Markets where firms hold algorithmic advantages see fewer new entrants and rising concentration; concentrated data and cumulative learning deepen incumbents' edge, while interoperability and data access can restore contestability and inform competition policy reform.

Main Finding

High levels of algorithmic advantage—driven by cumulative learning, concentrated proprietary data, AI talent, and compute/investment—are consistently associated with lower entry rates, longer time-to-entry for newcomers, and higher market concentration in AI-intensive markets. Data concentration amplifies these exclusionary effects, while interoperability and data-access measures mitigate them. Results are robust to alternative index weightings but are interpreted as strong associative (not strictly causal) evidence due to endogeneity constraints.

Key Points

• Algorithmic advantage is dynamic and cumulative: performance improves with continued market engagement, producing path-dependent learning benefits that differ from traditional scale economies (which are cost-based).
• Entry barriers in AI-driven markets are largely capability- and asset-based (data, compute, AI-skilled labor), not only cost- or regulation-based. These include a "cold-start" or data bottleneck, expertise asymmetries, network/ecosystem lock-in, and switching costs.
• Data concentration creates feedback loops: better algorithms attract users → more data → better algorithms, reinforcing incumbency and leading to "winner-takes-most" outcomes even absent explicit anticompetitive pricing or conduct.
• Manifestations of barriers: both fewer entrants (reduced entry rate) and delayed arrivals (longer time-to-entry), which cumulatively raise concentration (HHI, CR4).
• Regulatory interventions that lower frictions to data access—interoperability, mandated APIs, portability, and data-access obligations—reduce the translation of algorithmic advantage into persistent exclusionary barriers without necessarily destroying incentives to innovate.
• Opacity of modern ML systems complicates detection of exclusionary conduct; traditional antitrust tools centered on short-term price effects and narrow market definitions may miss structural, long-run harms.

Data & Methods

• Scope: Unbalanced panel of AI-intensive industry-country markets in 29 OECD countries, 2012–2024.
  • Market types: digital advertising/MarTech, e-commerce/marketplaces, fintech (credit scoring, payments, fraud detection), mobility/logistics platforms, cloud enterprise software and data-analytics services.
  • Sample: 142 industry-country markets, ~1,240 firm–market–year observations, 318 unique firms.
• Key variables:
  • Algorithmic Advantage Index (AAI): composite z-scored index combining AI investment intensity, AI-related patent activity (normalized), AI-skilled labor share, and algorithmic performance proxies (industry benchmarks / disclosures). PCA-based weighting used for robustness.
  • Data concentration: share of proprietary data held by top-four firms (volume and exclusivity); alternatives incorporating persistence and diversity used in robustness tests.
  • Interoperability/Data-Access index: composite of regulatory binary indicators (portability/API obligations), technical interoperability measures, and jurisdictional mandates (e.g., DMA-like provisions).
  • Outcomes: market entry rate (new firms per market-year), average time-to-entry (months), market concentration (HHI, CR4).
• Econometric approach:
  • Fixed-effects panel regressions with market and year fixed effects; lagged independent variables to reduce simultaneity.
  • Interaction terms to probe dynamic learning (e.g., firm size × time).
  • Missing data: listwise deletion for core vars, lag-based imputation for select controls; sensitivity checks across samples.
• Identification caveat:
  • Authors acknowledge possible endogeneity (AAI and data concentration may be endogenous to market structure). No strong external instruments available for proprietary algorithm/data assets; results presented as associative evidence consistent with cumulative learning and path dependence, not definitive causal estimates.

Implications for AI Economics

• Rethinking market power metrics: Standard competition analysis focused on short-run price effects can miss structural, quality- and data-driven exclusion. AI economics should develop and standardize non-price measures (e.g., AAI-like indices, data concentration metrics, time-to-entry) for monitoring competition.
• Dynamic contestability matters: Policies and models should account for path dependence—early advantages compound—so static equilibrium analysis underestimates barriers and lock-in. Empirical and theoretical models should incorporate cumulative learning processes.
• Policy levers beyond pricing remedies:
  • Interoperability, mandated APIs, and data portability can be effective, targeted tools to reduce exclusionary feedback loops while retaining innovation incentives.
  • Pro-competitive data strategies (regulated data-sharing frameworks, data trusts, access obligations for essential datasets) merit evaluation as complements to traditional antitrust enforcement.
• Regulation design & evidence needs:
  • Ex ante or structural regulation (e.g., obligations for dominant AI platforms) may be necessary because harms evolve over time and are hard to detect ex post due to opacity.
  • Regulators need richer, longitudinal firm- and dataset-level data to detect cumulative advantage; investment in data collection and technical auditing capacities is essential.
• Research agenda:
  • Causal identification: pursue instruments, natural experiments (e.g., sudden policy changes), or firm-level quasi-experiments to isolate causal effects of algorithmic advantage and data concentration.
  • Microdata and model validation: obtain proprietary performance and dataset measures to validate proxies and refine indices like AAI.
  • Welfare analysis: quantify consumer surplus, innovation incentives, and dynamic welfare trade-offs of interventions (e.g., mandatory data sharing vs. reduced R&D returns).
• Broader consequence: AI-driven market structures can produce durable incumbency even without predatory pricing or explicit collusion. AI economics must therefore expand focus to structural and dynamic sources of market power and to policy designs that preserve competition over time.