AI Is Becoming Critical Infrastructure, Yet Ownership Remains Concentrated

AI has crossed from a fast-growing technology into a dependency system woven through public life. It already affects how governments deliver services, how healthcare and finance operate, and how information circulates. 

Despite that, the foundations of AI remain concentrated in a small group of firms controlling compute, chips, cloud access, models, and distribution. 

This piece explores why regulation alone cannot fix a system built on concentrated ownership, what is at stake economically and politically, and why decentralized AI is gaining attention as a way to improve openness, verification, and resilience.

Societies have long treated essential systems with special care:

• Electricity keeps homes, hospitals, transport, and industry running. 
• The internet supports communication, trade, media, and public debate. 
• Cloud computing became the operating base for business, government software, and everyday digital life.

Believe it or not, AI now also belongs in this category. It influences how people work, what information reaches them, how services are delivered, and how decisions are made across the economy.

Even companies building in decentralized AI speak in terms once reserved for public systems. 0G describes its mission as making AI “a public good” through decentralized foundations rather than closed control.

In short, AI is becoming a system other systems rely on.

Ownership Has Not Kept Pace With Dependence

What feels strange is how normal concentrated ownership still seems. Few people would accept a world in which a tiny group of private firms owned the power grid and decided who could access electricity, at what price, and under what terms. 

Fewer still would feel comfortable if those same firms also controlled the inspection tools, the metering system, and the data on how every household used power. 

Yet with AI, public debate often treats concentrated control over models, compute, data access, and distribution as if it were simply the expected outcome of innovation. 

AI already influences productivity, information, and decision-making on a societal level. Ownership, however, remains concentrated in a narrow corporate circle. 

As dependence grows, so does the importance of asking who controls the base on which everyone else builds.

AI Now Underpins Public Systems

A simple test helps here. Can major institutions still function well without AI? Reliance is already spreading across government, healthcare, finance, and media.

The OECD’s 2025 review of AI in government offers a strong example. Governments are already using AI to automate and tailor public services, improve decision-making, detect fraud, and support forecasting. 

The report found 57% of government AI use cases in its review were tied to streamlining or tailoring services, while 45% supported decision-making, sense-making, or forecasting. 

Healthcare tells a similar story. The OECD has written about the growing role of digital health in care delivery, including AI-based systems.

AI can affect triage, diagnostics support, patient management, and administrative decisions within health systems. Once technology influences care delivery at this level, public consequences follow.

Finance adds another dimension. The Bank of England (BoE) has treated AI as relevant to the resilience of the financial system itself. That is an important distinction. AI is being discussed as part of systemic stability, not simply as a productivity tool for internal teams.

The information environment is changing too. UNESCO chose AI’s impact on press freedom, media freedom, and the free flow of information as the theme for World Press Freedom Day 2025. 

AI is now a technology embedded in public services, health systems, financial systems, and the information sphere, where dependence is forming. 

Concentrated Control Creates Structural Risk

The main danger in AI comes from ownership and control. A small number of firms hold strong positions across model training, advanced chips, large-scale cloud compute, and commercial distribution.

The OECD’s work on competition in AI points to concentration across the supply chain, from advanced lithography and chip fabrication to GPUs and cloud services. 

In the GPU market used for AI tasks in data centers, estimates place Nvidia above 80% market share. Cloud access is also dominated by a small group of providers with global reach.

When so much capability depends on a narrow set of vendors, leverage over who can build, train, deploy, and compete becomes concentrated too.

Compute is especially important. 

Training advanced models requires enormous processing power, large capital expenditure, and access to specialized hardware.

Firms with deep capital reserves and preferred access to chips and cloud capacity hold a major advantage over smaller builders, public institutions, independent labs, and open ecosystems.

Distribution creates another pressure point. Even strong models still need routes to users, developers, and enterprise clients. 

Large incumbents often control the interfaces, hosting environments, and partnership networks through which AI reaches the market. That gives them influence over both development and adoption.

Opacity adds another problem. NIST’s AI Risk Management Framework treats transparency, explainability, accountability, resilience, and reliability as core traits of trustworthy AI.

Those qualities become harder to secure when high-impact systems remain closed and difficult to audit from the outside. 

The Federal Trade Commission’s (FTC) January 2025 report on cloud providers and AI developers warned that these partnerships can strengthen incumbent power over compute, talent, distribution, and sensitive business information. Concentration in AI can therefore reinforce itself over time.

Regulation Cannot Fix Ownership Concentration

Regulation can raise standards, increase reporting obligations, and limit reckless deployment. It can also improve accountability when harms appear.

Still, regulation alone cannot solve a problem rooted in concentrated ownership of core inputs. Most regulatory tools focus on products, conduct, or consumer outcomes. They do far less to change who controls compute, cloud access, model hosting, or distribution.

That is the limit of much of the current debate. A closed provider can publish principles and satisfy compliance requirements while still preserving strong control over the base on which others depend. A market can become more supervised without becoming more open.

The FTC’s 2025 report adds another piece by showing how cloud and AI partnerships can deepen dependence through privileged access, switching costs, and information asymmetries. 

Regulation may restrain abuse at important points, but ownership remains where it was.

Economic Stakes Extend Beyond Technology

The stakes here are economic, civic, and geopolitical.

The firms controlling AI increasingly influence who captures productivity gains, who sets terms of access, and who gets to build on top of the next generation of digital tools.

That affects startups trying to enter the field, enterprises choosing dependencies, and governments deciding whether to rely on outside vendors for important functions.

There are also consequences for information control. UNESCO’s focus on AI and press freedom points to the growing influence of these systems over media production and distribution.

When models and platforms help organize what people see, summarize complex topics, recommend content, or assist with editorial work, ownership starts to affect the information environment itself.

Financial stability belongs in the same discussion. The BoE has warned that heavy reliance on a small number of AI providers could create systemic vulnerabilities for the financial sector. Dependence on a narrow supplier base can create common points of failure and concentrated operational risk.

Security comes into view as well. Heavy concentration can make key parts of the AI stack attractive targets for state pressure, commercial coercion, or strategic failure. When essential capability sits behind a narrow set of private gateways, power over access becomes unusually concentrated.

Decentralized AI Offers an Alternative Model

If ownership concentration sits at the heart of the issue, the answer has to address ownership, access, and verification at the system level. 

Decentralized AI offers one route. Its value lies in opening core parts of the stack to more participants and making them easier to inspect.

Under this model:

• AI outputs can be verified rather than accepted on trust alone. 
• Compute can come from a wider pool of contributors. 
• Data input can be opened beyond a single corporate gate. 
• Model execution and validation can also become more transparent, improving auditability and reducing reliance on closed systems.

One path leads to closed systems controlled by a small group of firms. Another supports open participation, stronger verification, and more transparent rules for the systems people increasingly depend on.

The internet already offered a lesson in what happens when an open technological environment consolidates around a few dominant platforms. AI now presents a similar decision, with deeper implications for decision-making, productivity, and information. 

This debate needs to happen before concentration becomes too entrenched to challenge.