Brazil’s Lula and Mexico’s AMLO are floating state-backed AI ventures modeled after Petrobras and Pemex. The pitch is democratic control of a transformative technology. The reality is a playbook from the 1970s: politically connected insiders will run the models, award the cloud contracts, and decide who gets compute access. Chile’s Codelco tried this with lithium AI and ended up with a $200M pilot that produced zero revenue and a revolving door of executives now lobbying for the very firms they once regulated.

AI concentration is a real problem but state capture is worse. Public ownership removes the one force that disciplines tech: the threat of creative destruction. When an AI model underperforms in the private sector, capital reallocates. When it underperforms in government, budgets expand and critics get labeled anti-development. The second order effect is a brain drain. Argentina’s INVAP lost 40 percent of its AI team to U.S. firms after the 2023 devaluation because state salaries could not match dollar-linked offers.

Winners: Local conglomerates like Slim’s América Móvil and Brazil’s Odebrecht who can leverage political ties to secure exclusive data deals. Losers: Regional startups like Argentina’s Mural or Colombia’s Frubana that need neutral infrastructure to compete with U.S. giants.

Watch for a 2025 pilot in Brazil where BNDES funds a national LLM trained on Caixa Economica’s financial data. If the model’s benchmarks lag Llama 3 by 15 percent but the contract rolls over anyway, the experiment is already corrupt.