Rationed Intelligence

Inside every modern computer, smartphone, and data centre sits a component called a semiconductor chip. It is a wafer of silicon, roughly the size of a fingernail, engraved with billions of microscopic switches. These switches process information. The more powerful the chip, the more information it processes and the more sophisticated the tasks it can perform.

Artificial intelligence, in its current form, is essentially a mathematical operation run billions of times across billions of such switches. Without access to sufficiently powerful chips, a country cannot train competitive AI models. It cannot build the infrastructure to run them at scale. It cannot participate in the frontier of a technology that is reshaping medicine, agriculture, finance, and public administration.

Since October 2022, the United States has systematically restricted access to these chips through export control measures administered by the Bureau of Industry and Security (BIS), a body within the Department of Commerce. The measures began as a targeted response to China’s military AI ambitions. They have since expanded into a global tiered classification system that sorts every country on earth into one of three categories: trusted allies with unrestricted access, a middle group facing quotas and licensing requirements, and a restricted group facing near-total denial.

This piece is not about whether these export controls are good policy. They may serve legitimate security interests. The question is rather: what happens when a single state unilaterally constructs a global hierarchy over the foundational infrastructure of the twenty-first century economy, with no international legal rule governing how that power is exercised, no forum where affected states can challenge the classifications, and no obligation to justify the distinctions drawn?

The Architecture of Control and How It Is Enforced

Export controls are not new. States have long regulated the cross-border transfer of weapons, dual-use materials, and sensitive technologies under domestic law and through multilateral arrangements such as the Wassenaar Arrangement and the Nuclear Suppliers Group. What is new about the current American rules is their ambition and their genuinely global reach.

In January 2025, BIS issued what became known as the AI Diffusion Rule. The rule established a three-tier country classification for the export of advanced AI chips.

Tier 1 encompasses the United States and eighteen close allies, including the United Kingdom, Germany, Japan, South Korea, Australia and Canada. These entities access controlled chips without a licence. Tier 3 encompasses twenty-three arms-embargoed nations including China, Russia, Iran, and North Korea. Exports to these destinations are presumptively denied. Tier 2 covers every other country and includes most of the Global South.

How does the United States enforce this system globally? The answer lies in the structure of the global semiconductor supply chain. Every advanced chip produced anywhere in the world relies, at some stage, on American technology. The design software used to architect chips is predominantly American. The machines used to manufacture them depend on American components. Even ASML, the Dutch company that holds a monopoly on the extreme ultraviolet lithography machines essential to cutting-edge chip production, uses significant American-origin components in its equipment. This technological dependence in turn allows BIS to assert jurisdiction over foreign-made goods under its Export Administration Regulations pursuant to the Foreign Direct Product Rule (FDPR). If a chip is manufactured in Taiwan using American design tools and manufacturing equipment, the United States claims the right to stop its sale even if no American company is directly involved. A foreign company that violates the rule faces placement on the Entity List. Once listed, no American person or company can do business with it. Because almost every global technology company relies on American software, cloud services, customers, and access to the US banking system, being cut off from the American market is, for most firms, an existential threat. The practical result is that the key players in global semiconductor supply chains comply with American export rules not because they are bound by any domestic or international legal obligation to do so but because the cost of non-compliance is too high.

For Tier 2 countries, the rules established quantitative caps on chip imports. For 2025 through 2027, a Tier 2 country could receive approximately 50,000 H100-equivalent chips in aggregate before licence requirements were triggered. A Tier 2 government could double its cap by signing a bilateral agreement with the United States committing to align its own export controls with American security standards.

Recent Developments: The 2026 Rules and Congressional Pushback

The Trump administration formally rescinded the AI Diffusion Rule in May 2025, though the underlying policy objectives continued to govern licensing decisions.

On January 15, 2026, BIS issued a Final Rule that introduced a significant shift in policy toward China specifically. This rule replaced the previous presumption of denial for certain high-performance integrated circuits with a case-by-case review process. The chips covered include, but are not limited to, the Nvidia H200 and the AMD MI325X. Qualifying applicants must meet strict conditions, including volume caps relative to domestic sales, independent third-party chip testing, robust know-your-customer procedures, and a twenty-five percent tariff on chips fabricated abroad before re-export.

Congressional reaction was swift. On January 21, 2026, the House Foreign Affairs Committee passed the AI OVERWATCH Act by a vote of 42 to 2. The bill would impose a two-year ban on exports of next-generation Blackwell-class chips to China and would require congressional approval for individual AI chip export licences. The bipartisan margins signal that legislative oversight of executive export control decisions is becoming a live political issue, not merely an academic one.

These developments reveal that the rules are changing rapidly under domestic political pressure from multiple directions simultaneously. Firms in Tier 2 countries invest in AI infrastructure against a backdrop of legal uncertainty they cannot resolve through their own compliance efforts. The conditions governing their technological development are determined unilaterally in Washington by a process in which they have no formal voice.

The Limits of Existing International Law

States have a generally recognised right to control exports from their territory that derives from sovereignty: the authority of a state to regulate what leaves its jurisdiction. While not unlimited, this right’s limits under international law are modest.

WTO disciplines, primarily under the General Agreement on Tariffs and Trade (GATT), govern certain export restrictions on goods. These rules contain broad exceptions for measures taken in the interest of national security under Article XXI. The scope of that exception was partially addressed when the WTO Panel in Russia — Measures Concerning Traffic in Transit held that Article XXI is not entirely “self-judging” or “non-justiciable” in contradiction to the United States’ position. Even so, the ruling left substantial discretion to the invoking state and no WTO panel has yet applied GATT disciplines to AI semiconductor export controls.

