Frontier AI could widen global divides unless developing countries get capacity and compute access; coordinated compute governance, skills-building and knowledge sharing offer a practical path to more inclusive outcomes.

Main Finding

The paper argues that AI’s global diffusion—especially the shift from narrow AI toward more general-purpose systems—creates interdependent risks and opportunities for the Global South. Rather than a simple catch-up problem, equitable outcomes require a cooperative international framework centered on three pillars: capability building, responsible innovation, and knowledge/compute sharing. Compute governance (the measurement, monitoring, and allocation of computational resources) is highlighted as a concrete, tractable lever to reduce asymmetries and enable inclusive AI-driven development.

Key Points

• Structural asymmetries

  • Frontier compute and research are highly concentrated: the paper cites >90% of frontier AI compute capacity in the U.S. and China and <1% of African research output (as of 2025). Over 80% of advanced semiconductor fabrication capacity is in Taiwan, South Korea, and the U.S.
  • These concentrations risk creating “AI-enabled dependency” for the Global South (data extraction, imported models, weak domestic capability formation).

• Service-led development opportunities

  • AI-enabled services (digital finance, health, education, creative industries, digital outsourcing) provide realistic, high-return pathways for developing countries to leapfrog structural bottlenecks.
  • Examples: India’s digital public infrastructure (Aadhaar, UPI); rapid ASEAN digital economy growth; M-Pesa and GCash as inclusive finance cases.

• Compute governance as a policy lever

  • Compute is physically measurable (e.g., FLOPS), geographically instantiated, and subject to supply-chain controls, making it more tractable for multilateral governance than data or algorithms alone.
  • Proposed mechanisms include compute monitoring regimes, permit or allocation systems, and multilateral compute-sharing arrangements.

• Cooperative framework & policy recommendations

  • Three strategic priorities: capability building, responsible innovation/regulatory alignment, and knowledge sharing/digital commons.
  • Six policy directions: (1) global capability-building partnerships, (2) open & multilingual AI ecosystems, (3) sustainable digital infrastructure (regional data centers/clouds), (4) shared ethical/regulatory standards, (5) triangular and South–North–South cooperation, (6) international compute governance architecture.

• Risks of fragmentation

  • Without cooperation: deepened inequality, data/algorithmic colonialism, duplicated infrastructure, incompatible regulation, coordination failures, and reduced global welfare. Fragmentation could also increase safety risks from uncoordinated frontier development.

Data & Methods

• Type of study: Perspective / conceptual synthesis.
• Methods used:
  • Literature review of technological change, development economics, global value chains, and emerging compute-governance scholarship.
  • Conceptual argumentation linking theoretical literatures (capability accumulation, digital divide, services-led growth) to policy design.
  • Use of secondary empirical facts from cited sources (e.g., AI Index 2024, arXiv papers on compute governance) to motivate claims.
• Data availability:
  • No original empirical data were created or analyzed; the paper is a normative and theoretical contribution based on existing literature and public reports.

Implications for AI Economics

• Compute as a factor of production: Economists should treat compute as a scarce, geographically concentrated input that shapes innovation rents, entry barriers, and cross-country comparative advantage in AI. This calls for models that endogenize compute distribution and access.
• Distributional & welfare impacts: The paper implies large cross-country distributional effects—automation and AI could exacerbate inequality absent welfare or reskilling responses. Macro models should incorporate heterogeneous country responses (welfare systems, institutions, human capital) to AI shocks.
• Trade and global value chains: AI shifts comparative advantage toward services and digital trade. Trade models need to account for digital public infrastructure, platform-mediated services, and bilateral complementarities (Northern frontier tech × Southern service supply).
• Policy instruments & governance design: Compute governance opens a new suite of regulatory instruments (monitoring/reporting, permit allocation, multilateral compute-sharing) that can be modeled for incentive and welfare consequences. Analyses of these instruments should weigh innovation incentives, safety externalities, and equity.
• Research agenda suggestions for economists:
  • Quantify the effect of compute access on national AI research/output and productivity.
  • Evaluate welfare impacts of compute-allocation mechanisms vs. market outcomes.
  • Model dynamic paths for service-led development under varying degrees of compute sharing and regulatory coordination.
  • Empirically assess reskilling, digital public infrastructure, and regional data centers as mitigants of AI-driven labor dislocation.
• Safety and systemic risk: Lack of coordinated standards increases global accident/misalignment risk; economic models of international coordination and common-pool resource governance are relevant for AI safety policy analysis.
• Practical policy evaluation: Cost–benefit and general-equilibrium studies of proposed cooperative measures (e.g., subsidized cloud partnerships, shared compute consortia, data trusts) are needed to guide multilateral investments and treaty design.

Limitations to consider when applying this paper: - It is conceptual and normative rather than empirical; policy claims require quantitative validation. - Proposals (especially compute governance) will face political economy constraints—modeling should incorporate strategic state and firm behavior.

If useful, I can convert these implications into a brief research agenda with specific empirical approaches and datasets (e.g., cloud usage statistics, patent/research output, trade in digital services) for pursuing the most urgent open questions.