How to Outperform China with a Glass-Steagall Act for Defense

The US defense acquisition system exhibits a persistent and widening gap between technological innovation and fielded military capability—a gap China is rapidly closing. This paper argues that the root cause is not insufficient R&D funding or bureaucratic inefficiency *per se*, but a structural monopoly: the vertical integration of design, production, and legal leverage within a small number of prime contractors. Drawing on historical parallels from World War II (UK, Nazi Germany, and the Soviet OKB system) and contemporary commercial analogs (TSMC/fabless semiconductors, SpaceX/Tesla), we propose a three-pillar reform architecture. The centerpiece is a 'Glass-Steagall Act for Defense': the permanent structural separation of design/innovation entities from production/manufacturing monopolies. We estimate that such a model could compress the innovation-to-conversion timeline from 10–15 years to 2–3 years—comparable to commercial-sector scaling velocities. Beyond the defense reform proposal, this paper makes a broader contribution to institutional economics. The mathematical model developed in Appendix B provides the first formal stability analysis of the defense acquisition pipeline, demonstrating that the system exhibits a phase transition governed by a structural separation parameter. This formalization addresses persistent critiques of the Acemoglu-Robinson institutional framework by operationalizing 'elites,' providing a testable theory of institutional change, identifying a critical threshold (σ^* ≈ 0.55) that separates slow and fast conversion regimes, and proving that the current US architecture (σ ≈ 0.15) is structurally locked into a slow-conversion regime from which incremental reform cannot escape. The paper thus transforms a qualitative diagnosis into a mathematically supported prescription. We do not address political feasibility; we present a technically coherent model and pose the question of implementation as a political-economy problem for further research Erratum v0.0.1: publication date 26 August 2026