The addition of new controls on emerging technologies including advanced semiconductors under the Wassenaar Arrangement that operates by consensus, has effectively been blocked by Russia since its invasion of Ukraine in 2022. The United States has accordingly moved ahead with unilateral controls and has sought to build what some analysts call plurilateral coalitions, sometimes described as a “Wassenaar-minus-one” approach, with key allies such as the Netherlands and Japan. The American chip control regime is, therefore, a unilateral exercise of sovereign regulatory authority, buttressed by the extraterritorial reach of the FDPR.

The practical result is that a Tier 2 country has no international legal forum to contest its classification, no right to explanation, and no avenue of appeal.

The Tier System as a Constructed Hierarchy

There is a longer, consequential history behind this arrangement. International law has repeatedly encountered situations in which powerful states construct sorting systems that divide the world into categories of trust and restriction without the consent of those being sorted. The Non-Proliferation Treaty is one example. It formalised a distinction between nuclear-armed states and all others. Developing countries contested this arrangement with some justification, understanding the Treaty as an instrument that locked in an existing power hierarchy under the appearance of a universally applicable legal regime.

The AI chip tier system does something structurally similar. The classification of most of the world’s population as Tier 2 was not negotiated but rather decided unilaterally. A comment period was formally open, but it was not a process through which foreign governments could meaningfully contest the basic premises of the classification. The result is a hierarchy of computational access that maps, with uncomfortable fidelity, onto existing hierarchies of geopolitical alignment with the United States.

This hierarchy, established by the computational sorting mechanism, undermines the principle of sovereign equality, recognised in Article 2(1) of the UN Charter, and the right to development articulated in the 1986 UN Declaration on the Right to Development and affirmed in subsequent instruments. That right can be derived not only from the Declaration itself but also from the foundational principle that all states are sovereign equals entitled to chart their own economic futures. If access to the computational infrastructure of modern economic development is rationed by the security preferences of a single state, without international legal constraint, the right to development becomes an aspiration that can be effectively cancelled by executive order.

The Case for Multilateral Discipline

While security concerns pose a legitimate interest of states, control over a globally consequential technology that shapes the economic trajectories of most of the world’s states should be subject to some form of multilateral legal discipline. At a minimum, that discipline would require a principled and publicly stated basis for access classifications, meaningful participation by affected states in setting those criteria, and some procedure for review.

However, multilateralism is under strain. The institutions that might host such a regime are weakened, as demonstrated by the collapse of the WTO Appellate Body. The political will among powerful states to subject their own technological advantages to binding international oversight is, at present, close to absent. The International Atomic Energy Agency (IAEA) provides a model that is instructive, not as a template but as a proof of concept. It shows that even technologies with genuine dual-use security implications can be placed under international institutional oversight rather than governed exclusively through the domestic law of the most powerful states. Still the IAEA embeds the very hierarchy it purports to manage: nuclear-armed states sit permanently on its Board of Governors – the most influential organ of the organisation – while non-nuclear states must submit to comprehensive safeguards inspections to verify that they are not developing nuclear weapons, without any equivalent verification obligation applying to the nuclear-armed states regarding their existing arsenals. But the IAEA also represents the basic proposition that international oversight is possible. For AI compute, no such institution yet exists. The question that follows is: what principles should govern such an arrangement, and what existing norms already limit how the current vacuum is managed?

The Contradiction Within Any Proposed Solution

This piece has argued that the tier system is objectionable partly because it employs a sorting mechanism constructed without the consent of those being sorted. Yet the minimum standards proposed in the existing literature — including transparency, access to review, and reciprocal obligations – would themselves constitute a new sorting system. They would simply be one based on publicly stated and principled criteria rather than opaque administrative decisions.

The question is whether principled criteria can avoid reproducing the power imbalances that the current system entrenches. The 1986 UN Declaration on the Right to Development was itself meant to express a principled criterion. It has not prevented the development of arrangements that effectively cancel the right it protects. A new sorting mechanism, however transparently constructed, will still be a sorting mechanism. States with greater technological capacity will still be in a stronger position to shape what counts as principled.

This tension does not dissolve the argument for reform. It complicates it. History offers instructive, if imperfect, examples of reconfigured sorting mechanisms that provided better balancing of interests than their predecessors. The transition from the Coordinating Committee for Multilateral Export Controls (CoCom) to the Wassenaar Arrangement in 1996 is one such example. CoCom operated as a Cold War bloc instrument with no participation by affected states. The Wassenaar Arrangement, despite its serious limitations, brought more states to the table, published its control lists openly, and created a forum for regular review. The shift did not eliminate hierarchy. But it distributed participation more broadly and made the basis for controls visible to those affected by them. The same logic applies here. The case for multilateral discipline over AI chip controls does not rest on the claim that any achievable arrangement will be perfectly just. It rests on the more modest claim that some form of shared governance, with affected states present at the table, is better than none.

Conclusion

International law has always struggled with the gap between formal equality among states and the material inequalities that determine what that equality is actually worth. The story of the twentieth century’s international legal order runs through decolonisation, the New International Economic Order debates, and the WTO’s promises about development. In significant part it is the story of that struggle. The AI chip export control regime represents its latest episode.

The question is not whether the current arrangement is sustainable. It is whether international law will develop a response before the hierarchy it reflects becomes too entrenched to revise.