In the autumn of 2022, a relatively obscure regulatory update from the U.S. Commerce Department sent shockwaves through boardrooms in Santa Clara, Beijing, and Taipei. The Biden administration had quietly drawn a new kind of weapon: not a missile, not a tariff, but a technical specification. By defining performance thresholds measured in teraflops and memory bandwidth, American officials decided which graphics processing units could legally leave the country and which ones could not. Overnight, chips became geopolitical instruments.
Four years later, that weapon has been sharpened, contested, circumvented, and debated with an intensity that rivals any conventional arms control debate. The story of semiconductor export controls is not simply a trade dispute. It is a fundamental argument about whether advanced AI capabilities can be contained at all, who gets to make that call, and what happens to the global technology industry caught in between.
The Architecture of Restriction
The foundational logic of U.S. chip export controls rests on a straightforward premise: modern AI systems require enormous computational power, that power is concentrated in a handful of advanced chips, those chips are manufactured through supply chains that run through American technology and allied territory, and therefore Washington holds meaningful leverage over who can build frontier AI.
The October 2022 rules were sweeping in scope. They targeted not just finished chips but also the equipment used to make them, and they extended licensing requirements to a broad category of semiconductor manufacturing tools. Crucially, they also imposed restrictions on U.S. persons working in China’s chip industry, a provision that prompted a quiet exodus of engineers and technical staff.
The Biden administration followed with further tightening in October 2023, introducing a tighter framework that attempted to close loopholes that had allowed slightly downgraded chips to flow to China. Nvidia’s A800 and H800, chips the company had specifically designed to sit just below earlier thresholds, were swept into the restricted category. The Commerce Department also extended entity list controls to additional Chinese firms and extended licensing requirements to a wider set of destinations seen as potential transshipment risks.
Through 2024 and into early 2025, the Biden administration continued layering restrictions, culminating in the so-called “diffusion rule,” which attempted to cap the total amount of AI computing power that could reach certain countries even through third parties. The Trump administration rescinded that rule in May 2025, before its compliance date, promising a replacement. The rules grew so complex that compliance lawyers became nearly as valuable to chipmakers as their electrical engineers.
Nvidia and the Revenue Reckoning
No company has felt the squeeze more acutely, or watched the numbers more anxiously, than Nvidia. The Santa Clara-based firm had built a commanding position in AI accelerators through its CUDA software ecosystem and its H100 and subsequent chip architectures. China had historically represented a significant portion of its data center revenue, with some estimates placing it at roughly 20 to 25 percent of that segment before the controls took effect.
CEO Jensen Huang has navigated this terrain with notable candor. He acknowledged publicly that the controls have cost Nvidia billions in potential revenue and argued repeatedly that they risk pushing Chinese customers toward domestic alternatives rather than actually preventing AI development. That is a pointed argument that the controls may accelerate indigenization rather than impede it.
The numbers have been complicated to interpret. Nvidia’s overall revenue surged dramatically through 2023 and 2024, driven by insatiable demand from American hyperscalers and Gulf state sovereign AI projects, meaning the China restrictions were masked by explosive growth elsewhere. But the company’s position in China’s AI market has been structurally damaged. Customers who once defaulted to Nvidia H100 clusters have been forced to evaluate alternatives, and some have committed to domestic suppliers for reasons that now go beyond mere patriotism.
For investors and analysts tracking the sector, a useful resource for understanding GPU architecture comparisons is a technical reference book on GPU computing. The underlying architecture decisions that make certain chips controllable and others not are genuinely technical matters that reward deeper study.
The Huawei Wildcard and the Indigenization Problem
The central counterargument to export controls has always been that China will simply build its own chips. Critics of the policy, including some former U.S. national security officials, argue that the controls provide a temporary delay measured in years, not decades, while simultaneously galvanizing Beijing’s investment in domestic alternatives.
The evidence for this concern is substantial. Huawei’s Ascend series of AI chips, developed through its HiSilicon semiconductor design arm, has become the most prominent domestic alternative to Nvidia’s products. When Huawei unveiled the Mate 60 Pro smartphone in 2023 containing a chip manufactured by SMIC at a 7-nanometer class process, it demonstrated that Chinese firms could achieve advanced fabrication under restrictions. The chip was not at the frontier of global manufacturing, but it was far ahead of what many analysts had predicted was possible under the export control regime.
Huawei’s Ascend 910B and subsequent iterations have reportedly been deployed in training clusters at Chinese internet companies. Performance benchmarks suggest these chips remain behind Nvidia’s best available hardware in raw throughput, but the gap is narrowing, and Chinese engineers have become adept at optimizing software to work around hardware limitations.
Beijing has responded to the restrictions with a multi-pronged industrial policy effort. The “Big Fund,” China’s state-backed semiconductor investment vehicle, has channeled enormous capital into domestic chip design and manufacturing. Estimates of total government and quasi-government investment in the semiconductor sector since 2020 run into the many billions of dollars, though precise figures are disputed. The result is a parallel technology ecosystem developing on an accelerated timeline, driven precisely by the pressure the controls were meant to apply.
Allies, Loopholes, and the Enforcement Gap
Export controls are only as strong as the coalition enforcing them, and building that coalition has been one of Washington’s most persistent diplomatic challenges. The United States persuaded the Netherlands and Japan to restrict exports of certain semiconductor manufacturing equipment, bringing ASML’s extreme ultraviolet lithography machines and Tokyo Electron’s equipment into the control regime. These were significant wins. ASML’s EUV machines are effectively irreplaceable for manufacturing chips at the leading edge, and the Dutch government’s decision to restrict their export to China represented a genuine constraint on SMIC’s ability to advance.
