Hypothesis: National Monopoly of AI

Artificial intelligence is usually discussed as a technology race: who has the best model, the most GPUs, the most talent. In China, it is increasingly useful to view large language models (LLMs) through a different lens: political economy.

A useful analogy is not the smartphone industry or the cloud market. It is the state monopoly system that has governed products like tobacco and salt, not because the state must train every model itself, but because it can control the gateways that decide which models may legally reach the public.

When the gate is the power, the market can look competitive while remaining fundamentally monopolized.

The Core Logic: Gateway Control, Not Production Control

A state monopoly is often misunderstood as the state ‘producing everything’. In practice, monopoly power is usually exercised by controlling:

• Licenses
• Distribution channels
• Pricing rules
• Enforcement against unlicensed supply

China’s tobacco system is the clearest example. The State Tobacco Monopoly Administration does not run every cigarette factory, but it sets the boundaries: who can produce, who can sell, where products can circulate, and how the system extracts revenue.

Salt historically followed the same pattern: the state did not need to crystallize every grain of sodium chloride. What mattered was controlling the trade routes and the right to operate.

LLM regulation maps neatly onto this logic. China does not need to build one single ‘national model’ to monopolize AI. It only needs to decide, through filing, safety reviews, and compliance requirements, which models are permitted to serve the public. A model can be technically superior and still be commercially nonexistent if it cannot legally reach users.

In that sense, the checkpoint becomes the asset.

Why ‘Harm’ Narratives Strengthen Monopoly Power

Monopolies are easier to justify when a product is framed as dangerous.

Tobacco is harmful to health, yet it is not merely regulated. It is monopolized. The system performs two functions at once:

1. Extracts stable tax revenue
2. Argues that unified management prevents a worse, uncontrolled black market

Salt monopolies historically used similar language: fighting smuggling, ensuring supply, maintaining standards (such as iodization).

AI now has a ready-made ‘harm narrative’. Deepfakes, scams, disinformation, and ideological risks are not theoretical problems. They are real, legible threats to governance. And once a technology is widely labeled as potentially harmful, the state gains moral and administrative justification to:

• raise compliance thresholds
• limit who can operate
• enforce centralized standards
• punish unlicensed distribution

The stronger the danger story, the easier it is to defend gatekeeping.

This is not simply censorship as a technical feature. It is a regulatory rationale that naturally produces monopoly structure.

Rent, Not Innovation, Is the Prize

One reason tobacco is so durable as a system is that it produces enormous rents. The competition that does exist is not ‘free market competition’. It is competition within a bounded framework.

Brands compete. Regional players fight for share. Yet no private actor can ‘win’ into an unregulated position, because the system is designed so that the most important leverage point, legality of distribution, is not up for grabs.

That is what a mature monopoly looks like: lively surface-level rivalry under immovable institutional constraints.

China’s AI market already resembles this. Baidu, Alibaba, Huawei, Tencent, and other players can compete fiercely on model capability, infrastructure, and developer ecosystems. But they compete inside an environment where the ability to serve the public is conditional and revocable.

In other words, the arms race is real, but the borders of the battlefield are set from above.

No one can innovate their way out of compliance.

Competition as Discipline: The Monopoly That Still Looks Like a Market

There is a common mistake in analyzing monopoly: assuming it eliminates competition. Some monopolies do the opposite. They use competition as an organizing tool.

Bounded competition can be useful for the state because it:

• drives continuous investment without surrendering control
• prevents a single private winner from becoming politically unmanageable
• creates multiple ‘national champions’ whose incentives remain aligned with licensing systems

This is how you can get an industry that feels dynamic while still functioning like a monopoly.

From the outside, it looks like a market. From the inside, it behaves like a regulated utility with a small number of approved operators.

How This Might Play Out

If the ‘national monopoly’ framing is correct, the next phase is not primarily about who builds the smartest model. It is about who owns the infrastructure and chokepoints that make models usable at scale.

Filing becomes the operating license

The model filing and evaluation framework becomes what a tobacco license is to cigarettes: the legal prerequisite to exist in public commerce. This does not eliminate open-source models or research models, but it sharply differentiates:

• what is allowed in labs
• what can be deployed to the public
• what can be sold as a service

Distribution channels become more important than capability

In a pure technology market, the best product wins distribution. In a gatekept market, distribution is itself the product.

The key asset becomes access to cloud deployment, enterprise procurement pipelines, public-sector clients, compliant content safety systems, and approved consumer-facing apps.

A smaller model with the right approvals can beat a stronger model that lacks a legal route to users.

The market produces ‘regulated innovation’

Innovation still happens, but it becomes shaped by what is legible to regulators and acceptable to licensing systems. The industry optimizes for controllability, auditability, standardized safety processes, and predictable compliance outcomes.

This changes what ‘winning’ means. The winning company is not necessarily the one with the most impressive benchmark numbers. It is the one that turns compliance into scale.

What Has Already Happened: Telecoms Smell the Chokepoint

A revealing signal is coming from the telecom sector.

On March 25, 2026, China Telecom (one of the four state-owned telecom operators in China) released its 2025 annual performance report, in which the term ‘Token’ reportedly appeared for the first time, mentioned several times across the document. In the earnings briefing, management emphasized token services as a future business focus, including strengthening proprietary token products, expanding ecosystem tokens, and exploring internationalized token operations.

Even if we set aside the specifics of that statement, the direction is telling: telecom operators naturally sit near the chokepoints. They are positioned to package access, billing, identity, routing, and enterprise distribution: the very places where ‘gateway control’ becomes monetizable.

If LLMs become a regulated monopoly-like product, the most valuable players may be those who can industrialize access and compliance, not only those who can train the largest model.

The telecoms are not trying to become the best AI lab. They are trying to become part of the toll booth.

A More Precise Claim (and a Necessary Counterargument)

To be clear, calling this a ‘national monopoly’ does not mean China will ban private AI companies or nationalize every model.

It means something narrower and more structural:

• public-facing LLM services may increasingly resemble a licensed industry
• the right to distribute becomes scarcer than the ability to build
• rents concentrate around gatekeeping institutions

Counterargument: ‘Isn’t this just normal regulation?’

A fair objection is that every country is regulating AI now. So why call China’s approach a monopoly logic rather than simply safety governance?

Because there is a difference between regulating harms while leaving market entry broadly open, and regulating via a licensing architecture that limits legal distribution to a small set of approved actors.

In the latter case, regulation does not merely constrain behavior. It shapes the market into a structurally concentrated form, even if the state never declares a monopoly in name.

The question worth watching is not whether the monopoly label is formally attached. It is whether the structural outcomes (concentrated access, rent extraction at the gateway, bounded competition) arrive regardless of what the policy documents say.

EOF