How Washington is drafting the most sweeping technology export controls since the Cold War — and why every nation building AI infrastructure should pay attention
Executive Summary
- The Trump administration is drafting regulations that would require every country in the world to obtain U.S. government approval before purchasing Nvidia or AMD AI chips — transforming existing China-focused export controls into a universal licensing system.
- The proposed tiered system — from cursory reviews for small orders to intergovernmental treaties for mega-clusters — would give Washington unprecedented power to determine who can build AI infrastructure and where.
- This represents a fundamental shift from the Cold War's COCOM regime and Biden's AI Diffusion Rule: instead of blocking adversaries, the U.S. is asserting control over allies — effectively creating an "OPEC of compute" where Washington sets the production quota for global intelligence.
Chapter 1: The Draft Rule That Changes Everything
On March 5, 2026, Bloomberg reported that U.S. officials have written draft regulations that would restrict AI chip shipments to anywhere in the world without American approval. The rule, still in draft form within the Commerce Department's Bureau of Industry and Security (BIS), goes far beyond existing export controls that target specific adversary nations.
Under current regulations, companies like Nvidia and AMD cannot ship high-performance AI processors to China, Russia, Iran, and North Korea. Sales to Middle Eastern nations require intergovernmental arrangements and matching investments in U.S. AI infrastructure. But most of the world — Europe, Japan, South Korea, India, Southeast Asia — has been able to purchase advanced AI chips with minimal friction.
The proposed rule would end that era.
The Three-Tier Architecture
The draft establishes a graduated licensing system based on deployment scale:
| Tier | Scale | Process | Requirements |
|---|---|---|---|
| Light | ≤1,000 GB300 GPUs | Simplified review | Minimal; possible limited exemptions |
| Medium | 1,001–199,999 GPUs | Pre-authorization required | Compliance obligations, operational transparency, potential on-site U.S. inspections |
| Heavy | ≥200,000 GPUs (single company/country) | Direct intergovernmental arrangement | National security assurances + mandatory investment in U.S. AI infrastructure |
The implications are staggering. A single hyperscaler data center — the kind currently operated by AWS, Microsoft, Google, Oracle, OpenAI, or xAI — easily exceeds 200,000 GPUs. Under the proposed rules, building such facilities outside the United States would require what amounts to a bilateral treaty between Washington and the host government.
Chapter 2: From COCOM to Compute Control — A Historical Lens
The United States has a long history of using technology export controls as instruments of geopolitical power. But the proposed AI chip licensing regime represents something qualitatively different from its predecessors.
The COCOM Era (1949–1994)
The Coordinating Committee for Multilateral Export Controls was a Western bloc agreement to restrict technology flows to communist states. It was multilateral, consensus-based, and focused on a clearly defined adversary. When Toshiba sold submarine-quieting milling machines to the Soviet Union in the 1987 Toshiba-Kongsberg scandal, the backlash was fierce — but the system was designed to police exceptions, not to control all trade.
The Entity List Era (2019–Present)
The Trump and Biden administrations expanded export controls dramatically, placing Huawei, SMIC, and hundreds of Chinese entities on the Entity List. But these were still targeted restrictions aimed at specific companies and countries.
The AI Diffusion Rule (2025)
Biden's AI Diffusion Rule, finalized in January 2025, created a three-tier country classification system. Tier 1 allies could purchase chips with few restrictions; Tier 2 nations faced caps; Tier 3 (adversaries) were blocked. The Trump administration initially criticized this as too complicated — then suspended it.
The Proposed Universal Licensing (2026)
The new draft rule is qualitatively different. It doesn't classify countries; it classifies transaction sizes. Every shipment, to every destination, requires some form of U.S. government review. This is not an export control — it is an export permit system. The distinction matters: controls define what is prohibited; permits define what is allowed. Under a permit system, the default is denial unless approval is granted.
This represents the most expansive assertion of extraterritorial technology jurisdiction since the Manhattan Project's classification regime.
