The Sovereign AI Capability Imperative Why American Dependence on Commercial Frontier AI Constitutes a Structural National Security Vulnerability

Abstract The United States leads the world in artificial intelligence, but the leadership rests on four privately-held companies whose foundation models the federal government cannot replicate, cannot fully control, and cannot afford to lose access to without operational degradation. This paper argues that the resulting dependency constitutes a structural national security vulnerability comparable in character to Cold War-era reliance on a small number of strategic suppliers, and that the United States requires a Manhattan-style sovereign artificial intelligence capability — government-directed, government-funded, with government-controlled deployment authority — to address it. The paper situates the argument against the People's Republic of China's military-civil fusion doctrine, in which all elements of national capacity are explicitly postured to support military objectives. It steel-mans the principal counterarguments — cost, talent acquisition, government efficiency, and the risk of suppressing commercial innovation — and answers them. It concludes with three concrete policy recommendations: a Federally Funded Research and Development Center adapted for frontier AI, an Intelligence Community-resident capability for the most sensitive workloads, and a public-private partnership architecture that combines government direction with commercial execution under sovereign deployment terms. The recommendations are scoped for implementation within a single administration cycle.