Cheap Energy Might Not Be Enough: A Trade Model of AI Compute Services

Can energy-rich developing countries convert cheap electricity into AI compute exports? This paper develops a capacity-constrained trade model of AI compute services with bilateral frictions in delivery, regulation, and trust. Calibrating the model across 85 countries shows that several developing economies can produce compute at low cost, but this advantage rarely becomes export competitiveness. Because hardware dominates unit costs and is globally priced, cross-country production costs differ by only 12–20 percent. Modest regulatory, financing, and trust frictions can therefore erase the gains from cheap power. The binding constraint is institutional credibility rather than electricity prices, including enforceable data governance, stable regulation, credible power contracts, access to finance, and geopolitical alignment with buyers.