AI innovation is concentrated in a handful of advanced economies and shows only a weak statistical link to aggregate productivity, while gains instead appear tied to localized accumulation of intangible capital; the result implies growth from AI may be concentrated rather than broadly diffused.

Main Finding

The paper introduces the Relative Economic Growth Illusion (REGI): contemporary AI-driven technological expansion often produces concentrated, intangible‑intensive gains (firm valuation, localized productivity, market power) that do not translate into broad aggregate Total Factor Productivity (TFP) growth. Empirically, AI patent intensity shows weak/insignificant association with aggregate TFP, while accumulation of intangible capital is strongly associated with localized sectoral productivity gains and with geographically concentrated AI innovation.

Key Points

• REGI concept: distinguishes absolute growth (broad-based, inclusive productivity increases) from relative growth (redistributive gains concentrated in technologically dominant firms/sectors).
• Theoretical innovation: extends a Romer-style endogenous growth model to allow heterogeneous AI returns (A_i,t = A0 + φ_i AI_i,t) where φ_i increases with market share (φ_i = φ0 s_i^γ; γ > 1), formalizing how scale, data advantages, and network effects produce increasing returns and concentration.
• Empirical findings:
  • AI patent intensity is only weakly and statistically insignificantly correlated with aggregate TFP across the sample.
  • Intangible capital accumulation (software, data, organizational capital, IP) is positively associated with productivity gains, but primarily at the sectoral and local (cluster/hub) level.
  • AI-related innovation and intangible investment are geographically concentrated in a small set of technologically advanced economies and urban functional areas.
  • Evidence of rising market concentration and declining labor shares in AI-intensive industries is consistent with redistributive rather than broad productivity effects.
• Complementary mechanisms highlighted: economies of scale in AI, low marginal replication costs, network externalities, proprietary data/cloud infrastructure, and financialized valuation expectations.
• Policy warning: without redistribution, competition and diffusion policies, AI-driven efficiency gains may weaken aggregate demand and amplify inequality even as firm-level metrics improve.

Data & Methods

• Data sources: OECD Productivity database, OECD STAN (industry-level), OECD Patents, INTAN‑Invest (measures of intangible capital), and Functional Urban Areas (FUAs) datasets for geographical concentration.
• Empirical strategy:
  • Descriptive analysis to document concentration of AI patents, intangible investment, and sectoral productivity patterns.
  • Panel and robust regression techniques to estimate relationships between AI innovation intensity (patent-based measures and related proxies), intangible capital accumulation, market concentration indicators, labor shares, and aggregate/sectoral TFP.
  • Concentration-adjusted productivity estimation to account for uneven distribution of AI gains across firms/sectors.
  • Counterfactual macroeconomic simulations calibrated from the extended endogenous growth model to explore long-run outcomes under different diffusion, competition, and redistribution scenarios.
• Key variables: AI patent intensity, measures of intangible capital stocks/flows, sectoral output and productivity (TFP), market share/concentration metrics (e.g., sectoral HHI or s_i), labor share indicators, geographic clustering from FUAs.
• Robustness and limitations noted by author:
  • Patent counts and related indicators are imperfect proxies for AI diffusion and productive use.
  • Potential implementation lags and measurement issues in capturing AI’s full effect on productivity.
  • Causal identification is challenging given simultaneity between firm success and intangible accumulation; regressions are complemented by simulations and sensitivity checks.

Implications for AI Economics

• For researchers:
  • Move beyond firm-level success stories: analyze sectoral and geographic diffusion, intangible capital complementarities, and distributional effects.
  • Improve measurement of AI adoption (beyond patents), intangible stocks, and AI use intensity; collect richer micro-data linking firm AI use to output and employment.
  • Incorporate market structure and demand-side channels when modeling AI’s macroeconomic impact (not only supply-side TFP channels).
• For policy:
  • Competition policy and antitrust: mitigate winner‑take‑all dynamics that lock-in data and compute advantages and impede productive diffusion.
  • Data governance and public data sharing: facilitate access to complementary datasets and infrastructure to reduce barriers to diffusion.
  • Support for complementary investments: public incentives for intangible capital adoption (training, organizational change, software/Cloud access) in lagging sectors to realize productivity spillovers.
  • Redistribution and demand support: fiscal and labor-market measures (wage support, progressive taxation, social transfers) to maintain aggregate demand if productivity gains concentrate in capital income.
  • Measurement and monitoring: update national accounts and productivity statistics to better capture intangible-driven value and monitor whether observed growth is broad-based or concentrated.
• Normative conclusion: AI has strong potential to raise productivity, but realizing inclusive, aggregate gains requires active policies addressing diffusion, competition, and distribution. Without them, economic expansion may largely reflect concentrated intangible-driven growth and financialized valuations rather than generalized improvements in real productive capacity.