Chapter 3: The Strategic Logic — Why Now?
Three converging forces explain why Washington is moving toward universal compute licensing in March 2026.
1. The China Loophole Problem
Despite years of escalating export controls, advanced AI chips continue to reach China through intermediary countries. Singapore, Malaysia, UAE, and Central Asian nations have all been identified as transshipment points. Nvidia's own CFO confirmed zero China H200 revenue, but the company halted production precisely because it couldn't verify end-use destinations. A universal licensing system closes the loophole by requiring approval for every transaction, making it impossible to ship chips to Singapore "for domestic use" that end up in Shenzhen.
2. The Pax Silica Architecture
The Compute OPEC draft aligns with the broader Pax Silica technology alliance framework unveiled at the Delhi AI Summit. By conditioning chip access on security commitments and investment in U.S. AI infrastructure, Washington is building a loyalty-based compute distribution system. Nations that invest in American AI get preferential access; those that don't face bureaucratic friction that could delay projects by months or years.
3. The Iran War Catalyst
The ongoing conflict has dramatically underscored the strategic importance of AI-enabled military capabilities. Iran's use of AI-guided drones and missiles, and the U.S.-Israeli reliance on AI-powered defense systems, has made AI compute a national security commodity of the first order. The Pentagon's recent confrontation with Anthropic over military AI use — and its embrace of OpenAI, xAI, and Palantir — reinforces the view that AI compute is now a munition.
Chapter 4: The Stakeholder Map
Nvidia and AMD: Revenue at Risk
Nvidia generated $216 billion in revenue last year, with the vast majority tied to AI. AMD's AI-related revenue reached $35 billion. Under the proposed rules, every international sale faces additional compliance costs, delays, and uncertainty. The company's experience in China — where $17 billion in annual sales evaporated after the 2025 restrictions — serves as a cautionary precedent.
But the industry response is complicated. While Nvidia publicly opposes export controls, a universal licensing system actually strengthens its competitive position by raising barriers to entry for foreign competitors. If every Nvidia chip sale requires a U.S. permit, then customers have no incentive to switch to non-American alternatives — because those alternatives don't come with the U.S. security umbrella.
European Allies: Sovereignty vs. Access
The timing couldn't be worse for Europe. The EU is simultaneously trying to build sovereign AI capabilities through the European AI Act, the SAFE defense bond, and massive rearmament spending. Requiring U.S. government approval for European data center construction directly conflicts with European strategic autonomy objectives.
France's Macron, who just announced a nuclear deterrent expansion, is unlikely to accept U.S. veto power over French AI infrastructure. Germany's Merz, fresh from shuttle diplomacy between Beijing and Washington, faces the question of whether "derisking, not decoupling" extends to accepting American compute sovereignty.
China: Acceleration, Not Defeat
Paradoxically, universal export controls may accelerate China's semiconductor self-sufficiency drive. The 15th Five-Year Plan, unveiled at the Two Sessions this week, targets 70% semiconductor equipment self-sufficiency by 2030. If American chips become equally difficult for allies and adversaries to obtain, the incentive for other countries to develop or purchase non-American alternatives increases.
Huawei's Ascend 910C, SMIC's improving yields, and DeepSeek's optimization for domestic hardware all benefit from a world where American chips come with political strings attached.
The Global South: Digital Colonialism Intensified
For developing nations trying to build AI capabilities — India's $120 billion MANAV vision, Indonesia's digital economy ambitions, Nigeria's fintech ecosystem — the proposed rules mean that every AI infrastructure investment must first pass through Washington's approval process. The power asymmetry is stark: the U.S. controls the technology, sets the conditions, and can revoke access at will.
Chapter 5: Scenario Analysis
Scenario A: Enacted with Full Force (25%)
Premise: The draft rule is finalized largely as written, with the three-tier system intact.
Rationale: The Iran war has created a national security environment where expansive executive power faces minimal pushback. The SCOTUS IEEPA ruling actually strengthens the case for using existing statutory authorities (Export Control Reform Act of 2018) rather than emergency powers.