But the coalition has edges and gaps. Some allied nations have been slower to align their controls with U.S. standards, creating potential transshipment routes. Enforcement of entity list requirements across complex global supply chains is operationally difficult. Reports of chips appearing in restricted locations through intermediary countries have surfaced repeatedly, prompting the Commerce Department to add layers of verification requirements and expand investigations.
The so-called “chip smuggling” problem became a significant enforcement concern through 2024 and 2025. Investigations revealed instances of restricted Nvidia chips appearing in Chinese data centers through third-party distributors in Southeast Asia and the Middle East. The scale of these flows was disputed, but the existence of the problem was not. It underscored a core tension: the chips that are most valuable for AI training are also among the most portable, expensive, and in-demand products in the global technology economy, creating powerful incentives for evasion.
For policy researchers and students trying to follow the documentary record of these regulations, a policy analysis subscription service for technology governance can provide access to primary source materials and expert commentary that newspaper coverage often cannot.
The Global South Dilemma and Allied Resentment
One of the less-discussed dimensions of the export control debate concerns what might be called the collateral damage to countries that are neither adversaries nor formal allies. The short-lived diffusion rule framework introduced tiered country categories that would have treated much of the developing world as a high-risk zone for AI chip exports. Countries in Africa, Southeast Asia, Latin America, and the Middle East found themselves facing aggregate compute caps and licensing requirements that had nothing to do with their own military ambitions.
This created genuine diplomatic friction. India, which has complex relationships with both the United States and China, pushed back on classifications it found insulting to its sovereignty and its status as a strategic partner. Saudi Arabia and the UAE, which had become major customers for data center buildouts and sovereign AI projects, were initially placed in restrictive tiers that constrained their ambitions, prompting intensive negotiations that eventually resulted in more accommodating arrangements for trusted Gulf partners.
Critics argued that the blunt application of controls to the global south would simply push those countries toward Chinese technology providers, precisely the outcome the policy was meant to prevent. If a government in West Africa or Southeast Asia cannot easily procure American AI computing infrastructure, it may find Huawei’s offerings more accessible, even if less technically capable. The geopolitical logic cuts both ways.
The Trump administration, returning to office in January 2025, kept the core controls on advanced chips while scrapping the diffusion rule’s country tiers and emphasizing bilateral deals with Gulf partners. In December 2025 it also announced that Nvidia could sell its H200 chips to approved Chinese customers, with 25 percent of the revenue going to the U.S. government. Execution proved uneven: Chinese customs reportedly held up early shipments in 2026, and in May 2026 the U.S. cleared only a limited set of named Chinese buyers. The fundamental direction of policy remained consistent across administrations, which itself was notable: export controls on advanced chips had achieved something rare in American political life, a genuine bipartisan consensus rooted in both national security concerns and industrial policy ambitions.
What the Controls Have and Have Not Accomplished
Assessing the effectiveness of semiconductor export controls requires separating several distinct questions. Have they slowed China’s AI development? Have they strengthened American AI leadership? Have they damaged American chip companies competitively? And have they achieved their ultimate strategic goal of preventing a rival from developing AI capabilities that could threaten U.S. national security or military advantage?
On the first question, the honest answer in 2026 is: somewhat, for some time. Chinese AI labs, by most independent assessments, are working with hardware that is one to two generations behind what American frontier labs can deploy. Training the largest models takes longer, costs more, and requires more ingenuity in distributed computing architectures. That is a real constraint. But Chinese researchers have also demonstrated remarkable adaptability, and the models produced by firms like DeepSeek and others have in some cases achieved competitive performance relative to resource inputs, suggesting that hardware disadvantage is not a death sentence for AI competitiveness.
On American leadership, the controls have had mixed effects. U.S. hyperscalers have benefited from privileged access to the most advanced chips. But the controls have also fragmented the global technology market in ways that are beginning to create alternative ecosystems. A world with two partially separate AI supply chains is not obviously better for long-term American technological leadership than a world where U.S. companies dominated global markets.
For hardware professionals and engineers who need to stay current with the technical landscape that underlies these policy debates, a professional certification course in semiconductor technology can provide the foundational vocabulary that makes regulatory documents comprehensible.
The Road Ahead: Escalation, Adaptation, or Negotiation
The trajectory of chip export controls through the mid-2020s suggests a dynamic of continuous escalation and adaptation on both sides. Each tightening of American controls has been met with increased Chinese investment in domestic alternatives, and each advance in Chinese chip capabilities has been used to justify further controls. It is a regulatory arms race layered on top of a technological one.
Several scenarios compete for credibility as analysts look further ahead. In one, the controls succeed in maintaining a durable capability gap, buying time for American and allied AI development to cement advantages that become self-reinforcing. In another, the controls accelerate Chinese indigenization to the point where China achieves rough parity in chip manufacturing, rendering the controls irrelevant and having sacrificed significant revenue and diplomatic goodwill in the process. In a third, some form of negotiated arrangement emerges that trades certain access for verification mechanisms and behavioral commitments, though the political conditions for such a deal remain distant.
What seems clear is that the era of treating semiconductor supply chains as purely commercial matters is over. Chips, particularly the advanced AI accelerators that have become the defining infrastructure of the intelligence economy, are now explicitly classified as strategic assets by major governments. The implications of that classification will ripple through investment decisions, research collaborations, workforce mobility, and geopolitical alignments for years to come.
The October 2022 rules that started this particular chapter were drafted by U.S. officials who understood that they were making a consequential bet. Whether that bet pays off depends not just on the rules themselves but on whether the industrial, diplomatic, and technical ecosystem can be mobilized to enforce them, adapt them, and sustain the coalition they require. Silicon, it turns out, is not just a material. It is a policy instrument, and like all such instruments, its effectiveness depends entirely on the hands wielding it.
As an Amazon Associate, The Rough Idea earns from qualifying purchases.