusion", establishing global performance benchmarks. Trump revoked it in July 2025, then reimposed restrictions on Blackwell chips in November 2025.

Meanwhile, according to Al Jazeera, the H20 chips — the downgraded version Nvidia had designed specifically for the Chinese market — continued to ship, before being banned from state-funded data centers. A gray market for Blackwell chips took shape, with servers selling at more than double their U.S. price on the black market, according to June 2026 reports. The history of these restrictions looks less like a strategy than a series of reactive measures.

The exploited loopholes

The sanctions were full of holes from day one. Jensen Huang estimated that 75% of the chips powering AI model training in Chinese data centers were still running on Nvidia's CUDA platform in 2025–2026, and that the company had delivered more than one million export-compliant H20 chips to China since late 2024. Chinese companies were also using AWS and other American cloud providers to access H100s remotely — a loophole the U.S. government took years to close.

The American Bureau of Industry and Security had to publish a clarification in June 2026 confirming that the ban applied to foreign subsidiaries of Chinese companies — after months during which some had operated in a legal gray zone. Every delay in enforcing restrictions was a gift to Beijing to strengthen its self-reliance.

The rise of Huawei and Chinese domestic players

Huawei: the unexpected beneficiary

Huawei had been on the American trade blacklist since 2019. Cut off from access to cutting-edge technology — including TSMC's manufacturing — the company was forced to develop its own capabilities. In 2024, it unveiled the Ascend 910C — an AI processor that rivals the H100's inference capabilities by some estimates. In 2026, its AI chip revenue is projected at $12 billion, up 60% year over year.

ByteDance committed $5.6 billion in Huawei chip orders for 2026 alone. Alibaba and Tencent placed significant orders. In May 2026, the Chinese government officially approved nine categories of domestic AI chips for deployment across government and sensitive sectors — including Huawei's Ascend series, Biren Technology's BR100, Alibaba's Hanguang 800, and Moore Threads' MTT S80.

The real performance gap

It would be misleading to claim that Chinese alternatives have achieved full parity with Nvidia. Tom's Hardware notes that the 80% domestic chip requirement imposes a ceiling: China will be constrained by the production capacity of SMIC — China's domestic semiconductor foundry — which operates with less advanced equipment than TSMC or Samsung. Huawei's Ascend 910B delivers roughly 60% of the H100's performance on inference tasks — acceptable for many applications, insufficient for the most demanding models.

But the gap is narrowing. And crucially, for the AI applications China prioritizes — mass surveillance, intelligent weapons systems, Chinese-language models — current domestic chips are sufficient. Technological perfection doesn't matter for those applications: what matters is availability, cost, and independence from American supply chains.

Strategic implications: a technologically bifurcated world

Two incompatible AI ecosystems

The current trajectory leads to a world where two AI ecosystems coexist with almost no overlap. On the Western side: Nvidia, AMD, the CUDA architecture, models trained on American and allied infrastructure, applications deployed under democratic norms (in theory). On the Chinese side: Huawei Ascend, CANN (China's CUDA equivalent), $295 billion in state-owned data centers, applications optimized for the Communist Party's priorities — surveillance, social control, military superiority.

This bifurcation carries profound implications for countries that haven't yet committed to either camp. Developing nations, emerging economies, non-aligned states: which ecosystem will they turn to? The cheapest one? The one delivered without human rights conditions? China is offering its AI infrastructure through its Digital Silk Road — a significant vector of geopolitical influence fed directly by the gap created by American sanctions.

Nvidia loses $30 billion in potential market

The rupture is not only geopolitical — it is economic and large-scale for American companies. Analysts estimate that Chinese demand for advanced AI chips amounted to somewhere between $15 and $30 billion in potential orders for 2026 alone. That market is now lost to Nvidia. Trump and Jensen Huang had reportedly tried personally to convince Xi Jinping to allow Chinese companies to purchase H200 chips — they returned empty-handed, according to available reporting.

China refused the H200 not because it couldn't afford them, but because the national 80% domestic content plan makes purchasing them politically untenable. In other words: even when Washington tries to sell, Beijing refuses to buy. The sanctions have created a rupture so deep it is now irreversible for this technology cycle.

