The Geopolitics of AI Export Controls: Rethinking Digital Chokepoints

The United States government’s use of export controls has shifted from physical hardware to the direct regulation of artificial intelligence models. This change in strategy, highlighted by recent federal directives targeting advanced AI software, tests the limits of “weaponized interdependence”—the theory that states can leverage control over central network nodes to exert geopolitical influence. By moving beyond semiconductors to restrict access to the models themselves, Washington has inadvertently blurred the lines between national security, international commerce, and the reliability of American-led digital infrastructure.

The Evolution of Export Controls

For years, U.S. technology policy focused on the “small yard, high fence” approach, which sought to restrict adversarial access to critical hardware like advanced semiconductors and high-end cloud computing resources. However, the regulatory environment changed as artificial intelligence models became the primary drivers of technological capability.

According to research on global economic networks, states with jurisdiction over central nodes—such as cloud providers or AI platforms—can convert market centrality into strategic leverage. When the U.S. government restricts access to these nodes, it effectively creates a chokepoint. While these measures were traditionally designed to target specific foreign adversaries, the recent expansion of these controls to include AI models has introduced new complexities. By treating the software model itself as a controlled item, the U.S. has extended its regulatory reach to include foreign-national employees, allied organizations, and domestic firms, complicating the traditional “friend-enemy” distinction.

The Risks of Non-Discriminatory Restrictions

The primary challenge of modern AI export controls is the difficulty of maintaining precision. When a policy is applied via a broad nationality screen rather than a targeted list of entities, it creates significant friction for global enterprises.

* Allied Relations: Governments in the G7 and other allied nations have expressed concern regarding the discriminatory nature of these controls. Critics argue that such broad restrictions hinder collaboration and force partners to reconsider their reliance on U.S.-based technology.
* Commercial Disruption: For domestic companies like Anthropic, compliance with sudden, broad directives can lead to the total suspension of services, causing uncertainty for customers and internal staff alike.
* Administrative Discretion: The unpredictability of these controls leads users to question whether relying on a U.S. frontier model implies an ongoing dependence on American administrative decisions rather than purely technical stability.

This lack of discrimination threatens the very network centrality that gives the U.S. its leverage. If users perceive that access to a critical tool can be revoked overnight due to political shifts in Washington, the incentive to migrate to alternative, politically independent ecosystems increases.

The Rise of Open-Weight Alternatives

As Washington tightens its control over proprietary AI models, a “counter-chokepoint” has emerged through the diffusion of open-weight models. Rather than attempting to replicate the U.S. model of centralized control, developers and firms in other regions—most notably China—are increasingly relying on open-source ecosystems.

Data from the platform OpenRouter indicates a significant shift in developer behavior. By mid-2026, Chinese-developed open-weight models accounted for a majority of routed tokens on the platform, signaling that developers are increasingly prioritizing accessibility and political predictability over the frontier performance of closed models. Reports from the U.S.-China Economic and Security Review Commission highlight that Chinese labs are actively publishing source code and weights, creating a feedback loop of adoption and iteration that reduces the effectiveness of U.S. denial-of-access strategies.

Systemic Implications for Global Technology

The Anthropic case serves as a limit case for weaponized interdependence. While the U.S. retains the power to deny access to its central technological nodes, the exercise of that power now carries higher costs.

When a chokepoint is used too broadly, it ceases to be a surgical instrument and becomes a systemic hazard. It teaches global users that dependence on a centralized hub is a vulnerability. Consequently, the strategic irony of current U.S. policy is that by aggressively fencing off advanced models through broad administrative discretion, the government may be accelerating the global adoption of alternatives. Moving forward, the effectiveness of U.S. technology policy will likely depend on whether it can balance national security requirements without eroding the trust and predictability that sustain the global appeal of American digital infrastructure.