For more than a year, competition regulators around the globe have been unified in issuing a clarion call that artificial intelligence (AI) risks becoming dominated by just a handful of firms.
International competition agencies issued a joint statement in December warning that AI could entrench “market power” and reduce competition. Andreas Mundt, president of Germany’s Bundeskartellamt, earlier worried that AI would lead to “even deeper concentration of digital markets.”
Meanwhile, the European Commission has been scrutinizing exclusive deals between AI firms and cloud providers, while UK regulators worry that only the richest tech giants could afford to train the most powerful models. Authorities in other countries, like Brazil, have taken note of the imperative to act quickly and decisively to “avoid the mistakes of web 2.0.”
But these fears are increasingly at odds with reality. In recent months, we have seen a surge of alternative AI models—including China’s DeepSeek, the French Mistral AI, and open-weight versions of Meta’s Llama—that challenge the idea that AI is a closed, monopolistic system. The landscape is evolving rapidly, and AI is proving to be one of the most dynamic, competitive industries in tech today.
It is DeepSeek that has of late drawn almost breathless coverage for having been developed outside the traditional Silicon Valley powerhouses. Its emergence shows that AI development is not limited to a few American firms with deep pockets, and may be possible with far lower levels of investment than previously imagined. But in fact, this is not a new story at all. Generally speaking, the breakthroughs we have seen in generative AI have not come from supposed incumbents with “insurmountable data advantages,” but from startups like OpenAI and Midjourney.
For its part, demonstrating that innovation is still fully possible in the European Union, Mistral has raised significant funding and released competitive models that are on par with—or even better than—those from OpenAI and Google DeepMind. And the open-source AI movement more generally is thriving, with thousands of developers collaborating on decentralized AI projects that challenge the idea that only a handful of companies can drive innovation.
Yet regulators continue to treat AI as if it is inevitably a winner-take-all industry. The U.S. Federal Trade Commission (FTC) and U.S. Justice Department (DOJ), for example, have warned that AI firms could use strategic partnerships and data-access agreements to create insurmountable barriers to entry. The European Commission’s investigation into cloud-AI partnerships suggests an assumption that the market is already captured by a few firms. This regulatory lens—borrowed from past and ongoing battles against tech platforms—fails to account for AI’s fluid and rapidly evolving nature.
To be sure, competition concerns in AI should not be ignored entirely. The cost of training frontier models is high (although the example of DeepSeek tentatively suggests that the cost may not as high as we thought). And some AI firms have secured exclusive deals that merit scrutiny.
But assuming that AI is destined for monopoly ignores the diversity of players entering the field. If anything, aggressive intervention at this stage could freeze a market that is still taking shape. The effect could be to discourage new entrants and render AI development riskier and more expensive. This is compounded when intervention is pursued via regulation, rather than the more flexible case-by-case approach of traditional competition law. Ex-ante market-conduct rules often prove both inflexible and difficult to repeal, even when they show themselves to be wrong-headed.
Rather than regulating AI as if it were already a captured market, policymakers should focus on ensuring that competition continues to flourish. Among other things, that could mean fostering international AI development and removing unnecessary regulatory barriers to entry.
AI is emerging as one of the most competitive fields in technology. Regulators should take note before trying to craft solutions for a problem that does not exist and may never arise.
