The Rent-Seeking Singularity: How AI Monopolies Are Converting Intelligence Into Intellectual Property

By Yuto Yamaguchi Sep. 3, 2024

The rhetoric of artificial intelligence is that of revolution—creative destruction, democratization, disruption. Yet beneath the spectacle of innovation lies a subtler transformation: the monopolization of intelligence itself. By 2025, five firms—OpenAI, Google DeepMind, Anthropic, Meta, and Amazon—control over 85 percent of global large-model computing capacity (OECD AI Market Concentration Index, 2025). These entities now function less as technology companies than as privatized knowledge regimes—extracting rents not from production, but from cognition.

AI’s future is not just about who builds the smartest model, but who owns the intellectual scaffolding of thought.

From Innovation to Enclosure

The early internet promised open access to information. The AI era has reversed that ethos, enclosing the cognitive commons within proprietary architectures. Large language models (LLMs) and diffusion systems are trained on public data but privatized in output. The MIT Center for Digital Governance (2025) reports that over 60 percent of the world’s open-access scientific datasets have been commercially repurposed by AI firms without compensation to originators.

This is digital enclosure—knowledge as capital accumulation. Every training token, from academic papers to social media posts, is absorbed into trillion-parameter vaults controlled by corporations that charge subscription fees to access the recombined intelligence of humanity.

AI is no longer a tool; it is a toll.

The Economics of Cognitive Rent

Traditional monopolies control scarcity—oil, land, bandwidth. AI monopolies create scarcity out of abundance. Computation is abundant, but high-quality data and model infrastructure are not. As a result, dominant firms now extract cognitive rents—payments for access to predictive capacity rather than for any tangible good.

The IMF Digital Rentier Economy Report (2025) estimates that 23 percent of global digital GDP is now generated through “intellectual infrastructure leasing”—API access, model licensing, and compute rental. These revenues behave like rent, not profit: passive, recurring, and reinforced by network effects.

Innovation has become a subscription service.

The Barrier of Compute Capital

If data is the fuel of AI, compute is its real estate. Training a frontier model now costs between US$600 million and US$2.1 billion, requiring access to specialized chips, energy, and parallelized infrastructure. The World Bank Computational Economics Study (2025) found that only nine organizations worldwide possess the capacity to train models above 500 billion parameters.

This creates a form of technological feudalism—where smaller firms must rent compute power from the same giants that dominate the market. The result: even innovation becomes vertically indebted.

AI’s meritocracy is a myth built on megawatts.

Intellectual Property in the Age of Prediction

AI models are trained on copyrighted works—books, images, films, and music—yet produce outputs that blur the line between imitation and invention. The legal system has not kept pace. In 2025, the European Court of Justice ruled that AI-generated works cannot qualify for copyright unless “human creative control is demonstrably significant.” Meanwhile, U.S. courts continue to debate whether datasets scraped from the public web violate fair use.

The London School of Economics Intellectual Property Review (2025) notes that AI firms now file an average of 4.3 patents per day related to “model architecture innovations,” effectively patenting the structure of cognition itself.

Knowledge—once cumulative and communal—is now fenced, metered, and monetized.

Algorithmic Colonialism and Global Asymmetry

The new divide in the AI economy is not North vs. South, but owning vs. training. Developing nations supply the raw data—social media content, linguistic corpora, and sensor inputs—while Western corporations monetize predictive intelligence. The United Nations Development Programme (UNDP) Digital Sovereignty Report (2025) found that 92 percent of African and South Asian languages represented in major AI models were collected without consent or compensation.

This replicates the logic of historical colonialism: extraction without equity. Data is the new raw material; the algorithm, the refinery; the model, the empire.

The Myth of Open Source Salvation

Open-source AI projects like Mistral, Falcon, and LLaMA promise democratization. Yet open-source releases often rely on closed infrastructure—NVIDIA GPUs, AWS cloud servers, or proprietary datasets. The OECD Artificial Intelligence Transparency Audit (2025) shows that 78 percent of self-proclaimed open models are either partially closed (restricted weights) or monetized through cloud APIs.

Open AI, in practice, has become Open Access—until Payment Required.

The tension between openness and ownership defines the moral crisis of modern machine intelligence.

Regulatory Capture and the Illusion of Safety

While policymakers debate AI “safety,” regulatory frameworks risk entrenching existing monopolies. The U.S. AI Executive Order (2024) and the EU AI Act (2025) impose compliance costs exceeding US$25 million per model deployment, effectively barring startups and academic labs from participation. The Brookings Institution Technology Policy Study (2025) calls this “compliance capture”—when regulation designed for safety inadvertently consolidates power among incumbents.

The paradox of AI governance is that the more we regulate risk, the more we institutionalize it.

Rethinking Ownership of Intelligence

If intelligence is humanity’s collective output, then its artificial replication should not belong to a few firms. The World Intellectual Property Organization (WIPO) Future of Knowledge Framework (2025) proposes a radical alternative:

1. Global Data Dividend – Redistribute a share of AI firms’ profits to the populations whose data trained their models.

2. Public Compute Infrastructure – Establish national AI supercomputers accessible to academia and SMEs.

3. Commons-Based Licensing – Mandate that publicly funded research cannot be privatized through proprietary model architectures.

4. Transparency Mandates – Require disclosure of training data provenance and weighting.

Such measures would convert AI from a rentier system into a civic one—a shared infrastructure rather than a privatized oracle.

The Political Philosophy of Prediction

AI is not merely a technology but an epistemology—one that transforms knowledge into prediction and prediction into power. When prediction is owned, thought itself becomes property. In that sense, AI’s economic model is not artificial at all—it is the logical conclusion of late capitalism: privatizing the future to sell it back to the present.

The singularity may never arrive, but the rent-seeking singularity is already here—quietly enclosing the mind behind a paywall.

Works Cited

“AI Market Concentration Index.” Organisation for Economic Co-operation and Development (OECD), 2025.

 “Center for Digital Governance Report.” Massachusetts Institute of Technology (MIT), 2025.

 “Digital Rentier Economy Report.” International Monetary Fund (IMF), 2025.

 “Computational Economics Study.” World Bank, 2025.

 “Intellectual Property Review.” London School of Economics (LSE), 2025.

 “Digital Sovereignty Report.” United Nations Development Programme (UNDP), 2025.

 “Artificial Intelligence Transparency Audit.” Organisation for Economic Co-operation and Development (OECD), 2025.

 “Technology Policy Study.” Brookings Institution, 2025.

 “Future of Knowledge Framework.” World Intellectual Property Organization (WIPO), 2025.

 “AI Ethics and Regulation Review.” European Commission, 2025.