Set as preferred source
Insights from State of AI in Europe: The Invisible Giant.
Europe’s position in artificial intelligence is frequently mischaracterized as lagging. The data suggests something more complex—and more structurally consequential.
The region has reached parity in frontier AI talent, leads in enterprise and consumer adoption, and continues to generate a high volume of startups. In 2025 alone, European AI companies attracted $21.8 billion in venture capital, representing a 58% year-over-year increase and over 30% of total VC funding in the region.
Yet this momentum does not translate into ownership.
The defining feature of Europe’s AI trajectory is not capability, but conversion. The system generates innovation, but cedes control once companies reach the stages where ownership, governance, and exit value are determined. This dynamic reflects an imbalance across capital allocation, compute infrastructure, and ecosystem integration that compounds at scale.
Europe is not behind. It became structurally constrained as capital, compute, and scale economics concentrated outside the region.
A system that originates value—but externalizes it
At the earliest stages, Europe performs competitively. The number of AI startups is comparable to that of the United States, and early-stage funding volumes are broadly aligned.
The divergence begins when companies move from technical validation to capital-intensive expansion.
European investment is approximately 3x lower at breakout stage and around 9x lower at late stage (based on $12B vs. $141B in 2025), with some analyses indicating gaps of up to 12x depending on segmentation.
This gap increasingly determines who owns the company, who sets its direction, and where long-term value accrues.
More than 50% of late-stage capital in European AI companies originates from foreign investors, predominantly US-based funds. At the largest funding rounds ($100M+), foreign investors lead the majority of deals.
As companies scale with external funding, control over governance, strategic direction, and exit pathways migrates with it.
The pattern is persistent: Europe builds companies, while others scale and own them.
Capital is not scarce—it is misallocated
The capital gap is often framed as a shortage. The data suggests otherwise.
European pension funds and insurance companies collectively manage over €20 trillion in assets, with funding ratios often exceeding 120–140%. Yet allocation to venture and growth capital remains minimal—often below 1%, compared to approximately 5% in US endowment models.
A reallocation to just 3% would unlock approximately €100 billion in additional growth capital.
The constraint is not capital availability; it is the absence of mechanisms that deploy capital into scale-stage opportunities.
This misalignment creates a reinforcing dynamic: European capital flows into global funds, which then re-enter the European ecosystem at later stages—capturing disproportionate ownership in the process.
Compute as a strategic gatekeeper
If capital defines ownership, compute defines possibility.
Europe hosts approximately 16% of global data center capacity, yet accounts for less than 5% of AI-specific compute. This gap limits the region’s ability to train frontier models and compete in compute-intensive domains.
In response, Europe is increasingly treating compute as a strategic layer of infrastructure, with growing public-sector involvement aimed at expanding sovereign AI capacity.
The implication is clear. Control over compute infrastructure is becoming a prerequisite for retaining strategic autonomy in AI. Without it, even well-funded companies remain dependent on external platforms for training and deployment.
Infrastructure cycles are long. Competitive cycles in AI are not.
The usage paradox: demand without capture
Europe leads in AI adoption across many markets, with usage levels in several countries exceeding those of the United States. In aggregate, interaction with large language models is significantly higher—approximately 2x relative usage levels.
This creates a self-reinforcing loop:
- High usage generates data and interaction signals
- These signals improve underlying models
- Improved models increase dependence
- Dependence further increases usage
The dominant models, however, are largely built outside Europe.
In effect, Europe’s demand subsidizes the improvement of external AI systems without securing corresponding control over the underlying models or platforms.
Demand functions as a strategic asset that is not captured domestically.
Talent: parity without platform formation
Europe’s AI talent base is substantial. The region hosts approximately 325,000 AI professionals, with an estimated ~50–55,000 classified as AI researchers (frontier talent, depending on definition).
However, distribution differs.
European AI talent is largely absorbed by incumbent industries, while US talent is more heavily concentrated in platform creation.
