DeepSeek and Trump 2.0: Can Europe Keep Up in AI?

Spoiler: No, this does not spell the end of concentrated market power in AI. Even if training models becomes more efficient, existential questions about the direction of AI remain.

As we write this, the DeepSeek shock is still reverberating across financial markets and the minds of policymakers. By now, you have probably read 100 hot takes on what the Chinese model’s success means for AI.

We have been thinking about a different question: what does the AI market cooldown mean for Europe’s AI ambitions—especially in light of Trump’s first two weeks in office?

Also in this issue: AI investments in EU member states, UK replaces Competition and Markets Authority chair with ex-Amazon boss, and a first glance at the EU’s Competitiveness Compass.

Next week, we’ll cover the AI Action Summit in Paris and France’s AI ecosystem.

Why has DeepSeek sent a shockwave through markets? And why now?

Leading AI researchers have long questioned the supposedly robust and law-like relationship between using more computation for training and better model performance (along conventional benchmarks). Before this week, however, these widely known concerns about scaling laws were studiously evaded by the largest companies that have everything to gain from the entrenchment of a bigger-is-better paradigm (and by governments that are caught up in the mythologies and pervasive “AI race” dynamics).

DeepSeek’s advancements have cut through this manufactured consensus—making it more common sense that the eye-watering sums being invested in AI infrastructure—chips and cloud—are still a highly speculative and inherently vulnerable bet. As governments (especially Trump 2.0) position themselves as “all-in” on supporting AI infrastructure build-out, this warning of the dangers of too-big-to-fail dynamics for AI could not be better timed.

It’s important to add that the mismatch between costs and profits is not the only reason why researchers have questioned the “bigger is better” approach to AI. This approach also comes with collateral consequences, particularly for the environment. We’ve previously argued that large-scale AI’s current trajectory stands in irreconcilable tension with Europe’s environmental and green transition goals.

Also, DeepSeek still requires large amounts of computing power

While models like DeepSeek allegedly need less compute for training, they still require significant computational resources for inference (i.e., running a model at scale). We expect that this will shift the debate from compute for developing large-scale AI models to compute for running large-scale AI models.

This also explains why Microsoft CEO Satya Nadella was exuberant that more efficient AI could become a commodity “that we cannot get enough of,” as the dominant cloud providers are best positioned to reap the rewards of this commoditization—as are those serving the inputs and the cloud infrastructure on which AI is developed and shared with end users.

Is the EU now able to “compete” in an AI arms race? Four reasons why it’s not

We have heard from several EU officials (and opinion pieces) that the DeepSeek moment has been interpreted as a sign that Europe is now finally able to compete. The AI market seems to be in flux, and that is seen as evidence that not all is lost.

Let’s unpack these claims one by one:

1. Big tech firms still enjoy a competitive advantage in AI

It is too early to say what this week’s news means for model developers. Theoretically, companies that have created moats around closed models or privileged access to compute could potentially lose their competitive edge. More high-performing “open” models could also mean that European Mistral sees its moats and margins narrowing. We can’t stress enough that it’s too early to make predictions like these.

What we do know is this: if the DeepSeek moment leads to greater commoditization, model providers that are deeply integrated into the distribution networks of big tech firms will still enjoy a significant competitive advantage. In other words, the market remains fundamentally skewed in favor of dominant U.S. players.

2. Europe has never been able to compete with U.S. AI giants on their terms

Europe’s AI industrial policy proposals have been steered by the assumption of scaling laws: that more compute leads to better models. Case in point: the EU just committed €1.5 billion to develop high-performance computing capacity for training new models—a drop in the ocean compared to the investments made by hyperscalers in recent years.

In reality, Europe has never been able to compete with U.S. AI giants on their terms. Yet, this realization still hasn’t fully sunk into EU policy circles, where Trump touting $500 billion for AI infrastructure has been widely interpreted as a sign that Europe needs to do the same. Even if this week’s news sparks a rethink of the implicit assumption of scaling laws, Europe still needs a strategy that lays out how, where, and whether it invests in AI.

3. Europe is still dependent on U.S. cloud providers

Since models like DeepSeek’s still need large amounts of computing power, Europe’s structural dependence on U.S. cloud providers will continue to bite. That’s because Europe’s existing high-performance computing capacity for model development is unlikely to serve the need for inference (running models after they have been trained).

Mistral AI is the perfect example to illustrate Europe’s persistent structural dependency. Partly trained on European high-performance computing infrastructure and alternative cloud computing providers, the company ultimately had to enter a deal with Microsoft to serve its customers at scale.

