The Moment a Chip Becomes a Weapon
Imagine a cargo container full of graphics processing units sitting at a port. To a casual observer, the crates hold consumer electronics, the kind of hardware that renders video games or accelerates scientific simulations. To a national security analyst, that same container might represent a meaningful upgrade to a foreign military’s ability to train artificial intelligence models for autonomous weapons, cryptanalysis, or large-scale surveillance. The difference between those two readings is precisely why export controls on semiconductors have become one of the most fiercely contested arenas in global commerce.
Since the Biden administration issued a sweeping set of chip export restrictions in October 2022, the United States government has attempted something genuinely ambitious: using its leverage over the global semiconductor supply chain to slow the development of advanced AI capabilities in countries it considers adversaries, primarily China. The policy has since evolved through multiple rounds of revisions, drawing objections from chipmakers, trading partners, and civil liberties advocates alike. Understanding how the rules actually work requires looking at the machinery behind them, the technical thresholds that define a controlled chip, the agencies that enforce the rules, and the cat-and-mouse dynamic between regulators and the companies that build the hardware.
The Regulatory Architecture: Who Makes the Rules and How
Export controls in the United States are administered primarily by the Bureau of Industry and Security, an agency within the Department of Commerce. The bureau operates under the authority of a law commonly known as the Export Control Reform Act of 2018, which modernized a framework that had existed in various forms since the Cold War era. The core instrument the bureau uses is the Export Administration Regulations, a dense body of rules that classifies goods, software, and technology according to their potential military or strategic significance.
Every product subject to these regulations receives an Export Control Classification Number. Items with certain classification numbers require a license before they can be shipped to particular countries or end users. License applications are reviewed against a set of policies that weigh the technical capabilities of the item, the identity of the buyer, the stated end use, and the broader foreign policy context. For most commercial goods shipped to most countries, no license is required. For advanced semiconductors destined for China, the calculus has become dramatically more complicated.
The key concept linking chips to national security is what regulators call “dual-use” technology: hardware or software with legitimate civilian applications that also has significant military potential. A high-performance GPU originally designed to train a commercial recommendation algorithm can, with the right software, train a model that guides a missile or processes battlefield imagery. The dual-use nature of AI accelerators is precisely what made the 2022 restrictions so sweeping and so controversial.
Performance Thresholds and the Problem of Drawing Lines
The technical heart of the semiconductor export control regime is a set of numerical performance thresholds. Rather than naming specific products directly, the rules originally defined controlled chips by metrics such as aggregate performance measured in a unit called the Total Processing Performance and a related measure of interconnect bandwidth. Any chip exceeding both thresholds simultaneously triggered the license requirement for exports to China.
This approach has significant appeal. By writing rules around performance specifications rather than product names, regulators hoped to create a framework that would remain relevant even as specific chip generations came and went. The problem is that semiconductor engineers are extraordinarily good at their jobs. After the October 2022 rules were published, Nvidia moved with notable speed to redesign its data-center chips in ways that brought them just below the threshold values, producing variants such as the A800 and H800 specifically tailored for the Chinese market.
The Commerce Department responded to this workaround by issuing updated restrictions in October 2023 that tightened the thresholds and added a “performance density” metric intended to close the redesign loophole. The cycle illustrated a structural tension in the entire project: the regulators are writing rules about hardware whose capabilities are advancing faster than the typical pace of government rulemaking.
Nvidia’s situation has drawn particular attention because the company has, at various points, been generating a substantial share of its data-center revenue from Chinese customers. The company has stated publicly that complying with U.S. law is its obligation and that it designs its products accordingly. Critics of the restrictions argue that the effect is simply to push Chinese buyers toward domestic alternatives, strengthening companies such as Huawei’s semiconductor division, while critics of Nvidia’s workarounds argue that redesigned chips still provide meaningful capability to entities the controls are meant to restrict.
The Entity List, the End-User Problem, and Enforcement
Performance thresholds address what can be exported. A parallel instrument addresses who can receive controlled technology. The Bureau of Industry and Security maintains what is called the Entity List, a roster of foreign companies, research institutions, government bodies, and individuals to whom exports of specified items are presumed to require a license that will generally be denied. Placement on the list is an administrative determination, not a criminal conviction, but its commercial consequences can be severe.
Huawei was added to the Entity List in 2019, a decision that cascaded through the global supply chain because the company was not only a chip buyer but a buyer of chip-making equipment, software, and intellectual property. The addition effectively cut Huawei off from leading-edge chip fabrication at foundries such as TSMC, which manufactures chips designed by companies all over the world using U.S.-origin equipment and technology.
This points to one of the most powerful and least understood levers in the entire framework: the Foreign Direct Product Rule. Under this rule, the U.S. government can assert jurisdiction over products made entirely outside the United States if those products are manufactured using American equipment, software, or technology above a specified threshold. Because nearly every advanced semiconductor fabrication process in the world relies heavily on U.S.-origin tools, particularly lithography and design software, the Foreign Direct Product Rule gives Washington substantial reach beyond its own borders. When the rule was expanded in 2022 to cover advanced computing chips and semiconductor manufacturing equipment more broadly, it effectively enrolled foreign chipmakers, including TSMC, in the U.S. export control system whether or not they had any particular interest in geopolitics.
