AI acts less like a neutral productivity tool and more like a rent-bearing asset, concentrating returns through access control and eroding the value of higher education; fixing labour-market harms requires tackling ownership and access, not just skill policies.

Main Finding

Tholen (2026) argues that AI should be analysed not only by what it can technically do to jobs but as a productive rentier asset whose ownership and access control—rather than capability alone—determine who captures productivity gains. Under contemporary rentier dynamics, AI infrastructure and proprietary models enable enclosure and constructed scarcity, decoupling returns from university-level human capital and shifting rewards toward asset owners. As a result, credential value and returns to educational investment can decline even where AI augments productivity.

Key Points

• Core claim: AI functions as a productive rentier asset—expensive to develop, concentrated in few firms, and both constitutive of production and a source of extractable rent through enclosure (access pricing, APIs, platform lock-in).
• Two mechanisms driving labour outcomes:
  • Substitutability: When capabilities embodied by credentials become available as cheap, ownable AI tools, workers’ bargaining fallback is weakened because employers can substitute asset access for credentialled labour.
  • Enclosure: Concentrated infrastructure (cloud, proprietary models, hardware) captures productivity gains through access control and ecosystem lock-in; gains are appropriated upward as rent rather than diffused to labour.
• Distinction between asset types: “Productive rentier assets” (AI systems, platforms, data infrastructure) actively shape the labour process and extract rents, unlike inert passive assets (land, financial claims).
• Consequence for human capital theory: The standard assumption that educational investment reliably secures higher wages via productivity gains is regime-dependent. Under rentier assetisation, returns accrue to asset controllers, producing credential devaluation and declining returns to education.
• Empirical signatures noted: declining labour share since 1970s; concentration in Big Tech (winner-takes-all/network effects); evidence of AI exposure being higher in degree-intensive occupations; documented declines in freelance incomes post-AI adoption in some markets.
• Analytical stance: Moves beyond capability/exposure studies that treat technological capability as exogenous; instead foregrounds ownership, organisational strategy, and political-economic regime.
• Limitations acknowledged: Uses an asset-centred definition of rent (rather than a Marxian profit-rate test); does not claim AI firms universally achieve persistent supernormal profits; does not fully adjudicate whether rentierism is a new epoch or intensification of existing tendencies.

Data & Methods

• Methodological approach: theoretical and conceptual synthesis grounded in political economy and sociology; systematic literature review and integration of empirical findings from multiple recent studies (exposure studies, empirical labour-market analyses, platform/competition research).
• Evidence types used: secondary empirical results (e.g., Henseke et al., Felten et al., Bloom et al., Hui et al.), macro trends (labour share decline, Big Tech market shares), and case examples illustrating enclosure and platform dynamics.
• Analytical devices: articulation of a two-layer model (competitive application layer vs concentrated infrastructure layer) and the productive rentier asset concept; engagement with critiques from Marxian and heterodox traditions and clarification of conceptual choices.
• No original microdata or novel econometric estimation presented; argument is inferential and interpretive, mapping mechanisms to observed patterns and proposing empirical implications.

Implications for AI Economics

• Rethink models of technological change and labour: incorporate ownership, access control, and institutional regime into assessments of how AI affects wages and employment; move beyond capability/exposure metrics alone.
• Empirical research agenda: measure enclosure and assetisation (market concentration, API dependence, access pricing, licensing regimes); compare wage and employment outcomes across sectors/countries with different ownership regimes; test predicted divergence between productivity gains and labour compensation in AI-intensive activities.
• Policy and regulation focus:
  • Competition and antitrust: address platform monopolies and ecosystem lock-in to reduce enclosure effects.
  • Access and infrastructure policy: promote open, public or cooperative alternatives to proprietary AI infrastructure (public models, shared compute, data trusts).
  • Intellectual property and licensing: reform IP/IP-like regimes that create constructed scarcity around AI capabilities and datasets.
  • Taxation and redistribution: capture unearned rents through targeted taxes or royalties on AI infrastructure rents and direct proceeds toward labour-supporting measures.
  • Labour-side measures: strengthen collective bargaining, portable credentials, and worker access to AI tools (e.g., mandated interoperable APIs, workplace governance of AI deployment).
• Education policy: treat upskilling as necessary but not sufficient; protections for workers cannot rely on credential accumulation alone when ownership and access determine rent capture.
• Normative and measurement shift: AI economics should explicitly quantify rent extraction channels and differentiate productivity effects that increase worker productivity from those that primarily enlarge asset-owner returns.

Summary takeaway: To understand AI’s labour-market consequences—especially for university-educated workers—research and policy must foreground who owns and controls AI assets and how access is priced and enclosed. Technological capability tells which tasks are exposable; ownership and institutional regime tell who benefits.