China pursues centralized, state‑led data governance to mobilize large, sovereign data projects; the United States relies on decentralized, rights‑based institutions emphasizing transparency and accountability. Rather than strict opposites, openness and security are treated as layered, adaptable governance processes with distinct implications for data access, innovation and competition.

Main Finding

China and the United States pursue distinct institutional logics to manage the openness–security tension in scientific data governance. China favors a centralized, security‑driven “techno‑sovereignty” model that treats scientific data as a strategic state asset and emphasizes unified control. The U.S. favors a decentralized, rights‑and‑transparency‑oriented model in which federal openness mandates coexist with sectoral exceptions and market-led infrastructures. The authors conceptualize openness–security trade‑offs as layered, adaptive coordination processes (modular coordination) rather than a binary choice.

Key Points

• Research questions
  • RQ1: How do China and the U.S. structure policies to coordinate openness and security?
  • RQ2: Which coordination mechanisms enable or constrain alignment across competing public values?
• Comparative framework (four analytical axes)
• Coordination objective: openness-for-innovation vs. securitized control for national interests.
• Actor type: ministries/regulatory bodies vs. polycentric actors (federal, state, private).
• Coordination mechanism: statutory mandates, portals, funding incentives, audits, etc.
• Stakeholder legitimacy: how policies are justified to publics and actors (collective development vs. individual rights/procedural legitimacy).
• Country contrasts
  • China: centralized, developmentalist, security-first. Key features include national laws and ministerial regulations, emphasis on data sovereignty, centralized oversight (e.g., MOST, CAC), and policy layering that privileges dominant security layers.
  • United States: polycentric, transparency-and-rights-based. Features include legislative acts, executive orders, interagency plans, and sectoral autonomy; openness is promoted but implementation is fragmented by security and proprietary exceptions.
• Theoretical contribution: extends modular coordination theory by framing openness–security as multi-level, adaptive institutional processes embedded in political regimes and legitimacy economies.
• Important caveat: analysis is based on national‑level formal policy texts; the study does not directly observe enforcement, implementation outcomes, or public reactions.

Data & Methods

• Comparative most-different-systems design: China vs. United States.
• Corpus: 36 national-level policy documents (18 China; 18 U.S.), purposively sampled for relevance to scientific data governance and legal/strategic standing (laws, ministerial regs, executive orders, strategic plans).
• Timeframe: documents from 2003–2025 with policy bursts around 2019 and 2021–2024.
• Data processing: manual conversion to machine‑readable format and qualitative coding in NVivo.
• Analytical approach: qualitative content analysis mapped to the four-dimensional comparative coordination framework and interpreted through institutional logics and modular coordination literatures.
• Limitations: reliance on formal texts (no enforcement data); variation in subnational/sectoral implementation not directly observed; findings interpret discursive and design features rather than observed outcomes.

Implications for AI Economics

• Data availability and competitive advantage
  • Regime type shapes the supply and cross-border availability of scientific datasets that are inputs to AI development (e.g., biomedical, geospatial, genomics). China’s sovereignty‑centric controls can create concentrated national datasets for domestic AI firms while restricting international access. The U.S. system may permit broader private access domestically but produces fragmentation that can hinder large consolidated datasets.
  • Models of AI productivity and innovation should explicitly account for regime-dependent data access constraints and potential data localization effects on training data pools, sample diversity, and model generalizability.
• R&D costs, investment decisions, and risk pricing
  • Regulatory complexity and uncertainty (policy layering, security vetos, variable enforcement) raise compliance costs and regulatory risk premia for firms investing in data‑intensive AI R&D or cross‑border collaborations.
  • Investors and firms will price in political/regulatory regime risks differently: China may offer predictable state support but limited international markets; U.S. markets may enable global commercial paths but expose firms to fragmented rules and private‑property constraints on data.
• Market structure and industrial organization
  • China’s centralized approach can favor incumbent, state‑linked actors and platform firms aligned with state priorities; it can speed national deployments but reduce international interoperability.
  • The U.S. polycentric model gives strategic roles to private intermediaries (data aggregators, cloud providers, compliance vendors), potentially creating market opportunities for data‑governance services but also risks of proprietary lock‑in that limit open research ecosystems.
• Innovation diffusion and global competition
  • Divergent governance raises frictions for international scientific collaboration and for the global diffusion of AI capabilities. Data‑sharing barriers and incompatible modules (legal/technical) increase transaction costs for multinational research and model training.
  • Scenario analyses in AI economics should include “governance regimes” as parameters affecting knowledge spillovers, cross‑border talent flows, and coalition formation (e.g., public‑private consortia).
• Policy design and incentive alignment
  • Modular coordination (interoperable legal/technical modules, audits, third‑party certifiers) offers a policy lever to reduce frictions while preserving security goals. Economists and policymakers can evaluate cost‑benefit tradeoffs of modular standards, certification markets, and subsidy designs to encourage sharing under controlled regimes.
  • Designing incentives (grants, procurement, standards) that reward compliant data sharing could increase usable shared datasets for AI without undermining legitimate security safeguards.
• Legitimacy, adoption, and demand for governance services
  • Public legitimacy and trust affect adoption of data‑sharing arrangements; legitimacy deficits can suppress participation in open data initiatives, reducing dataset sizes available for AI. Governance that foregrounds procedural transparency and accountability (as in the U.S. discourse) may support adoption but requires mechanisms to avoid symbolic transparency.
  • Demand for compliance, auditing, and interoperability services will grow—markets for data governance, certification, and privacy‑preserving tooling are economically consequential.
• Modeling recommendations for AI economists
  • Incorporate regime‑specific parameters: data access probability, compliance cost distributions, enforcement variance, and international data‑transfer frictions.
  • Use modularity in counterfactuals: model effects of introducing interoperable policy “modules” (e.g., standard data use agreements, certified enclaves) on innovation output, social surplus, and competitive dynamics.
  • Consider endogenous firm responses: data‑sharing strategies, vertical integration (to internalize data access), and investment in data‑synthesis or synthetic data technologies to mitigate access constraints.
• Broader implications for public value and global welfare
  • Tradeoffs between openness and security affect not just firm profit but social benefits from AI (e.g., reproducible science, public health AI). Policy choices that overly securitize data could slow socially valuable AI innovations; conversely, lax controls can create national security or privacy harms that impose systemic costs.

Short concluding note: For scholars and practitioners in AI economics, the paper highlights that data governance regimes are foundational determinants of data flows, costs, and incentives. Quantitative models, investment analyses, and policy evaluations should explicitly model institutional logics and modular governance features to produce realistic projections of AI innovation trajectories and market outcomes.