The data refuses a simple story

AI is supposed to do two opposite things to competition. It gives a small company tools it could not previously afford. It also rewards the companies that already have money, data, talent and distribution.

A new OECD study finds evidence for both.

Competition in the age of AI, released on Thursday, combines official business surveys from France and Portugal with employment data, global patent records and information on AI start-ups. The report runs to 75 pages.

Its first useful choice is to separate companies that use AI from companies that develop it. It also separates generative AI from older forms of prediction, classification and automation.

Without those distinctions, almost any claim about 'AI and competition' becomes too broad to test.

Older AI adoption did not raise markups in this sample

The cleanest result comes from France and Portugal between 2011 and 2022, before generative AI became a common business tool.

Companies using non-generative AI were generally larger and more productive than non-users. Yet their markups — the gap between a firm's price and its estimated marginal cost — were not substantially higher.

The OECD therefore finds no systematic link between adopting that kind of AI and greater market power in this sample.

That is not proof that AI adoption never strengthens a company. It covers two countries, a defined period and mostly pre-generative systems. Markups are also a proxy, not a complete picture of power over suppliers, workers or customers.

Still, it is a useful correction to the idea that using AI automatically lets a firm charge more.

Generative AI may lower one barrier and raise another

For generative AI, the report has less direct adoption history. It uses the mix of occupations inside Portuguese firms to estimate how exposed each company could be to the technology.

That measure suggests smaller and younger firms may have room to enter markets or challenge incumbents. Access to a general-purpose model can replace some software, research and production work that once required a larger team.

But the same analysis points to an advantage for firms with more human capital and complementary assets. Buying access to a model is easy. Reorganising work around it, checking the output and connecting it to customers is not.

The implication is awkward but familiar. A cheaper tool can widen access while the largest gains still flow to companies that know how to absorb it.

Because exposure is a proxy for potential adoption, this part of the study is descriptive. It does not yet measure the long-term effect of everyday generative AI use on market share.

Patents and acquisitions lean toward concentration

The development side looks less even.

Using global patent data, the OECD finds a positive correlation between concentration in AI invention and concentration in sales. Its measure looks at the share of AI patents held by the four largest patentees in a year.

In information and communications technology, firms with an AI-related patent also show faster markup growth. The same pattern is not clear in sectors where AI is mainly an input rather than the product being sold.

Start-ups keep entering the market and generative AI attracts substantial venture capital. At the same time, large incumbents frequently acquire those firms. An acquisition can fund founders and move technology into a larger product. It can also remove a future rival.

The report raises that risk without claiming every deal is a so-called killer acquisition. Intent and the future that did not happen are hard to observe in a dataset.

What the study shows — and what it cannot

Confirmed by the study: non-generative AI use was not associated with substantially higher markups in the French and Portuguese sample. Patent concentration, sales concentration and ICT markup growth move together in some of the wider data.

The OECD's interpretation is that AI markets are dynamic but uneven. It sees opportunities for entry alongside advantages that may reinforce established firms, and recommends continued monitoring.

Still open is causation. A strong company may patent more AI because it already has market power. A patent may help it gain power. Both can be true, and this study does not fully separate them.

The timing matters too. The OECD says 20.2% of enterprises across its members used AI in 2025. Generative AI diffusion is young, while the firm-level adoption result covers 2011 to 2022.

The honest conclusion is not that AI will break monopolies or create them. It is that access, skills, ownership and acquisitions pull in different directions. Competition authorities will need better data before those early lines harden into a market structure.

Sources

  1. OECD — Competition in the age of AIPrimary publication page for OECD Artificial Intelligence Papers No. 64, released 30 July 2026.
  2. OECD — Full 75-page competition studyPrimary research report used to verify the datasets, time periods, empirical findings, caveats and policy interpretation.
  3. OECD — Competition topic and related AI workPrimary OECD policy hub used to confirm the paper's place in the organisation's current competition programme and related market-monitoring work.