Trump and technology policy: coherence or chaos?

Revoking the Biden framework

One of Trump's first technology decisions upon returning to power was to revoke the Biden administration's "Framework for AI Diffusion" in July 2025. This framework had established precise rules on AI chip exports to different country categories. Its revocation created a period of uncertainty — months during which companies did not know exactly which rules applied, and during which chips were delivered into legal gray zones.

This decision reflects the fundamental tension in Trump's technology policy: on one side, pressure from American companies (including Nvidia) that want access to the Chinese market; on the other, the strategic necessity of limiting China's technological capabilities. Trump oscillates between these two imperatives — lifting restrictions then reimposing them, negotiating then abandoning deals — creating a permanent uncertainty that neither American companies nor Chinese customers can manage effectively.

Taiwan's chip strategy: a lesson to follow

Taiwan drew its own conclusions from the situation. According to June 2026 reports, Taipei is considering aligning its own export controls with American rules to restrict sales of advanced AI chips to all Chinese customers, and criminalizing chip smuggling — making circumvention of restrictions a criminal offense. This is a more coherent approach than American hesitation: clear rules, clear penalties, rigorous enforcement.

Taiwan's integration into the Western technological control strategy is all the more urgent because TSMC not only manufactures chips for Nvidia but also for Huawei Ascend through processes that predate the sanctions. Closing that tap — which has already partially closed — is critical to limiting the long-term progression of Chinese AI capabilities.

China's acceleration: beyond chips

The GLM-5.2 model: competition in language models

The impact of sanctions extends beyond the chips themselves. China's major technology companies — Alibaba, Tencent, ByteDance, Baidu — were forced to redirect their investments toward AI models less dependent on cutting-edge hardware. DeepSeek-V3, GLM-5.2, and other Chinese models have demonstrated performance levels comparable to the best Western models on certain tasks — using architectures optimized for the chips available, not the ideal ones.

This is not an American victory. It is a demonstration that the technological gap can be partially bridged through algorithmic innovation. Chinese engineers, denied access to the best chips, learned to do more with less. This is precisely what the sanctions had not anticipated: they restricted access to hardware but not the creativity of developers.

The $740 billion figure: when you factor in the power grid

China's $295 billion plan for AI data centers is imposing on its own. But when you add investments in integration with the national power grid, the total rises to $740 billion — and some estimates reach $800 billion by 2030 when including private investments from Alibaba, Tencent, and ByteDance. According to abhs.in, this infrastructure plan is "the largest state-directed AI infrastructure investment in history."

For comparison, the United States committed $725 billion in AI for 2026 alone, by some estimates. But American investment is private, fragmented, without centralized state coordination. China's is planned, directed, coordinated. These are two competing philosophies of innovation — and neither has clearly won yet.

The U.S.' tech allies: a unified front?

Japan, South Korea, and the Netherlands: the critical links

The semiconductor war is not strictly an American affair. American export controls on advanced chips are partially ineffective without coordination with the other key players in the supply chain. Japan controls critical equipment for chip manufacturing — notably Tokyo Electron. South Korea is home to Samsung and SK Hynix, major memory chip producers. The Netherlands is the headquarters of ASML, the world's only manufacturer of extreme ultraviolet (EUV) lithography machines essential for producing the most advanced chips.

These governments have progressively aligned their export controls with those of the United States — but with delays, exemptions, and compromises that created windows of opportunity for China. The Netherlands in particular resisted American pressure for a long time before agreeing to restrict ASML machine exports to China. Those delays allowed Chinese companies to stockpile critical equipment and advance in their mastery of alternative technologies.

Taiwan and the systemic vulnerability

Taiwan sits at the heart of the technological war's paradox: TSMC — its industrial crown jewel — manufactures some of the most advanced chips used in American military equipment. The same company that is a strategic asset for the United States in its competition with China is physically located a few dozen kilometers from the Chinese coast. This is an unprecedented geostrategic vulnerability.