As European AI companies attempt to scale to a global customer-facing market, they are pulled into the orbit of the hyperscalers. To reach a sufficient customer base for a functioning business model and gain access to the computation needed to run large-scale AI inference at scale, the path to profitability goes through Big Tech. This explains why European AI companies like Mistral form partnerships with Microsoft, or why Finnish Silo AI offers its Viking models in Google Cloud.

4. Europe is still dependent on the U.S. for chips

While we cautiously anticipate that chip companies focusing on inference (e.g., SambaNova, Groq, and partly AMD) may find an opening to challenge the fiefdoms of Nvidia, the hardware dependency of the European Union remains significant.

The AI news cycle is moving so quickly that you may have already forgotten that Biden introduced export controls on advanced AI chips that affected several European countries. Commissioners Henna Virkkunen and Maroš Šefčovič expressed concern, and Poland was particularly upset, as this may pose a threat to its ambitions for military expansion. There’s a possibility that Trump, who called DeepSeek “a wake-up call for our industries that we need to be laser focused on competing to win,” may further expand chip export control rules.

All taken together, market concentration and the EU’s dependency on U.S. companies remain a major reason why European AI companies are structurally disadvantaged in the current race for large-scale AI.

Should Europe try to compete in an AI arms race?

The much more interesting question has always been whether Europe should, in fact, accept the premises of an AI arms race in the first place.

We’ve already mentioned the collateral consequences, especially for the environment and the public, who are on the receiving end of poorly designed (and often unfit-for-purpose) applications of AI. The handling of personal data (and the persistent and inherent security vulnerabilities in AI) is equally at odds with even baseline European regulations. Incidentally, the Italian Data Protection Authority has already sent DeepSeek a request for information on how it handles users’ personal data, and the AI Office is reportedly also monitoring the company. These concerns remain valid even if more efficiently trained models are used more widely.

The metaphor of a race suggests a winner-takes-all logic and the need to act as soon as possible. It is true that Europe is under immense pressure to boost its economy, particularly in light of looming U.S. tariffs. Our main concern with the EU’s AI industrial policy proposals so far is that they lack a comprehensive strategy. What kind of digital future does Europe want? What role can and should AI play in this future? And what’s the strategy that will get us there?

Even if training models have become much more efficient, these existential questions about the direction—not only the pace—of innovation remain. This is particularly true when it comes to the hope that investments in AI infrastructure and boosting AI adoption will lead to exactly the kind of economic growth and job creation that Europe wants and needs.

What’s next for AI markets?

Much will be written about DeepSeek in the upcoming weeks and months. It would be tempting to over-index on this change. It is likely that markets will correct, as more widespread diffusion of AI models could compensate for the decrease in computational demand for training. In its wake, however, the retreating waves will reveal shipwrecks.

“Tactical silence” after Trump’s first weeks in office

It’s hard to believe that it’s not even been two weeks.

Here are just a few signals and key announcements from across the Atlantic that will have effects on AI in Europe:

• The tech industry’s widespread embrace of Trump, signaling a merging of corporate power with state authority.

• Deregulation of AI, including Trump’s repeal of Biden’s Executive Order and his passing of an Executive Order on “Removing Barriers to American Leadership in Artificial Intelligence.”

• A clear signal to prioritize U.S. AI dominance, with the announcement of a $500 billion “Stargate” partnership and an acceleration of the energy infrastructure platform for AI expansionism.

What does this mean for Europe?

Threats to European sovereignty (and economy)

The biggest threat of the merging of corporate and state power remains the real risk that the U.S. administration could retaliate against the EU for enforcing laws against tech companies that are on good terms with Trump. This is about more than just enforcement of the DSA against X. Trump has described the EU’s competition policy as a “tariff,” and it’s no stretch to think he may react similarly if the EU were to get serious about industrial strategy.

We’ve seen early signs that the Commission may relent, at least on enforcement (yesterday, 40+ civil society organizations, across the EU and the US asked the Commission to stand up to Trump). Behind closed doors, however, we’re now hearing that the tone has changed significantly. There’s now talk of “tactical silence” and a commitment to not back down. We will see if that’s true over the coming weeks.

Meanwhile, European start-ups are expressing worries over Trump’s tariff rollout, prompting discussions about relocating to the U.S.

Deregulatory winds blow over to Europe

Changing winds in the U.S. will give further momentum to deregulatory voices at home. Those who were always opposed to regulation like the GDPR or the AI Act often assume that these rules are to blame for Europe’s lack of competitiveness.

There’s an obvious counterargument: from data to cloud, Europe has failed to effectively rein in dominant tech companies and now finds itself in a condition of structural dependence. Under such conditions, blanket deregulation would primarily benefit those who are already dominant.