Trigger conditions: Continued Middle East escalation; bipartisan congressional support for tech controls; no major allied revolt.
Historical precedent: The Manhattan Project's "born classified" concept — where nuclear technology was controlled from inception, not after proliferation.
Investment implications: Nvidia and AMD face short-term compliance costs but long-term oligopoly protection. Non-U.S. chip designers (Arm, TSMC custom designs) could see demand surges. Compliance-tech firms benefit.
Scenario B: Diluted Compromise (45%)
Premise: After allied pushback, the rule is softened. The light tier is expanded, the heavy tier is limited to non-allied nations, and allied nations receive blanket pre-authorizations.
Rationale: Europe, Japan, South Korea, and India — all critical security partners — would resist the full version. The Trump administration, needing allied cooperation on Iran and trade, accepts a compromise that maintains control over adversaries and swing states while giving allies a streamlined path.
Trigger conditions: EU threatens retaliatory measures (digital services taxes, Arm access restrictions); Japan and South Korea push back through defense cooperation leverage; Nvidia lobbying succeeds.
Historical precedent: The Wassenaar Arrangement (1996), which replaced COCOM with a looser multilateral framework after allied resistance to overly restrictive controls.
Investment implications: Limited disruption for major chip companies; compliance costs manageable; China remains the primary target.
Scenario C: Shelved or Abandoned (30%)
Premise: The rule is never finalized, killed by a combination of industry lobbying, allied resistance, and shifting administration priorities.
Rationale: The Iran war consumes policy bandwidth; the April Trump-Xi summit in Beijing requires trade concessions; the 150-day Section 122 tariff clock creates competing priorities; midterm election calculations favor deregulation narrative.
Trigger conditions: Ceasefire in Iran; successful Beijing summit; Nvidia lobbies effectively; congressional opposition from tech-friendly Republicans.
Historical precedent: Biden's AI Diffusion Rule was finalized but then suspended by the Trump administration, suggesting a pattern of ambitious drafting followed by practical retreat.
Investment implications: Return to status quo; Nvidia/AMD maintain current market dynamics.
Chapter 6: Investment Implications
Winners
| Asset | Rationale |
|---|---|
| Compliance/RegTech | Global export licensing creates massive demand for automated compliance systems |
| Non-U.S. chip designers | RISC-V, Arm custom designs, European alternatives gain strategic value |
| U.S. data center REITs | "Heavy tier" rules make U.S.-based compute most frictionless option |
| Nvidia/AMD (long-term) | Paradoxically, universal licensing cements oligopoly — no competitor can match U.S. security clearance |
Losers
| Asset | Rationale |
|---|---|
| Non-U.S. hyperscalers | European, Asian cloud builders face delays and uncertainty |
| Emerging market digital economy | India, Indonesia, Brazil AI ambitions face bureaucratic headwinds |
| European sovereign AI plays | Mistral, Aleph Alpha face compute access risk if French/German governments resist U.S. terms |
Key Metric to Watch
The Commerce Department's average license processing time. Under current rules, export licenses take 30-90 days. If the new system maintains that timeline for the "medium" tier, the practical impact on allied nations would be severe — a 90-day delay in chip deliveries could derail billion-dollar data center projects.
Conclusion
The proposed global AI chip licensing regime represents a watershed moment in technology geopolitics. For the first time since the dawn of the semiconductor industry, Washington is asserting that no country — ally or adversary — has an inherent right to purchase American AI processors without government approval.
This is not merely an export control. It is the foundational architecture of a new global order where access to intelligence — artificial intelligence — is determined by political alignment with the United States. The question is whether America's allies will accept this bargain, or whether the Compute OPEC will fracture the very alliances it claims to protect.
The answer will shape not just the semiconductor industry, but the distribution of power in the 21st century.
Sources: Bloomberg, Tom's Hardware, Motley Fool, Seeking Alpha, Xinhua, The Guardian
Leave a Reply