The European Union aims to respond to the dominance of the United States and China in artificial intelligence with a large-scale investment: mobilizing €20 billion to establish up to five AI gigafactories, each equipped with more than 100,000 advanced processors. On paper, the plan has political weight and an obvious logic. Without independent computing capacity, there is no real technological sovereignty, and without European infrastructure, AI will continue to rely on clouds, chips, and platforms controlled from abroad.
The problem is that the initiative is surrounded by doubts. Several experts, legislators, and industry actors question whether Brussels is designing the right response or if it is late to copy a model that is already changing. The question isn’t whether Europe needs more AI capacity. It does. The question is whether concentrating billions in large centers for training giant models is the best way to help European companies, governments, researchers, and industrial sectors.
The European Commission defends that gigafactories are not just data centers. They are presented as large-scale facilities for developing and training next-generation models, with over 100,000 AI chips, energy capacity, advanced networks, efficiency, and automation. Additionally, they are part of a broader strategy: the AI Continent plan, the InvestAI initiative, the 19 AI Factories already planned, and a future Cloud and AI Development Act that aims to triple EU data center capacity over the next five to seven years.
The risk of building capacity without knowing who will use it
The most uncomfortable critique pertains to demand. Brussels received 76 expressions of interest to build gigafactories in 60 possible locations across 16 member states. The number demonstrates industrial appetite but does not address a key question: which European companies will fill these facilities, with what workloads, business models, and economic returns.
Mistral AI, the most prominent European company in foundational models, does not seem to be waiting for community infrastructure. The French firm announced a €1.2 billion investment in data centers in Sweden alongside EcoDataCenter and has also strengthened its capabilities in France. In May 2026, its CEO, Arthur Mensch, advocated for a vertical integration strategy that includes infrastructure, models, and applications, aiming for 200 MW capacity by late 2027 and 1 GW by 2030.
This move raises an open question: if Europe’s leading AI champion is building its own path, who exactly are the EU gigafactories for? They could serve startups, universities, research centers, and industrial companies, but training frontier models is not an activity that can be improvised. It requires talent, data, software, clients, continuous funding, and the capacity to keep pace with hardware updates.
There is also doubt about size. €20 billion seems substantial until compared to private US spending. Large hyperscalers and AI labs are announcing investments of hundreds of billions in infrastructure, energy, chips, and data centers. Europe cannot compete solely by matching these figures, as it lacks the same capital market strength, the same cloud platform concentration, and the same depth in advanced chips.
Dependence on NVIDIA doesn’t disappear by building in Europe
Another sensitive issue is technological dependence. While EU gigafactories aim to strengthen sovereignty, if they are built almost entirely on NVIDIA GPUs, NVIDIA networks, NVIDIA software, and a supply chain dominated by non-European manufacturers, autonomy will be limited. Infrastructure will be on European soil, but much of the critical technology will continue to come from outside.
This doesn’t mean Europe should reject NVIDIA. That would be unrealistic. Its GPUs and systems are currently the backbone of much advanced AI. But sovereignty cannot be reduced to just buying many accelerators and hosting them in European centers. It also involves controlling data, software, models, energy, operations, security, maintenance, funding, and the ability to replace technology in the medium term.
The computing market itself is evolving. While training large models remains relevant, inference—the deployment of pre-trained models for millions of users, companies, and processes—gains increasing importance. This inference doesn’t always require the same chips, architecture, or physical concentration as massive training. In many cases, it can benefit from distributed infrastructure, specialized accelerators, CPUs, European chips with lower power consumption, edge computing, or private cloud.
Europe might have a more realistic opportunity there: not trying to emulate US-scale giant model training but creating a sovereign computing network for industrial, healthcare, scientific, energy, administrative, and defense use cases. Less obsession with building “the next European GPT,” and more focus on solving specific problems with European data, under European rules, and with controllable infrastructure.
Industry demands concrete objectives, not just more GPUs
The gigafactory plan could fall short if it becomes just a showpiece policy. A single facility with 100,000 GPUs is impressive, but does not ensure competitiveness. Europe’s advantage might lie elsewhere: automotive, healthcare, robotics, chemical industry, energy, advanced manufacturing, telecommunications, public sector, defense, agribusiness, logistics, and science.
In these sectors, Europe already has companies, data, knowledge, processes, and clients. Often, what’s missing is not just raw computing power but the ability to bring models into production, share data securely, comply with regulations, integrate AI into legacy systems, and measure results. An AI factory useful for Europe should offer more than GPUs: secure environments, deployment tools, inference services, technical support, governance frameworks, SME access, and domain-specific models.
The European Commission is attempting to address part of this via the AI Factories network, designed for startups, industry, and research. But the scale of gigafactories could end up absorbing political attention and private capital if their specific problems and benefits aren’t clearly defined. The risk isn’t spending money on compute, but spending without a clear industrial architecture.
There is also a physical bottleneck: Europe has faced years of data center demand exceeding supply. Power availability, permits, grid connection, land, cooling, and social acceptance all influence deployment. Building gigafactories isn’t just about allocating funds; it requires steady energy, local agreements, fiber networks, qualified operators, and plans aligned with environmental goals and community acceptance.
Sovereignty doesn’t mean copying the US
Europe’s answer should start from a simple idea: sovereignty isn’t about having the same as the US but smaller and slower. It’s about deciding which capabilities are strategic for the continent and funding them coherently.
Building one or several gigafactories makes sense if linked to concrete objectives: scientific models, defense, healthcare, industry, European languages, public administration, security, climate research, or advanced simulation. Reserving some capacity for startups and universities can also be justified. But designing the project as a delayed replica of Stargate risks arriving when the market has already shifted toward distributed inference, business agents, more efficient models, and specialized value chains.
Europe needs computing, but also strong European clients, European cloud providers, alternative chips, open software, proprietary models, technical talent, and regulations that don’t push companies to deploy outside the EU. If a European company finds it easier to train, serve, and sell AI in the US than within the EU, no gigafactory will solve the core issue.
The €20 billion plan can be an opportunity if used to build a lasting industrial base. To do so, Brussels should avoid the temptation of easy symbols. Opening buildings filled with GPUs isn’t enough. It’s necessary to ensure that this capacity is used, accessible, meets real needs, and reduces dependencies rather than disguises or reinforces them.
European AI will not succeed solely through size. Success comes with focus, integration, and a demand-driven policy. Less racing to build the biggest facility and more addressing practical questions: which models are trained, who uses them, what data feeds them, which sectors benefit, which providers participate, and how to ensure public money doesn’t reinforce the dependencies it intends to reduce.
Frequently Asked Questions
What are the AI gigafactories the EU aims to build?
Large computing facilities for training and developing advanced AI models, with over 100,000 AI processors per center and a planned investment of up to €20 billion.
Why are they criticized?
Because it’s unclear who will use all that capacity, whether mass training focus is the best approach, and if Europe will truly reduce its technological dependence if centers rely on chips and software outside of Europe.
What’s the difference between AI Factories and AI Gigafactories?
AI Factories are supercomputing centers aimed at startups, research, and industry. AI Gigafactories are much larger facilities designed for next-gen AI models and far more intensive workloads.
What would be a more effective strategy for Europe?
Combine increased computing capacity with clear sectoral objectives, inference infrastructure, European providers, open software, SME access, data governance, and concrete applications in industry, health, energy, defense, and public administration.
via: techzine