Opening municipal government data measurably boosts local AI innovation and ripples into neighbouring cities; city-level staggered DID estimates across 281 Chinese cities (2008–2023) show significant direct and spatial effects, with private firms capturing the largest share of gains.

Main Finding

Open government data (OGD) significantly promote artificial intelligence innovation (AII) at the city level and generate positive spatial spillovers to neighbouring cities. The effect is strongest for enterprise-driven AII; enterprises capture most of the innovation gains from OGD compared with universities/colleges and research institutes.

Key Points

• The paper frames the OGD → AII link within the data-driven innovation (DDI) paradigm: data is a core production factor for AI, and OGD supplies, reduces the cost of, and improves access to training and evaluation data for models.
• Mechanisms identified: OGD (i) increases data supply and diversity (multimodal, cross-sector datasets), (ii) lowers data acquisition and preprocessing costs, (iii) enhances data quality/coverage for real-world scenarios and model iteration, and (iv) enables cross-boundary data circulation (public-good attributes) that fosters intercity knowledge/resource flows.
• OGD exhibits spatial externalities: because data are partially non-rival and transportable digitally, local openness benefits neighbouring jurisdictions’ AII activity.
• Heterogeneity: firms (enterprises) are the principal beneficiaries and drivers of OGD-facilitated AI innovation, more so than universities or research institutes.

Data & Methods

• Sample: 281 Chinese cities over 2008–2023.
• Empirical strategy: staggered difference‑in‑differences (DID) to identify the direct effect of cities adopting/opening OGD platforms over time, combined with a spatial DID (spatial econometric extensions) to detect cross-city spillovers.
• Outcome measures: city-level AII and entity-specific AII (decomposed into enterprises, universities/colleges, and research institutes) across the period (the paper constructs city- and entity-level indicators of AI innovation for the analysis).
• Heterogeneity analysis: separate estimation by innovation entity to assess which actors drove the OGD → AII relationship.
• Robustness: the combination of staggered DID and spatial DID addresses time-varying adoption and spatial dependence (as reported by the authors).

Implications for AI Economics

• Data as a factor of production: empirical evidence that public-sector data openness is a scalable input that raises AI innovation suggests models of AI production should explicitly treat data (and its public-good/externality properties) alongside compute and algorithmic capital.
• Local policy yields regional returns: OGD policies produce spillovers across jurisdictions, so optimal policy design should internalize spatial externalities (e.g., regional coordination, shared platforms, cost‑sharing).
• Firms as the primary channel: because enterprises capture most of the innovation gains, policies aimed at boosting AI innovation via OGD should prioritize firm access and capabilities (APIs, toolkits, data licensing clarity, incubation supports) while preserving privacy/security.
• Market structure and competition: abundant public data can lower entry costs for AI startups and affect incumbents’ competitive dynamics; regulators and economists should consider how open data reshapes barriers to entry, returns to scale in model training, and platform market power.
• Public investment priorities: investments in data quality, standardization, metadata, multimodal corpora, and interoperable platforms are likely high-return for AII. Complementary support—such as compute access, incentives for data reuse, and outcome‑oriented procurement—can amplify benefits.
• Evaluation and accounting: economic measurement of AI R&D and productivity should incorporate data availability and data‑sharing policies as explanatory variables; welfare assessments of OGD must account for local and cross‑regional innovation externalities.
• Policy risks to manage: while promoting openness, policymakers must balance privacy, security, and commercial confidentiality; designing licensing, governance, and access-control regimes that preserve reuse while mitigating harms is crucial.

(Concise recommendations implicit from findings: scale up high‑quality, multimodal OGD; enable cross‑jurisdictional data sharing; support firms’ ability to use OGD; and internalize spatial externalities in regional policy design.)