If China took control of Taiwan and TSMC — even briefly, even at enormous cost — it would instantly position itself as the dominant power in the global semiconductor supply chain. This is why American and European investments to diversify chip production — factories in Texas, Arizona, Germany — are not merely economic decisions. They are strategic insurance against a catastrophic Taiwan scenario. The geography of chip production is a matter of national security.

Artificial intelligence and the next front of competition

From chips to models: AI as a strategic stakes

China's $295 billion data center plan is not merely a matter of physical infrastructure. It is an ambition for leadership in artificial intelligence — the technology that will transform military forces, economic systems, and governance apparatus in the decades ahead. Beijing has understood that AI superiority will deliver decisive advantages across domains as varied as cybersecurity, military logistics, image recognition, and large-scale disinformation.

American technology players — OpenAI, Google DeepMind, Anthropic, Meta AI — today maintain a significant lead in the most advanced AI models. But that lead is directly tied to access to the highest-performance compute chips — precisely the Nvidia GPUs that China is working to replace with its own alternatives. If China succeeds in developing competitive AI chips in the coming years, the American technological window will close faster than expected.

The $740 billion figure: infrastructure as doctrine

China's $295 billion AI data center plan is imposing on its own. When power grid integration investments are included, the total climbs to $740 billion — and some projections reach $800 billion by 2030 when private investments from Alibaba, Tencent, and ByteDance are factored in. This infrastructure plan has been described as "the largest state-directed AI infrastructure investment in history."

For comparison, the United States committed $725 billion in AI investment for 2026 alone, by some estimates. But that investment is private, fragmented, and lacks centralized state coordination. China's investment is planned, directed, and coordinated by the state. These are two competing philosophies of innovation — and neither has clearly won yet.

Conclusion: the boomerang effect and its lessons for the West

Lessons from a partial failure

The history of American technology sanctions on China since 2022 teaches several lessons. First: partial and hesitant sanctions give the adversary time to adapt — and an adversary that adapts may emerge stronger than before. Second: unilateral controls without allied coordination (Europe, Japan, South Korea, Taiwan) create gaps that the adversary exploits. Third: the private market alone cannot manage national security stakes — public-private coordination is indispensable.

China's decision to invest $295 billion in domestic chips is, paradoxically, a testament to the partial effectiveness of the sanctions: Beijing would never have invested this much had American chips been freely available. But it is also proof that the adversary has adapted — and that this massive investment could significantly erode the West's technological lead by 2030.

What the West must do now

It is too late to reverse the sanctions. They exist, they have produced their effects — both good and bad. What the West can do is coordinate more effectively. Align American, European, Japanese, and Taiwanese export control regimes. Close the cloud and gray market loopholes. Invest in its own AI infrastructure with the same strategic intensity as China. And stop allowing short-term commercial decisions by companies like Nvidia to dictate long-term national security policy. The chip war is not lost — but it is harder than it needed to be.

By Maxime Marquette, columnist

Columnist's transparency note

Who I am and my acknowledged biases

I am Maxime Marquette, columnist. I believe that the technological competition between Western democracies and China is one of the most important strategic stakes of the coming decades. My bias: I want democracies to win this competition — not for technological supremacy in itself, but because I believe democratic values are preferable to the authoritarian model of AI governance.

What I don't know and my method

I am not an expert in semiconductors or international trade law. The technical details of chip architectures, performance thresholds, and process nodes exceed my specialized knowledge. My analysis rests on sources from June 2026: Bloomberg, Tom's Hardware, abhs.in, Al Jazeera, Instawhat.ai, and others.

Sources

Primary sources

ABHS.in — China plans a $295 billion AI data center grid, Nvidia locked out — June 23, 2026

Bloomberg — China prepares $295 billion plan to fund nationwide AI buildout — June 9, 2026

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

Secondary sources

Al Jazeera — US says ban on AI chip shipments applies to Chinese firms outside China — June 1, 2026

Instawhat.ai — China's $295 billion AI buildout locks out Nvidia, backs Huawei — June 9, 2026

Gadget Review — China's $295 billion AI data center could lock out Nvidia — June 10, 2026