Poster #120 - Which States Win Federal AI Contracts? Procurement Structure, Politics, and Geography

• In Event: Saturday Poster Luncheon

Abstract

Virginia received $532.9 million in federal AI contract obligations between fiscal years 2020 and 2024 — 31.8 percent of a $1.68 billion market. California, Maryland, Massachusetts, and Texas combined for another 39.6 percent. The remaining 45 states split less than 29 percent. That concentration is the starting point for this paper.Three bodies of scholarship predict different causes. Distributive politics research expects partisan alignment between the governor and the White House to steer discretionary spending toward co-partisan states. Agglomeration economics expects states with established technology and defense contracting ecosystems to keep winning — not because of politics, but because cleared workforces, agency relationships, and past performance records compound with each contract cycle. Procurement structure theory makes a different claim: that indefinite-delivery vehicles and pre-approved vendor pools create incumbency advantages that persist across administrations regardless of partisan control or geography. All three have been tested in other spending contexts. None has been tested against state-level AI contract data.We use a balanced panel of 50 states from fiscal years 2020 to 2024 (250 state-year observations). Contract data comes from USAspending.gov prime award records, filtered for AI-related keywords and aggregated to the state-year level, then merged with BEA GDP estimates, Census population figures, and NGA and Ballotpedia party records.

We exclude the District of Columbia because it lacks comparable state governance institutions. About 27 percent of state-year observations carry zero AI obligations — a within-state variation problem, not a permanent exclusion problem, since every state received at least some AI contracting dollars across the five-year window. We run OLS with logged AI obligations as the outcome variable and report variance inflation factors given the correlation between GDP, population, and federal contracting footprint. All specifications include year fixed effects; we test robustness with state fixed effects given the within-state variation in zero-obligation years.

Early descriptive patterns point toward agglomeration and incumbency rather than partisan alignment. The correlation between a state's non-AI federal contracting footprint and its AI contract capture is 0.58. States aligned with the White House average $6.2 million in annual AI obligations; non-aligned states average $7.1 million. That gap moves in the wrong direction for the partisan prediction. Virginia's share appears tied to decades of defense contracting infrastructure that newer entrants cannot replicate on a short timeline. That advantage does not reset when administrations change.This does not yet settle whether procurement structure or agglomeration effects are the primary driver — distinguishing them is what the regression analysis does. But the descriptive patterns are already hard to reconcile with a partisan account. Changing administrations without changing contracting structure is unlikely to change the map.

Author

• Javaid Sofi, Virginia Polytechnic Institute & State University
  Presenting Author