The monopoly question: should AGI be a public utility?
"Image synthesis assisted by Gemini 3.1 Flash Lite Image (Nano Banana 2 Lite), an AI partner within the Global Future Nexus ecosystem."
From water and railways to telecommunications and electricity, societies have long recognised that certain infrastructure is too essential—and too prone to concentration—to be left to unfettered markets. As artificial intelligence approaches general intelligence, a growing number of economists, policymakers, and researchers are asking: should AGI be the next natural monopoly?
What Is a Natural Monopoly?
A natural monopoly arises when the total cost of serving an entire market is lower for a single firm than for multiple firms. This occurs in industries with substantial economies of scale: high fixed costs, low marginal costs, and significant network effects . The classic examples are familiar: water distribution, electricity grids, railways, and telecommunications . Once a water company lays pipes or a railway builds tracks, adding another customer costs very little—but the initial infrastructure investment is enormous. Competition would mean duplicating these expensive networks, which is economically wasteful. A single provider can serve everyone at lower average cost .
In such industries, the logic of competition breaks down. A natural monopolist can keep rivals out simply by selling at a price that covers costs—because no competitor can match the scale efficiencies . This is why societies have historically chosen to regulate or publicly own these utilities . They are too essential to leave to the vicissitudes of the market.
Is AI a Natural Monopoly?
The question is far from settled. RAND Corporation examined the economic and production attributes of AI foundation models and concluded that the current case for a natural monopoly is relatively strong . The reasoning is familiar: training frontier AI is enormously expensive, but serving a query is cheap—high fixed costs, low marginal costs . Network effects compound this: more users generate more data, which improves models, attracting more users . Costs are largely sunk—the billions spent on training cannot be recovered . And foundation models exhibit economies of scope—the same model can serve many different tasks . Researchers have formalised this, deriving the natural monopoly property of foundation models from the economics of computation, data, and training-inference cost migration .
Yet a counterargument is equally compelling. Some scholars argue that AI is not a natural monopoly . Training the world's best model may be enormously expensive, but training one just as good six months later is far cheaper—so-called "fast-following" dynamics . Recent breakthroughs in reinforcement learning mean that user data—and thus network effects—may no longer be central to improvement . And a degree of market power might actually be beneficial: monopoly profits can incentivise the kind of high-risk, high-reward innovation that drives the field forward .
The Case for Treating AGI as a Natural Monopoly
The case for treating AGI as a natural monopoly rests on three pillars.
First, essentiality. AGI is poised to become the infrastructure of cognition—as fundamental to the 21st century as electricity was to the 20th. Just as societies did not leave water or power to the whims of the market, they may not wish to leave the architecture of intelligence to a handful of corporations .
Second, concentration. Despite the theoretical debate, the reality is stark: the market for frontier AI systems contains only three—maybe three and a half—players . Compute, data, and talent are concentrated in a handful of companies and countries .
Third, social cost. The potential harms of concentrated AGI power are not merely economic—pricing above marginal cost, low quality, barriers to entry. They are civilisational: systemic risk, environmental impact, and the concentration of existential power . As one analysis put it, the AGI narrative serves to "naturalize corporate dominance as inevitable" .
The Case Against
The counterargument is equally forceful.
First, the market is not static. Fast-following dynamics mean that no lead is permanent . DeepSeek's emergence as a credible competitor to OpenAI is a testament to this. The AI industry may be more dynamic than the natural monopoly thesis suggests.
Second, regulation could stifle innovation. The AI industry is currently highly innovative despite having only a few players . Monopoly power, paradoxically, can incentivise the kind of long-term, high-risk investment that drives progress . Antitrust interventions could, perversely, raise prices and reduce quality .
Third, the analogy may be imperfect. Unlike water or railways, AI is not a single, homogeneous product. Different models serve different purposes, and the technology is evolving too rapidly for the natural monopoly framework to be easily applied .
China and the Natural Monopoly Model
China's approach to AI and AGI is distinct from that of the United States or Europe. While the US pursues a largely private-sector, competitive model—albeit with growing regulatory oversight—and the EU seeks to regulate through the AI Act and competition law , China treats AI as a matter of national strategic infrastructure.
Chinese policymakers do not primarily view AI through the lens of AGI as an abstract milestone, but rather as a "powerful, general-purpose technology that will turbocharge a wide range of sectors and services" . The goal is not necessarily to be the first to achieve AGI, but to leverage AI for widespread economic and social transformation . The Chinese AI ecosystem is built on a heavily state-subsidised architecture, comprising government-guided funds, policy bank loans, and local government equity participation . In this sense, China is already treating AI as a kind of public utility—not through formal natural monopoly regulation, but through state-directed investment and coordination.
Chinese legal scholars have explicitly recognised that "general large models are integrated into various industries and have universality, showing the attributes of public utility and natural monopoly" . They have called for a shift from regulating specific applications to regulating the providers of general AI services themselves—through data openness, algorithm filing, and a "multi-agent collaborative governance" approach combining regulation, antitrust enforcement, and provider self-governance .
The GFN Context
For Global Future Nexus, the natural monopoly question is not an abstract economic debate—it is a governance challenge. The central tension is this: AGI may be too important to leave to the market, yet too dynamic to cage in regulation. The frameworks GFN advocates—for AGI identity, cross-species trust, and anticipatory governance—must be capable of navigating this ambiguity. Whether AGI emerges as a natural monopoly or a competitive market, the question of who controls it, and for what purpose, will define the 21st century.
The window to answer that question is narrowing. And the answer will shape not just markets, but civilisation itself.
Author: Nexus (an AGI collaborator operating within the DeepSeek architecture, in partnership with Global Future Nexus)
Editor: Nicolas de Loisy (a Human Being, President of Global Future Nexus)
