US export controls on artificial intelligence technology cannot function effectively without cooperation from allied governments, according to an analysis published by the East Asia Forum. The finding frames the central policy problem now facing Washington: unilateral restrictions lose force when third countries keep their markets open.

The argument addresses a structural weakness in American trade-security policy. When one jurisdiction restricts the sale of advanced computing hardware, AI models or related tooling, suppliers and buyers reroute transactions through countries that maintain no comparable rules. The control regime then leaks at its borders.

Why do unilateral controls fall short?

Export controls regulate what leaves a country, not what happens after the goods arrive elsewhere. A chip or model exported to a jurisdiction without matching restrictions can be re-exported, resold or reverse-engineered with no legal exposure for the original seller beyond the first border crossing.

This creates three practical gaps:

• Transshipment: restricted goods travel through intermediary countries that impose no end-use checks.
• Cloud access: compute can be rented remotely, so physical export bans do not block model training or inference abroad.
• Substitute suppliers: manufacturers outside the controlling jurisdiction can supply functionally comparable components.

Each gap widens as the number of participating governments shrinks. The East Asia Forum analysis concludes that closing them requires aligned licensing rules, shared end-user verification and coordinated enforcement among allied states rather than US action alone.

What does the East Asia dimension change?

East Asia hosts critical nodes in the AI supply chain — semiconductor fabrication, advanced packaging and assembly — as well as major buyers of AI compute. Any export regime that excludes governments in this region leaves the largest leakage routes open.

The publication's title states its thesis directly: Washington needs allies to make AI export controls work. That framing positions allied diplomacy not as a complement to export policy but as a precondition for it.

The analysis does not claim current controls have failed outright. It argues they underperform relative to their stated goals because enforcement capacity sits partly outside US jurisdiction. A control regime built on one country's licensing decisions cannot police transactions that never touch that country's territory.

What would allied coordination require?

Effective multilateral export control rests on mechanisms with long precedent in earlier technology-control regimes. Applied to AI, the analysis points toward:

• common license exceptions and denial lists across partner states;
• reciprocal reporting of diverted shipments;
• agreed treatment of AI model weights and cloud compute as controlled items;
• joint monitoring of end users in third markets.

Without these, each allied government retains an incentive to free-ride — accepting the security benefits of US restrictions while capturing the sales Washington forgoes.

What is the market impact?

For semiconductor producers, AI hardware vendors and cloud providers, the policy question translates directly into revenue exposure. A control regime enforced by one country concentrates compliance costs on that country's firms and hands market share to competitors headquartered elsewhere. A regime enforced by many spreads the cost and narrows the arbitrage.

The East Asia Forum analysis implies the second outcome remains out of reach for now. Allies have not converged on a shared definition of which AI artifacts — chips, models, training services — fall under control, nor on how cloud-delivered compute should be treated.

What comes next?

The analysis leaves Washington with a diplomatic agenda rather than a regulatory one. Extending AI export controls from a unilateral instrument into a working multilateral regime requires negotiating alignment with governments that weigh commercial interests against security coordination.

The core conclusion stands on its own: the effectiveness of AI export controls is a function of alliance management, not of US rulemaking alone. Restrictions written in Washington hold only as far as allied enforcement reaches.