Where AI Advantage Becomes Market Power
The direct competition question is whether firms that use or develop AI gain a durable position over rivals. The available evidence is more discriminating than a simple yes or no: earlier non-generative AI use is not associated with a uniform rise in markup growth, while GenAI exposure and AI innovation show conditions under which advantages could widen. Comparing these measures carefully helps separate selection, productivity, innovation, and market power rather than treating every sign of concentration as the same phenomenon.
What user evidence does—and does not—show
The OECD's firm-level user evidence comes mainly from official information-and-communications-technology surveys in France and Portugal, linked to financial data. The available survey waves identify whether a firm uses AI, but their timing means they largely capture non-generative systems rather than the post-2022 surge in generative AI. The analysis focuses on firms with at least 10 workers and examines two market-related outcomes: the firm's position in its industry's market-share distribution and its markup, a measure intended to approximate the gap between output price and marginal cost.
Those outcomes answer different questions. Market share describes a firm's relative sales position. A market-share decile indicates where the firm sits within a sector, but not why it sits there. A markup is closer to the ability to price above marginal cost, but its estimation depends on financial data and production assumptions. Neither measure alone is a complete measure of competition, and a high share or markup is not automatically proof of weak rivalry.
The descriptive starting point looks striking. AI-using firms had average market shares 7.5 times those of non-users in France and 3.2 times those of non-users in Portugal. That does not mean AI multiplied the market share of a typical adopting firm by those factors. It compares group averages, and adopting firms were already different: they tended to be larger, more productive, more digitalized, and more likely to have complementary infrastructure. A firm that is already near the top of its sector is also more likely to have the resources and incentive to adopt.
Once those differences are taken into account, the simple story weakens. In regressions controlling for broadband, other digital technologies, productivity, firm size, age, sector, and year, the association between AI use and a higher position in the market-share distribution lost statistical significance in the fuller specifications. In an ordered analysis, AI use was associated with a higher likelihood of being in the top market-share decile in both countries, but that relationship was statistically significant only for France. When the outcome was movement through the distribution over a five-year window, AI adopters were generally no more likely than non-adopters to move up.
This is evidence against a uniform early market-share acceleration, not evidence that AI has no competitive effect. The top of a distribution may respond differently from the middle. A binary use indicator may hide whether the system is a recommendation tool, a predictive process, or an integrated capability. Adoption may also matter only after a period of learning and organizational investment. The available window is short and the exact timing of adoption is not always known, so a delayed effect could remain outside the observation.
The markup result is more stable. From 2011 to 2022, markup paths for firms classified by their 2020 AI-use status did not significantly diverge within either France or Portugal.