AI readiness is broadly distributed across a geographically dispersed cluster of capable states rather than concentrated in a single hegemon, identifying latent institutional and infrastructural capacity for influence in future AI governance. Short-panel and event-study diagnostics show only modest and statistically uncertain short-run effects on governance and diplomatic proxies, so readiness signals potential rather than immediate geopolitical power.

Main Finding

AI readiness—measured across multiple published indices—loads on a single latent construct (high inter-index concordance: Cronbach’s α = 0.93; first principal component ≈ 76% of variance). That latent AI-capacity is geographically dispersed across world regions (a multi‑regional/multipolar pattern of readiness), even though key hardware, specialized chips, and hyperscale cloud remain concentrated in a few firms/jurisdictions. Readiness correlates positively with governance and diplomatic‑capacity proxies cross‑sectionally, but short panels and event‑study checks show only small and statistically uncertain temporal effects; therefore the authors interpret readiness as an indicator of latent capacity rather than evidence that it currently produces immediate diplomatic power.

Key Points

• Measurement: Five complementary indices (GARI, IMF AIPI, Stanford AI signals composite, UN EGDI, World Bank fixed broadband) converge on a common AI‑readiness construct.
  • Cronbach’s α = 0.93; first PC ≈ 76% variance explained.
• Global distribution:
  • Readiness is widely distributed across regions (not strictly a US–China duopoly).
  • Many mid‑sized and advanced economies (and some small states) show meaningful readiness.
  • Core compute, chip manufacturing, and cloud remain concentrated → dependencies and chokepoints persist.
• Associations with governance and diplomacy:
  • Cross‑sectional analyses: positive associations between readiness and proxies for governance effectiveness (EGDI, e‑participation) and diplomatic capacity.
  • Temporal diagnostics (country fixed effects panel 2019–2023; staggered DiD around first national AI strategy) show small, uncertain effects — interpreted as suggestive, not causal.
• Conceptual framing:
  • AI readiness = stock of enabling inputs (compute/connectivity, data & governance, models/software, human capital/organisation, institutions).
  • Readiness is analytically distinct from outcomes (state capacity, governance effectiveness, diplomatic influence).
• Policy recommendations (high level):
  • Treat AI readiness (skills, infrastructure, data governance, innovation capacity) as both domestic and foreign‑policy priorities.
  • Prioritise inclusive, interoperable global governance (OECD, UNESCO, GPAI, UN).
  • Invest in capacity building for lagging states; diplomats should adopt AI ethically and transparently; guard against widening divides and weaponised interdependence.
• Analytical caveats:
  • Short time series, data alignment (EGDI biennial), partial construct overlap, and potential omitted confounders → avoid strong causal claims.

Data & Methods

• Primary datasets
  • Government AI Readiness Index (GARI, 2023): 193 countries, 0–100 scale; covers gov’t capability, tech sector, human capital, data, infrastructure.
  • IMF AI Preparedness Index (AIPI): ~174+ countries, 0–100, macro digital/AI preparedness.
  • Stanford AI Index (selected signals): constructed composite from country‑level signals (private AI investment USD, share of global AI publications, talent proxy, count of national AI policy initiatives). Components standardized (z‑scores) then averaged.
  • UN E‑Government Development Index (EGDI): 193 UN members, 0–1; online services, telecom infrastructure, human capital. EGDI is biennial (2022 used for 2023 alignment where needed).
  • World Bank WDI: Fixed broadband subscriptions (per 100 people), latest available (~2021–2024).
• Samples and construction
  • Core cross‑section sample: ~149 countries with complete data across key indices (coverage table in supplement).
  • Short panel: 2019–2023, raw up to 899 country‑year observations; main lagged‑GARI FE sample ≈ 495 country‑year observations from ~177 countries (listwise deletion applied per model).
  • Harmonised country names; EGDI values carried forward one year to align vintages when necessary.
• Empirical strategy
  • Measurement validation: Cronbach’s alpha, principal component analysis to establish a common latent AI‑readiness construct.
  • Cross‑sectional associations: compare indices and correlate readiness with governance/diplomacy proxies (EGDI, e‑participation, broadband).
  • Panel analysis: country and year fixed‑effects regressions using lagged GARI to predict next‑year outcomes (clustered SE by country) to test within‑country changes.
  • Event study: staggered difference‑in‑differences (Sun & Abraham; Callaway & Sant’Anna) around each country’s first national AI strategy (event‑time coefficients up to 3 years before/after).
  • Robustness: pre‑trend tests, placebo checks, overlap‑controlled sensitivities, alternative alignments described in Supplementary Appendices A7–A8.
• Limitations explicitly acknowledged
  • Short post‑treatment windows and limited temporal coverage limit causal inference.
  • Partial construct overlap across indices; some indices emphasise governance, others R&D/market signals.
  • Listwise deletion may exclude some small/conflict states; some indices omit specific countries.

