Regulatory Fragmentation and Economic Efficiency: An Empirical Analysis of United States State-Level Artificial Intelligence Policy Divergence and Federal Preemption Dynamics

As of September 2026, the United States regulates artificial intelligence (AI) not through a single federal statute but through a widening patchwork of state law, raising urgent and unanswered questions about the economic consequences of regulatory fragmentation. This dissertation provides the first integrated empirical assessment of that fragmentation, combining three methods: natural language processing (NLP) of a corpus of 1,766 state AI bills spanning 56 jurisdictions (of which 280 were enacted), a heterogeneity-robust Callaway–Sant'Anna staggered difference-in-differences (DiD) design linking legislative enactment to technology-sector labor outcomes, and a bottom-up compliance cost model that prices multi-jurisdictional regulatory divergence and evaluates alternative federal preemption architectures. Three findings anchor the analysis. First, the estimated aggregate annual compliance cost of the current fragmented regime is approximately $52.2 billion, distributed regressively so that startups bear roughly 8.3 times the revenue-proportional burden of large incumbents. Second, the causal DiD analysis finds no statistically significant effect of state AI legislation on technology-sector employment (ATT = 0.003) or wages (ATT = 0.008); an apparently significant negative wage effect in a naive specification does not survive corrected treatment timing or heterogeneity-robust estimation. Third, the political economy of the federal preemption standoff is consistent with Stigler–Peltzman capture dynamics: AI-specific federal lobbying grew from $5.2 million in 2018 to $55.0 million in 2025, a 40.1% compound annual growth rate, with large incumbents favoring ceiling preemption and startups, state attorneys general, and civil-society groups favoring floor preemption. A welfare analysis of counterfactual federal architectures identifies partial preemption—federalizing frontier-model safety while preserving state authority over employment and consumer protection—as the optimal policy, achieving a composite welfare score of 7.15 and roughly $13.9 billion in annual savings while preserving state experimentation. These results speak directly to the preemption debate before Congress, Treasury, Commerce, and the State Department, and to scholarship on regulatory federalism in the algorithmic age.