The world’s most advanced semiconductors have become the most contested commodity in geopolitics. And the fight over who gets them — and who doesn’t — is reshaping alliances, supply chains, and the balance of technological power between the United States and China.
The Biden administration’s sweeping export controls on AI chips, first imposed in October 2022 and tightened repeatedly since, were designed to deny Beijing access to the computing horsepower needed to train frontier artificial intelligence models. The Trump administration has largely maintained and in some cases intensified these restrictions. But the consequences are proving far messier than Washington anticipated. What started as a targeted denial strategy has metastasized into a global regulatory thicket that’s frustrating allies, emboldening adversaries, and forcing the semiconductor industry into contortions that may permanently alter its economics.
The Architecture of Denial — and Its Unintended Consequences
The logic behind U.S. chip export controls is straightforward: advanced AI requires advanced chips, and if you control the chips, you control the pace of AI development. Nvidia’s H100 and its successors, manufactured by TSMC in Taiwan, represent the gold standard for training large language models. By restricting their sale to China, Washington aimed to freeze Beijing’s AI ambitions in place — or at least slow them dramatically.
It hasn’t worked out that cleanly.
As the Financial Times has reported extensively, Chinese firms have found workarounds. Huawei’s Ascend 910B processor, while not matching Nvidia’s top-end chips in raw performance, has proven capable enough for many AI training tasks. Chinese cloud providers have stockpiled chips purchased before restrictions took effect. And a gray market for diverted Nvidia GPUs has flourished across Southeast Asia and the Middle East, with chips originally sold to data centers in Singapore or the UAE finding their way to Chinese buyers.
The smuggling problem is real and persistent. U.S. Commerce Department officials have acknowledged that enforcement remains a challenge, particularly in jurisdictions where transshipment is difficult to monitor. Nvidia itself has been forced to design China-specific chips — the H800, then the downgraded A800 — that comply with export thresholds, only to see Washington move the goalposts and restrict those products too.
Meanwhile, China’s domestic chip industry is receiving unprecedented state support. Beijing has poured tens of billions into semiconductor manufacturing through its “Big Fund” and related initiatives. SMIC, China’s largest foundry, has managed to produce chips at 7-nanometer process nodes — not state of the art, but far more advanced than many Western analysts expected given the restrictions on equipment sales from ASML and other lithography toolmakers.
The result is a paradox. The controls have imposed real costs on China’s AI sector, particularly at the frontier. But they’ve also accelerated China’s drive toward self-sufficiency in semiconductors, a goal Beijing had pursued halfheartedly before 2022 but now treats as an existential priority.
Short-term pain, long-term strategic risk. That’s the tradeoff Washington made, and it’s not clear the math works out.
The diplomatic fallout has been equally significant. The original October 2022 rules required the Netherlands and Japan — home to ASML and Tokyo Electron, respectively — to impose parallel restrictions on semiconductor manufacturing equipment. Both countries eventually complied, but not without friction. Dutch officials bristled at what they perceived as American extraterritorial overreach. Japanese chipmakers worried about losing their largest growth market. And South Korea’s Samsung and SK Hynix, both operating advanced memory chip fabs in China, found themselves caught between their biggest customer and their security guarantor.
The January 2025 framework introduced a tiered system that categorized countries into groups based on their perceived risk of diverting chips to China. Tier 1 nations — close U.S. allies like Japan, the Netherlands, South Korea, and Taiwan — face minimal restrictions. Tier 2 countries, including much of the Middle East and Southeast Asia, face caps on the total computing power they can import. Tier 3 is effectively an embargo list, with China and Russia at the top.
This tiered approach was supposed to bring clarity. Instead, it’s generated new grievances. Several Tier 2 countries view their classification as an insult — a suggestion that they can’t be trusted with American technology. The United Arab Emirates, which has invested billions in AI through its G42 fund and the Mohamed bin Zayed University of Artificial Intelligence, has lobbied aggressively for upgraded status. India, despite its close strategic partnership with Washington, initially found itself subject to restrictions that its officials considered inappropriate for a fellow democracy.
Nvidia’s Tightrope and the Industry’s Strategic Calculus
No company has more at stake in this fight than Nvidia. The Santa Clara chipmaker derives a significant portion of its data center revenue from international sales, and China was once its third-largest market. CEO Jensen Huang has walked a careful line — publicly supporting the national security rationale for export controls while privately warning that overly broad restrictions will simply hand market share to competitors.
Huang isn’t wrong about the competitive dynamics. Every chip Nvidia can’t sell to a Chinese customer is a chip that Huawei, or a future Chinese competitor, will try to replace. And the margins on those lost sales don’t just disappear — they come out of Nvidia’s R&D budget, potentially slowing the pace of American innovation in the very technology Washington is trying to protect.
