The United States continues to lead commercial artificial intelligence through companies such as OpenAI, Anthropic, Google DeepMind and xAI, but national security advisors say that advantage is increasingly fragile as China accelerates efforts to build sovereign compute infrastructure designed to operate independently of foreign technology. While U.S. firms have set global benchmarks for frontier models, Washington has yet to establish a cohesive strategy to ensure those systems can be trained, deployed and scaled domestically without relying on international supply chains, a gap officials warn could undermine leadership as AI becomes tightly coupled with defense, energy, and critical infrastructure.
China has pursued a different trajectory, allocating state-backed funding toward supercomputing centers, domestic semiconductor alternatives, and deployment pipelines insulated from U.S. cloud platforms. The approach reflects a shift in the global AI race from algorithmic breakthroughs to control over the physical infrastructure required to support them, where sovereignty, manufacturing, and jurisdiction are emerging as strategic variables equal to research talent. According to a former Commerce Department official who worked on export-control policy, the concern is not competition in innovation but vulnerability in deployment: “Open development helped startups and academics, but it doesn’t help you if essential systems need chips or servers you can’t legally access during a conflict.”
Even advocates of open-weight models acknowledge that transparency alone cannot function as a national strategy if critical training and inference depend on foreign compute. Analysts say that although open models have broadened participation, particularly among independent labs and smaller firms priced out of proprietary systems, they also reduce strategic leverage by enabling foreign states to replicate domestic breakthroughs at minimal cost. Some defense contractors have begun shifting development to secured private clusters that can operate offline, citing concerns that cloud-based AI could create single points of failure in military supply chains.
Policy advisors argue that long-term competitiveness will hinge on physical capacity rather than model quality: federally backed compute clusters on U.S. soil, scalable foundry capacity that reduces reliance on Taiwan-based semiconductor manufacturing, regulated national datasets suitable for training mission-critical systems, and deployment pipelines that can operate without commercial cloud dependence. Several proposals circulating in congressional working groups would allocate infrastructure funding through mechanisms similar to the CHIPS Act, but with operating requirements designed specifically for high-security AI workloads rather than consumer devices.
Despite growing concern, the U.S. retains structural advantages: concentration of top-tier research talent, access to global capital markets that support rapid scaling of private AI firms, and an ecosystem that continues to set performance benchmarks across frontier models. Investors say continued dominance will require aligning commercial incentives with national strategy rather than treating AI as a purely consumer technology sector. One Senate aide involved in drafting AI legislation said the central problem is misalignment of priorities: “Everyone is focused on who builds the biggest model. The real issue is who controls the server farms when those models run power grids and defense networks.”
Analysts warn that without a coordinated infrastructure plan, the U.S. risks replaying the pattern seen in industries such as solar energy, where American research led early breakthroughs but manufacturing capacity and long-term economic power consolidated overseas. The stakes, they say, are far higher now because AI underpins systems that governments cannot outsource without risking geopolitical leverage. As one national security advisor put it, “This isn’t about winning a tech trend. It’s about who can operate their digital economy when the world is not cooperating.”