AI Boom Is Building a New Global Divide and Poorer Economies Could Pay the Price

The AI boom could leave poorer economies further behind even as the technology becomes more capable. Geopolitical barriers may prevent investment from reaching countries that need it to modernise production and raise productivity. For economies already constrained by expensive credit and limited technological capacity, restricted access to AI financing could turn an opportunity for economic catch-up into another source of disadvantage.

Researchers Fernando Toledo and Gabriel Montes-Rojas examine this risk in Global Reallocation of Capital in the Era of Geo-Economic Fragmentation and Artificial Intelligence. Their May 2026 paper, available as an arXiv preprint, investigates how political divisions could reshape international investment and distribute the resulting economic costs unevenly across countries and households.

The researchers connect international finance, AI investment and household inequality within an economic model informed by earlier research and economic data. Their results describe conditional scenarios rather than observed losses or forecasts for individual countries. The central proposition is consequential: political barriers to investment could compound existing technological disadvantages, leaving economies and households with fewer resources bearing disproportionate costs.

When Investment Barriers Become Barriers to Growth

The paper treats AI as a technology requiring specialised capital, including computing hardware, data-centre infrastructure and model-related assets. In its framework, geopolitical distance raises the cost of cross-border investment. Financing becomes less available, imported AI capital becomes more expensive through exchange-rate effects, and firms facing borrowing limits struggle to maintain investment and labour demand.

The authors report welfare losses equivalent to 0.42–1.02% of lifetime consumption under moderate fragmentation, rising to 2.03–5.20% under extreme fragmentation. The range depends on how readily AI substitutes for labour. These figures translate changes in economic welfare into consumption equivalents; they should not be read as annual GDP losses, immediate household spending cuts or predictions tied to a particular year.

Within the model, costs accelerate because several pressures reinforce one another. Reduced investment weakens capital accumulation, while financial constraints make subsequent adjustment more difficult. When AI is assumed to substitute more readily for workers, the simulated losses are larger. The paper does not establish which relationship between AI and labour will dominate across real economies, making that assumption a major source of uncertainty.

The study also examines what happens when AI technology improves. Under lower fragmentation, capital inflows peak about 0.85% above their steady-state level; under higher fragmentation, the response reaches only 0.51%. The roughly 40% reduction illustrates its proposed mechanism: stronger technological opportunities can generate a weaker investment response when geopolitical barriers obstruct the finance needed to exploit them.

Countries Behind the Frontier Face a Double Exposure

The model's distributional results suggest that fragmentation could reinforce advantages already held by economies with greater AI capacity and easier financing. Under moderate fragmentation, in a scenario where AI substitutes for labour, the authors report a welfare loss of 1.35% for the economy behind the AI frontier, compared with 0.21% for its more technologically advanced counterpart.

The difference reflects assumptions about both AI resources and financial capacity. The economy behind the frontier faces tighter borrowing constraints, which amplify the investment shock. For Global South policymakers, the implication is that efforts to strengthen domestic AI capabilities may deliver less than expected if access to international capital deteriorates. Technological readiness and financial resilience, hence, need to be considered together.

Household outcomes add another dimension. Within the model's advanced economy, the lowest wealth fifth experiences a reported welfare loss of 1.25%, compared with 0.50% for the highest wealth fifth, a 2.5-fold difference. The authors attribute much of the disparity to labour-income exposure and limited financial buffers. Families relying heavily on earnings have fewer ways to absorb weaker wages and greater uncertainty.

These stylised results need testing against the diversity of actual countries and households. Nevertheless, they identify a development concern: international investment restrictions could reinforce inequalities both across borders and within national economies. Assessing AI policy solely through investment totals would then overlook how financing conditions influence employment, income security and the distribution of technological gains.

Connector Economies Gain While Access to Both Blocs Survives

Fragmentation does create potential beneficiaries in the authors' framework. A three-bloc extension introduces a connector economy capable of maintaining closer relationships with competing groups. Using Mexico-related geopolitical distances for calibration, the model examines how such an intermediary could attract investment diverted away from direct links between rival blocs. Its results are illustrative, rather than a forecast for Mexico.

The simulated connector gains are equivalent to 0.51% of lifetime consumption under moderate fragmentation and 0.72% at an intermediate level. Under extreme fragmentation, however, the gain falls to 0.08%. The opportunity depends on continued access to both sides; deeper separation eventually raises barriers for the intermediary itself and erodes the advantage that initially attracted capital.

For governments seeking investment through relationships with competing blocs, the analytical implication is to distinguish temporary diversion from durable development. Attracting projects may generate opportunities, but lasting benefits would depend on domestic capabilities, supplier relationships and workforce development. Those outcomes are not demonstrated by this model. The study supports caution about treating connector status as a permanent route to technological advancement.

Policy Needs a Wider Toolkit and Firmer Evidence

The authors frame the policy challenge as a "Fragmented-AI Trilemma": economies cannot simultaneously preserve monetary autonomy, allocate AI capital efficiently and insulate themselves from geopolitical shocks. Their simulations suggest that responding more strongly to exchange-rate movements can improve some outcomes while increasing inflation volatility. The broader proposition is that monetary policy alone cannot resolve all the pressures created by disrupted technology financing.

The proposed response combines financial safeguards, multilateral AI-investment arrangements and targeted capital-flow measures. In practical terms, this points towards assessing firms' foreign-financing exposure, preserving access to productive technology and protecting households vulnerable to labour-income losses. Development institutions could investigate financing arrangements that support access across political divides, although the paper does not establish which programme designs would work best.

Its evidence also requires careful qualification. The model explicitly represents household differences and financing constraints, and the authors report numerous sensitivity exercises. However, its AI-capital calibration draws on broader information-technology statistics as an upper-bound proxy. Simplified economic blocs cannot capture the institutional diversity of developing countries, while the persistence of geopolitical shocks strongly influences the size of the estimated losses.

More seriously, some internal explanations do not reconcile with the stated calculations. One passage describes an almost fourfold rise in financial friction between scenarios, whereas the published formula and estimated parameters imply an increase of about 3%. The trilemma argument also makes conflicting claims about whether increasing or removing financial barriers maximises exposure to geopolitical shocks. These discrepancies require clarification before the precise estimates or theoretical proposition become reliable policy benchmarks.

The paper consequently offers a research agenda whose mechanisms need direct empirical testing. Comparable data on AI investment, international financing and household outcomes would help distinguish geopolitical effects from domestic constraints. Policy evaluation must also account for the security benefits governments seek from restrictions, which this analysis does not quantify. Without that comparison, the full costs and benefits remain unresolved.