AI Agent Market Concentration: The Oligopoly Forming
Three American companies, three Chinese labs, and everyone else is fighting for relevance.
Published: 28 July 2026 Category: AI Industry Analysis Sources: [France competition authority report, industry revenue estimates, July 2026]
The Numbers
The AI agent market is concentrating faster than most observers expected. France's competition authority put the figure at 84% for OpenAI, Google, and Anthropic combined. Add the Chinese labs — Moonshot, Alibaba, Zhipu — and the top six players control perhaps 95% of global AI agent revenue and deployment.
This is not a normal technology market. Normal markets have room for dozens of competitors, niche players, regional variants, open-source alternatives that capture meaningful share. AI agents are not behaving like normal markets. They are behaving like markets with massive fixed costs, strong network effects, and winner-take-most dynamics.
Why Concentration Happens
The reasons are structural. Training frontier models costs hundreds of millions of dollars. Few companies can afford this. The companies that can afford it need revenue to justify the expense, which means building products, which means integration, which means ecosystem lock-in. The result is a self-reinforcing cycle: more investment produces better models, better models attract more users, more users generate more revenue, more revenue funds more investment.
Network effects amplify the cycle. An agent that integrates with more tools is more useful. A more useful agent attracts more developers. More developers build more integrations. The platform with the most integrations becomes the default choice, and default choices are hard to displace.
Data flywheels complete the picture. Agents improve through usage — more interactions produce better training data, which produces better models, which attract more usage. The companies with the most users generate the most data, which produces the best models, which attract the most users. The cycle is not unbreakable, but it is very hard to break from the outside.
The Risks
Concentration produces predictable risks. Higher prices, as vendors exploit market power. Less innovation, as dominant players optimize for defence rather than advancement. Slower adoption of safety measures, as competition prioritises capability over caution. And systemic fragility, as the global AI infrastructure depends on a handful of companies that could fail, merge, or be disrupted.
The national security implications are also significant. If 84% of global AI capability depends on three American companies, then American policy decisions — export controls, sanctions, regulatory changes — affect global AI access. Countries that do not want to depend on American AI companies are investing in domestic alternatives, but the gap is large and growing.
The Verdict
The AI agent oligopoly is not inevitable, but it is the default path. Reversing it would require deliberate intervention: public investment in alternative models, antitrust enforcement against exclusive deals, open-source mandates for publicly funded research, interoperability requirements that reduce ecosystem lock-in.
Some of these interventions are being discussed. Few are being implemented. The political will to challenge the largest technology companies is limited, and the technical complexity of AI markets makes regulation difficult.
The likely outcome is a stable oligopoly: three to four American companies, two to three Chinese labs, and a long tail of niche players and open-source projects. Not a monopoly, but not a competitive market either. A landscape where choice exists in theory but meaningful alternatives are limited in practice.
The AI revolution promised democratisation. It is delivering concentration. Whether that concentration can be reversed — or whether it even should be — is the defining policy question of the next five years.