Enforcement is handled partly by the bureau’s own Office of Export Enforcement and partly by other federal agencies including the Department of Justice and Customs and Border Protection. Violations can result in civil penalties, criminal prosecution, and denial of export privileges. The harder enforcement problem is diversion: controlled chips being purchased through third-country intermediaries in places like Singapore, Malaysia, or the United Arab Emirates and then routed to restricted end users. The Commerce Department has signaled repeatedly that it is aware of diversion concerns and has added various intermediaries to the Entity List as a result, but policing a global distribution network for small, high-value objects is genuinely difficult.
The Allied Dimension and the Multilateral Gap
One structural weakness in the American approach is that chips are made and sold by companies in many countries, not just the United States. If the Netherlands, Japan, South Korea, and Taiwan do not apply equivalent restrictions, the effect of U.S. rules is blunted. This created intensive diplomatic pressure on allied governments to harmonize their own controls.
The Netherlands is home to ASML, the company that manufactures extreme ultraviolet lithography machines, tools so technically singular that they are essential to producing the most advanced chips in the world. Under pressure from Washington, the Dutch government in 2023 imposed its own export licensing requirements on ASML’s advanced deep ultraviolet immersion systems, adding to the restrictions on its EUV machines that had been in place since 2019. Japan similarly announced controls on certain semiconductor manufacturing equipment in 2023. These moves represented meaningful steps toward a multilateral regime, though the degree of genuine alignment among allies on exact thresholds and enforcement has remained a subject of ongoing negotiation.
The coordination challenge illustrates a broader tension. American semiconductor companies argue that unilateral restrictions hurt their revenues and market position without necessarily preventing adversaries from accessing alternative technologies. Industry groups have argued that U.S. chip firms risk losing market share permanently if the rules are more stringent than those applied to competitors in allied countries. On the other side, security-focused analysts argue that the revenue concern, while real, is secondary to the strategic imperative of denying adversaries the hardware needed to build frontier AI systems.
What the Controls Can and Cannot Accomplish
The strategic logic of semiconductor export controls rests on a specific theory of leverage. Advanced chip fabrication is concentrated in a small number of locations, depends on an intricate global supply chain, and cannot be replicated quickly. The United States and its allies, through a combination of domestic design leadership, control over critical manufacturing equipment, and the reach of the Foreign Direct Product Rule, occupy chokepoints in that chain. If those chokepoints can be held, the argument goes, the development of AI capabilities with potential military applications can be meaningfully slowed.
The counterargument takes several forms. First, China has made explicit and well-funded efforts to develop domestic alternatives in chip design and manufacturing, with government programs directing substantial resources toward that goal. Progress has been uneven and the gap between China’s domestic chip capabilities and the global frontier has remained significant in recent years, but the trajectory matters as much as the current position. Second, software efficiency improvements mean that some AI tasks that once required cutting-edge hardware can increasingly be accomplished on somewhat less powerful chips. Third, the controls create friction that is not evenly distributed: a well-resourced state actor has more options for navigating restrictions than a smaller country or non-state group might.
There is also the question of what “slowing down” actually means in practice and whether slowing down is the right frame at all. Some analysts argue the more achievable goal is ensuring that the United States and its allies maintain a sufficient lead in the most capable AI systems, particularly those with direct military applications, rather than hoping to prevent adversaries from developing AI altogether. Others argue that the restrictions, by accelerating Chinese investment in domestic semiconductors, could ultimately reduce American leverage over a longer time horizon.
The Road Ahead: Rules in Motion
Export controls on semiconductors are not a fixed policy but a living regulatory system in continuous revision. Each round of updated rules reflects new assessments of technical capabilities, new evidence about how previous rules are being circumvented, and new diplomatic negotiations with allies. The October 2022 package and its October 2023 revisions were followed by further updates, and the framework has continued to evolve.
The fundamental difficulty is that the technology is moving faster than the rulemaking. AI chip capabilities have advanced sharply in recent years, and the gap between what a consumer-grade GPU can accomplish and what a data-center cluster can accomplish has grown rather than shrunk. Meanwhile, the line between civilian and military AI applications has become blurrier, not clearer, making dual-use determinations harder to defend with precision.
For the companies involved, the uncertainty is as costly as the restrictions themselves. Chip designers must build product roadmaps without knowing exactly what regulatory environment they will face at launch. Customers in countries not subject to controls worry about being caught in regulatory crossfire. And governments in allied nations face pressure from both Washington and their own technology industries.
The semiconductor export control system is, at its core, an attempt to translate physical chokepoints in a global supply chain into lasting strategic leverage. Whether that leverage can be sustained as China’s domestic capabilities develop, as allied coordination faces commercial and political pressures, and as the technology itself keeps advancing is a question that engineers, diplomats, and national security officials are all trying to answer at the same time. The chessboard is large, the pieces keep changing shape, and no one is entirely sure of the rules.