Nvidia’s stock, which has soared on the AI boom, remains sensitive to any hint of tightening restrictions. When reports surfaced in recent months that the administration was considering further limits on chip sales to the Middle East, Nvidia shares dipped. When those concerns eased, they recovered. The company has become a barometer for the market’s assessment of how far Washington will go.
But Nvidia isn’t the only player feeling the squeeze. AMD, Intel, and a host of smaller AI chip startups are all navigating the same regulatory constraints. Broadcom and Marvell, which design custom AI accelerators for hyperscale cloud providers, face questions about whether their chips could be restricted under future rules. And the equipment makers — Applied Materials, Lam Research, KLA Corporation, and ASML — are watching their China revenues shrink as restrictions on semiconductor manufacturing tools tighten.
The financial impact is substantial. ASML reported that China accounted for roughly 29% of its sales in 2024, down from a surge in 2023 when Chinese fabs rushed to buy equipment ahead of anticipated restrictions. That pull-forward demand masked the underlying trend: a structural decline in the China market that will pressure revenues for years.
For the broader semiconductor industry, the export controls have introduced a new variable into capital allocation decisions. Building a fab costs $20 billion or more. Those investments are made on 10- to 20-year time horizons. When the rules governing who you can sell to change every six months, the risk calculus shifts dramatically. Several industry executives have said privately that regulatory uncertainty is now a bigger factor in investment decisions than interest rates or demand forecasts.
And then there’s the talent dimension. China has been aggressively recruiting semiconductor engineers, offering salaries that dwarf what’s available in Taipei, Seoul, or even Silicon Valley. Taiwan has responded with laws restricting the movement of engineers to Chinese firms, but enforcement is spotty. The U.S. has tightened visa screening for Chinese nationals working in sensitive technology areas, a policy that critics say is driving talented researchers to return to China rather than build careers in America.
The AI chip war is also accelerating a broader fragmentation of the global technology supply chain. For decades, the semiconductor industry operated on a model of extreme specialization: design in the U.S., fabrication in Taiwan, packaging and testing in Southeast Asia, equipment from the Netherlands and Japan, materials from everywhere. This model produced extraordinary efficiency and relentless cost reduction. It also created deep interdependencies that made the whole system vulnerable to geopolitical disruption.
Now, every major player is trying to reduce that vulnerability. The U.S. CHIPS Act is subsidizing domestic fab construction by TSMC, Samsung, and Intel. The EU has its own chips initiative. Japan is backing Rapidus, a startup foundry aiming to produce 2-nanometer chips by 2027. And China is throwing money at every stage of the supply chain, from silicon wafers to EDA software.
The end state won’t be full decoupling — the physics and economics of semiconductor manufacturing make that nearly impossible. But it will be a more fragmented, less efficient, and more expensive industry. Chips will cost more. Innovation may slow. And the consumers and businesses that depend on ever-cheaper computing power will eventually feel the impact.
What Comes Next
The trajectory of U.S. chip export controls depends heavily on how the AI race itself unfolds. If Chinese models continue to close the gap with American ones — as DeepSeek’s surprisingly capable open-source models suggested earlier this year — pressure will mount in Washington to tighten restrictions further. If, on the other hand, the controls appear to be working and China’s AI capabilities plateau, there may be room for a more calibrated approach.
The wild card is the Middle East. Saudi Arabia and the UAE are spending aggressively on AI infrastructure, building massive data centers and courting partnerships with American tech firms. These countries want access to the most powerful chips available. Washington wants to ensure those chips don’t end up in Chinese hands. The tension between these objectives will define the next chapter of AI export policy.
There’s also the question of whether multilateral controls can hold. The current framework depends on allied cooperation — the Netherlands restricting ASML, Japan restricting Tokyo Electron and Nikon, South Korea policing Samsung and SK Hynix. If any of these countries breaks ranks, the entire edifice weakens. And the economic incentives to break ranks are enormous. China is the world’s largest importer of semiconductors, spending more on chips than it does on oil. Walking away from that market is painful for any company, and by extension, any government.
Some industry veterans argue that the controls are already past the point of diminishing returns. Chris Miller, author of Chip War and a professor at Tufts University, has noted that export controls work best when they’re narrow, targeted, and supported by allies. Broad restrictions that affect dozens of countries and hundreds of products risk creating exactly the kind of backlash and workaround behavior that undermines their effectiveness.
Others counter that the stakes are too high for half-measures. AI isn’t just another technology — it’s a general-purpose capability that will shape military power, economic competitiveness, and societal governance for decades. Allowing China unrestricted access to the computing infrastructure that underpins AI development, this argument goes, would be a strategic error of historic proportions.
Both sides have a point. And that’s precisely what makes this problem so difficult.
The semiconductor industry spent half a century building a globally integrated supply chain that delivered miracles of miniaturization and cost reduction. Now, governments are pulling it apart in the name of national security. The chips themselves keep getting smaller and more powerful. The politics surrounding them just keep getting bigger.
WebProNews is an iEntry Publication