The European Union is investing 25 million euros in a network of public data laboratories to create the first open alternative to the closed datasets of OpenAI and Google. This EuroHPC initiative reveals a strategic shift: in tomorrow’s AI, scarcity will no longer be computing power but legal access to quality data.
The European bet stakes on transparency against opacity. While American giants rely on billions of web pages scraped in legal gray zones, Europe is methodically building an infrastructure of open and compliant data. A strategy that could reshuffle the deck if regulatory compliance costs explode on the other side of the Atlantic.
The Essentials
- The EuroHPC initiative mobilizes 25 million euros divided between coordination of AI centers (€2.5M), a network of data laboratories (€7.5M), and a federated European service (€15M)
- The objective: create the first public alternative to proprietary datasets from OpenAI, Meta, and Google, entirely compliant with European law
- The strategy aims to shift AI competition from hardware (GPU) to legal access to training data
- Initial deployment planned for 2026 with national supercomputing centers as anchoring points
AI Giants Attacked on Their Blind Spot
The American champions of artificial intelligence built their lead on a resource they believed was inexhaustible: the open web. OpenAI scraped billions of pages for GPT, Meta helped itself to social networks, Google indexed the digital planet. This bulldozer approach gave them a decisive advantage, but it rests on a fragile assumption: that the use of this data would remain legally acceptable.
Europe attacks precisely this Achilles’ heel. The General Data Protection Regulation and the AI Act create an environment where the use of non-consented datasets becomes risky and costly. Result: European companies developing AI models find themselves paralyzed by compliance costs, while their American competitors continue navigating murky but fertile waters.
The EuroHPC Joint Undertaking reverses this asymmetry. Rather than suffer the regulatory disadvantage, Europe turns it into a competitive advantage. The 25 million euros fund three pillars: 2.5 million to coordinate national AI centers (AI Factories), 7.5 million to create a network of data laboratories, and 15 million to develop a federated service for European web data.
This architecture pursues a simple objective: offer European AI developers training data as rich as that of the giants, but legally airtight. An open corpus, documented, and compliant by design with the strictest data protection standards.
Scarcity Shifts from Computing to Data
The timing of this initiative is not fortuitous. After years when GPU access constituted the main bottleneck in AI, the equation is changing. Europe is pooling its computing power to catch up with AI giants with its exascale supercomputers, but raw power is no longer enough.
Language models are reaching diminishing returns on public data. GPT-4 and its successors have already ingested the bulk of quality English-language web content. The next generations will have to turn to synthetic data, specialized datasets, or partnerships with premium content holders. This shift moves scarcity: fewer GPUs, more clean and legal data.
Europe anticipates this transition. Rather than race the giants on the hardware front, it is betting on creating an ecosystem of open data. The Data Labs funded by EuroHPC do not merely aggregate: they clean, structure, and guarantee legal compliance of datasets. Invisible but decisive work that could make the difference when copyright infringement lawsuits multiply.
This strategy relies on Europe’s regulatory advantage. While the United States still debates the rights to use web data, Europe is building the tools to do without it. The federated data service developed by EuroHPC will aggregate national datasets, European academic publications, and institutional content in an interoperable and reusable format.
Public Infrastructure Against Platform Oligopoly
The European approach breaks with the platform logic that dominates American AI. At OpenAI, Meta, or Anthropic, training data remain jealously guarded proprietary assets. It is impossible to verify their origin, quality, or legal compliance. This opacity generates considerable rents, but it also weakens the entire downstream ecosystem.
European Data Labs reverse this logic. Funded with public money, they produce open and documented datasets. Each dataset indicates its provenance, conditions of use, and quality measured according to standardized criteria. A model that recalls traditional scientific infrastructure: public, shared, and built for collective interest rather than private capture.
This philosophy appeals beyond Europe. American universities and Asian laboratories are already expressing interest in contributing to the network. The idea of a “CERN for AI data” is gaining ground, driven by growing frustration with platform oligopoly.
The 15 million allocated to the federated service precisely funds this ambition. The infrastructure will connect European supercomputing centers, partner universities, and research institutions in a decentralized network. Each node can contribute data, use shared datasets, and benefit from collective power without depending on a commercial intermediary.
The Limits of the Open Model Facing Private Capture
This public strategy is not exempt from classical tensions in open innovation. By creating freely accessible data, Europe also facilitates the work of its competitors. Nothing prevents Google or Meta from using European datasets to improve their own models, while keeping their proprietary datasets closed.
The risk of asymmetric capture is real. American giants have larger teams, greater capital, and superior technical mastery to quickly exploit open resources. They could transform European public infrastructure into mere input for their private platforms, without reciprocity.
The EuroHPC Joint Undertaking anticipates this difficulty by conditioning access to the most advanced resources. Basic datasets remain open, but value-added services—automatic annotation, advanced cleaning, optimized formats—will be reserved for actors contributing to the network. A credit system will measure contributions and balance usage.
This hybrid governance tests an unprecedented model: neither entirely open nor closed, but contributive. Intensive users will either have to bring their own data or finance the development of common tools. An approach that could inspire other public digital infrastructures, from mapping to climate models.
Europe’s Bet on Compliance as Competitive Advantage
The 25 million euro investment remains modest compared to the AI budgets of tech giants. Meta spends more each month on computing power. But Europe is not aiming for an arms race: it is betting on the systemic efficiency of shared infrastructure and on the lasting advantage of regulatory compliance.
This strategy could pay off when legal costs explode across the Atlantic. Lawsuits are already multiplying: publishers against OpenAI, artists against Midjourney, programmers against GitHub Copilot. Each dispute increases uncertainty about web data use and could force a complete revision of training practices.
Europe, meanwhile, is building the alternative now. Its Data Labs will produce datasets that are more limited, perhaps, but legally unassailable. An advantage that will count when companies prioritize legal certainty over raw performance, especially in regulated markets like healthcare, finance, or education.
The project’s success will depend on its ability to attract contributions. The 7.5 million allocated to data laboratories must convince universities, research institutes, and European companies to share their datasets. A challenge requiring credible financial incentives and equitable governance among contributors.
Deployment begins in 2026 with the first Data Labs connected to EuroHPC supercomputers. If the initiative delivers on its promises, it could redefine the political economy of AI by showing that a public and open alternative remains possible against platform oligopoly. A full-scale test to see whether Europe can still create its own rules of the technological game.