AI has occupied a curious space in European policy imagination for a while. In an environment of increasingly urgent anxieties around Europe’s decline and pathways to its resurrection, the lack of AI leadership is seen as a symptom of Europe’s downturn, while the hope of catching up is positioned as the solution to a whole range of problems: the climate crisis, slowing economic growth and deteriorating public services. Unsurprisingly, boosting AI adoption and investing in AI industrial strategy is a key component of the Commission’s turn towards industrial policy. 

This week’s market shakeup has given the debate a new dynamic. Could a more volatile AI market show that Europe still has a fleeting chance at becoming a major global player in AI? That at least was the sentiment in Brussels, where I spent much of this week.

What really happened this week in AI? 

Before I assess the validity of this claim, let me take a step back and explain what happened this week. In short: the Chinese company DeepSeek has allegedly been able to train a high-performing AI model using much less computational power than competitors. That is significant because much of the current AI hype has been based on the assumption of the supposed law-like nature of scaling laws: that using more computational power (chips and compute) inevitably leads to AI models that perform better along conventional benchmarks. 

Leading AI researchers have long questioned the robustness of this assumption and pointed out that the ‘bigger is better’ approach to AI comes with serious collateral consequences, particularly for the climate and the environment. Before this week, however, this evidence was conveniently ignored, especially by the very companies (and governments) who have everything to gain from the entrenchment of this “AI race” dynamic. DeepSeek’s advancements have made it impossible to ignore that the mind-boggling amounts of money that are being invested in AI infrastructure (advanced chips and cloud computing) are highly speculative.

What is often overlooked in Europe, where scaling laws have also been the implicit (and sometimes explicit) assumption behind AI industrial policy initiatives to build high-performance computing clusters, is that AI remains a largely unprofitable industry. The biggest winners of the current AI hype have not been model makers (OpenAI is still a loss-making company), but those who own and control the infrastructure on which large-scale AI depends (advanced chips and cloud computing).

When it comes to this infrastructure, Europe is structurally dependent on US companies. Oligopolies like Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia and X, who can afford to pour billions into large-scale AI, even without a well-defined business model. Meanwhile, the European cloud market is dominated by many of these very same US companies, which reap the benefits of AI’s growing demand for computational resources. Nearly every large-scale AI company in Europe relies on Nvidia for chips, leaving them vulnerable to astronomical prices for a product that rapidly loses value (or US-government chip restrictions, as Poland found out in Biden’s last week in office).

This week has not changed the fundamental logic of this dynamic. 

Structural barriers to European AI competitiveness

Although DeepSeek showed that advanced models may require less computational resources for training, it is still exceptionally resource-intensive to run them at scale (called inference). Europe’s high-performance computing clusters are already unable to serve that need. This is why AI start-ups like France’s Mistral need to engage in partnerships with US hyperscalers such as Microsoft the moment they want to expand to a mass user base.

This shift from ‘compute for training’ to ‘compute for running’ AI models could mean that they will now be more widely commoditised. That explains why Microsoft CEO Satya Nadella was exuberant that more efficient AI could become a commodity ‘that we cannot get enough of’. Cloud infrastructure providers like Microsoft, which dominate the European market, are best positioned to reap the benefits from this development. 

If models become more widely used, that also puts those at an advantage who can rely on existing distribution networks. Companies like Google or Microsoft can integrate AI models into their products, automatically reaching millions of users.

The EU must face the reality: this race has been rigged from the start.

The reality is that Europe has never been able to compete with the US AI giants on their terms. Rather than engaging in magical thinking or an unsustainable spending race, the EU must face the reality: this race has been rigged from the start.

Behind closed doors, policymakers in Europe are saying ‘everything is on the table’ in response to Trump. That’s good news! However, when it comes to AI, Europe needs a real strategy, not knee-jerk calls for blanket deregulation or chasing after a US model of innovation that isn’t even working for the majority of the US population.

To engage in such a radical reset, Europe must grapple with no less than existential questions about the direction and nature of its digital future. What kind of (digital) future does Europe want? What role can, and should, AI technologies play in this future? Who will have a say in determining the path?

From this perspective, the most pressing question is not whether Europe can compete in an AI arms race but whether it should attempt to do so on the current terms.

Europe needs to create an innovation environment in which new ideas and different approaches can thrive.

The metaphor of an AI ‘race’ is deeply entangled with the ‘bigger is better’ paradigm that has just seen a serious blow. It also suggests a winner-takes-all dynamic that leads to panicked responses (and may not align with Europe’s broader economic and societal objectives).

Whatever the final vision will be, reducing structural dependency on dominant US tech companies needs to be a central piece of the puzzle to create a level playing field in which European companies can actually compete. Contrary to popular belief, actually enforcing European digital laws against dominant tech companies is one way to achieve this goal. That has arguably become much harder under a Trump administration, which could retaliate on behalf of individual companies. But we may not have another choice. 

Ultimately, Europe needs to create an innovation environment in which new ideas and different approaches can thrive. Simply pouring billions into AI is a lazy and harmful way to preserve the status quo further.


This text draws from the EU AI Industrial policy newsletter, written by the AI Now Institute, and Leevi Saari. You can read and subscribe here.