Implications for AI Economics

• Readiness as latent productive capacity
  • AI readiness packages inputs (skills, data governance, infrastructure, institutions) that condition a country’s ability to adopt, deploy, and capture economic value from AI — analogous to human‑capital or technological capability endowments in growth models.
  • Because readiness is a stock (not immediate output), its economic effects (productivity gains, comparative advantage shifts) may materialise with lags and depend on complementary investments and regulation.
• Geography of comparative advantage and specialization
  • Multipolar distribution of readiness implies more geographically dispersed potential AI hubs and centers of innovation, which can diversify sources of AI‑driven growth globally.
  • However, concentration of compute, specialized chips, and cloud services creates choke points: countries or firms controlling these inputs can extract rents, shape prices for compute, and influence where value accrues in global value chains.
• Trade, investment, and global value chains
  • Countries with higher readiness are better positioned to attract AI‑related FDI, scale digital services exports, and climb value chains into higher‑value AI tasks (model development, data services, platform orchestration).
  • Lower‑readiness countries risk being relegated to data sources or low‑value tasks unless they invest in skills, data governance frameworks, and institutional capacity.
• Policy levers and economic development
  • Targeted investments in broadband, computing access, AI education, and public‑sector digitalisation can be economically productive interventions; multilateral development banks and donors should prioritise “AI readiness” elements to avoid widening global inequality.
  • Export controls, sanctions, and tech transfer policy (e.g., restrictions on advanced chips) will shape how readiness translates into domestic economic gains; such policies can create fragmentation and economic frictions.
• Market structure, firms, and rents
  • Concentrated cloud/compute implies platform and infrastructure firms may accrue outsized market power and profits; economic policy should consider competition, data portability, and interop standards to limit rent extraction and enable downstream competition.
• Regulation and institutions
  • Differences in data governance, privacy, and competition law will affect international flows of data and AI services — regulatory heterogeneity can produce arbitrage and reshape comparative advantage.
  • International standards and interoperable rules (OECD, UNESCO, UN fora) can lower transaction costs and enable broader diffusion of economic benefits.
• Distributional risks and surveillance economies
  • Rapid AI adoption without rights‑based governance risks concentration of gains, surveillance markets, and increased inequality — these outcomes have macroeconomic implications (consumption, political economy stability) relevant for growth models.
• Forecasting and economic modelling
  • Economists and modelers should treat AI readiness indices as plausible leading indicators of a country’s future AI‑driven economic potential, but incorporate lags, dependence on imported compute, institution quality, and policy environment when projecting outcomes.
  • Structural models of productivity should include both the enabling inputs (readiness) and mediating institutional variables that determine whether inputs convert to productivity.
• Research and policy priorities for AI economics
  • Empirically: track longer time series to identify causal pathways from readiness to productivity, trade, and wages.
  • Policy: align industrial policy, education, and development finance to strengthen readiness while addressing chokepoints (e.g., promoting regional data centres, diversified supply chains for semiconductors).
  • Multilateral cooperation to support readiness in low‑ and middle‑income countries can mitigate global inequality and integrate them into AI value chains.

In short: the paper provides a validated, multi‑index measure of national AI readiness and documents a geographically dispersed readiness landscape with positive governance correlations. For AI economics, readiness should be treated as a latent capability shaping future comparative advantage, investment flows, and productivity—but its economic payoff depends on complementary assets, institutional quality, and exposure to concentrated hardware/cloud markets.