More than 50% of European AI talent is embedded in traditional sectors, while a smaller share operates within digital-native companies.
This results in a divergence in outcomes. One system optimizes existing infrastructure. The other builds new platforms.
At the same time, early signals suggest a shift in global talent flows. Migration from Europe to the US has slowed, and certain countries—notably the Netherlands—exhibit disproportionately high talent density (approximately 2x relative to population share).
Talent is no longer the constraint. Its concentration is.
Vertical strength without upstream control
Europe’s AI ecosystem is increasingly defined by vertical specialization. More than 75% of investment flows into applied domains such as healthcare, energy, fintech, and industrial systems.
One segment is accelerating particularly fast:
Defence, Security & Resilience (DSR)—an aggregated category spanning defence, cybersecurity, and broader resilience applications—accounted for approximately $8.7 billion in 2025, making it one of the fastest-growing segments with 55% year-over-year growth and a 4x increase over five years.
This reflects structural advantages:
- Industrial depth
- Regulatory alignment
- Public-sector demand
Vertical strength provides near-term defensibility, but it does not eliminate dependence on externally controlled model, compute, and distribution layers.
The next frontier: world models
The first wave of generative AI has been dominated by US-based players, driven by scale, compute, and capital concentration.
The next phase—world models—remains less consolidated.
These systems extend beyond language into physical interaction, simulation, robotics, and industrial environments. They require integration with real-world systems rather than purely digital inputs.
Europe’s capabilities in robotics, autonomous systems, and industrial AI position it closer to this frontier than current narratives suggest.
Unlike large language models, which reward scale and compute concentration, world models reward integration, domain expertise, and physical-system alignment.
This creates a narrow window of opportunity.
Fragmentation as a structural constraint
Europe’s ecosystem is defined by specialization without aggregation.
Key hubs demonstrate distinct strengths:
- London — model development, fintech, AI-native SaaS, and research concentration
- Munich — robotics, autonomous systems, industrial AI
- Zurich — deep research density and technical talent concentration
- Paris — emerging model development and enterprise AI
Collectively, these hubs form a distributed system of excellence. Individually, they lack the scale of US clusters.
Fragmentation is evident across:
- capital markets
- regulatory frameworks
- labor mobility
- corporate structures
Scaling across borders introduces friction that increases the cost and reduces the speed of company formation and expansion.
The missing layer: scaling architecture
The absence of a unified scaling framework remains a core constraint.
The report highlights the need for a more integrated European system—often framed through initiatives such as EU-INC—to reduce internal friction and enable cross-border scaling.
Key gaps include:
- fragmented stock option frameworks
- inconsistent labor laws across jurisdictions
- limited cross-border capital mobility
- regulatory overlap across multiple frameworks
Without alignment across these layers, scaling remains constrained within national boundaries.
In a system where competitors operate within unified capital markets and regulatory environments, fragmentation becomes a structural disadvantage.
Strategic risks
The current configuration creates a set of reinforcing risks:
Ownership leakage
European companies transition to foreign control at scale, shifting long-term value capture.
Compute dependency
Limited infrastructure reinforces reliance on external providers.
Capital misallocation
Domestic capital continues to strengthen external ecosystems.
Regulatory drag
Fragmentation slows scaling and reduces competitiveness.
Application-layer compression
Vertical solutions risk commoditization without upstream control.
A narrow window for rebalancing
Europe’s position in AI is not fixed. The region retains significant structural advantages:
- talent parity with global leaders
- strong adoption across sectors
- deep industrial integration
- influence over regulatory standards
However, the current trajectory positions Europe as an application-layer economy within externally controlled AI systems.
Rebalancing does not require new capabilities. It requires structural reconfiguration:
- aligning capital with scaling needs
- treating compute as strategic infrastructure
- reducing fragmentation across markets
- concentrating talent within platform-building environments
The next phase of AI—defined by infrastructure control, industrial integration, and world models—remains open.
But that window will narrow as capital, infrastructure, and distribution advantages continue to compound elsewhere.