Behind closed doors, we’re hearing that “everything is on the table.” The key question is whether that means tackling structural dependency at its root cause, for instance, by using the tools that existing laws provide. From designating cloud services as gatekeepers to including foundation models in the DMA—there’s a long list of things that can be done quickly. It needs the political will to follow through.

Europe is falling for Trump’s inauguration theater

Trump’s $500 billion announcement happened before markets reacted to DeepSeek. Nonetheless, Europe’s knee-jerk reaction has been revealing: panic, anxiety, and fear of missing out.

What got lost in the frenzy is a fact-check of Trump’s PR spin: this isn’t public investment. The project also hasn’t yet secured the funding it requires (and the real commitments, as Musk noted publicly, are closer to figures in the ballpark of $10 billion instead). Reporting by the Financial Times also suggests that the deal exclusively serve ChatGPT maker OpenAI.

For the U.S. AI industry and the Trump administration, this kind of photo op is a win-win. Trump gets to create the impression that, seemingly overnight, his administration is bringing in big investments. Meanwhile, the U.S. AI industry becomes further entrenched as “too big to fail” and “too important to hold back in any way”—at a time when its future business viability remains as ambiguous as ever. (Even before the DeepSeek correction, OpenAI was a loss-making firm.)

For Europe, this is a reminder not to fall for cheap PR and to move on from its knee-jerk reactions to copying whatever the U.S. is doing on AI.

In Other News

Commission releases Competitiveness Compass

• After EU-typical leaks, the Commission officially presented its Competitiveness Compass, effectively a translation of the Draghi report recommendations into a roadmap of the Commission.

• Proposals include: ‘AI Gigafactories' and ‘Apply AI' initiatives to drive development and industrial adoption of AI in key sectors, a Data Union Strategy, the EU Cloud & AI Development Act, and a dedicated EU Start-up and Scale-up Strategy.

AI industrial policy

• The European Commission also launched its pilot lines under the Chips for Europe initiative.

• For its six-month presidency of the European Council, Poland shared plans for research and innovation. On the agenda was AI in science, the interim evaluation of Horizon Europe, and arming the EU.

• The Eurostack coalition has published a pitch for building logical and physical infrastructures to secure Europe’s role in competitive digital value chains.

• In a column in "Le Monde", economists Cédric Durand and Cecilia Rikap outline a path to restore autonomy to the states and companies of the European Union.

Competition and market concentration

• The Open Markets Institute has shared an expert brief on AI & Market Concentration.

• Former EU presidents urge the EU commission to break up Google’s adtech monopoly to protect democracy.

• UK ministers push out chair of Competition and Markets Authority (CMA) and replace the position with the former head of Amazon UK. Read Cory Doctorow’s analysis on why this has implications far beyond the UK.

• The CMA also published its provisional decision on cloud. The 16-month market investigation into the supply of public cloud infrastructure services in the UK provisionally finds that competition in the cloud services market isn't working “as well as it could be” (AWS and Microsoft Azure each control 40% of the UK market). The decision merely suggests that the CMA use its new digital markets powers to determine whether Amazon and Microsoft should be classified as having “strategic market status” due to their influence. The CMA must publish a final decision by August 4.

EU and UK AI business news

• Aleph Alpha and Silo report a new approach to tokenization that should make training more efficient.

• Multi million-euro deal between Mistral and AFP for a fact-based chatbot is announced in response to Silicon Valley rivals withdrawal from moderation.

• UK-based generative AI start-up Synthesia secured a valuation of 2.1 billion dollars.

• The European Space Agency and OVHcloud continue their partnership with an announcement to support space start-ups in AI.

AI investments in EU member states

• Spain’s government releases a new initiative to boost the adoption of AI technologies across businesses with 150 million euros.

• Poland announces 1 billion EUR investment on development of domestic AI technologies in 2025. A 100 million EUR initiative the Polish Defence Fund seeks to boost Polish start-ups working on local defense and dual-use technologies, particularly in areas such as AI and unmanned systems.

• France’s AI investment has nearly doubled in 2024 but is still behind the UK and Germany for the amount of venture capital raised overall.

• UK tech minister announces increase in procurement deals with homegrown start-ups.

• Investment into Italy’s data centers will double 10 billion euros in 2025-2026 (largely due to Microsoft and Amazon).

Public sector uses of AI

• The UK government drops multiple AI prototypes for welfare usage after false starts have been reported. The news comes amidst the UK government’s AI Action Plan to boost AI adoption and drive growth in the UK.

• The UK government also launched Parlex, an AI tool described as a “parliamentary vibe check” for policy teams. Its aim is to provide insight on political climate and policy popularity amongst MPs.

Next week, we’ll cover the AI Action Summit in Paris and France’s AI ecosystem.