The European Commission is shifting its approach to AI from regulation to industrial policy initiatives – most notably through five proposed AI Gigafactories. In this policy brief we argue that this initiative should place more weight on demand and we suggest two operational models for the AIGFs. 

This year marks a clear shift in the European Commission’s approach to artificial intelligence – from a primary focus on regulation towards competitiveness and industrial capacity. In April, the Commission published its AI Continent Action Plan, outlining how the European Union intends to close the gap in the global AI race. 

Expansion of compute infrastructure as a priority for the European Commission 

At the heart of the plan is a major expansion of compute infrastructure – the key ingredient for training and deploying AI models. The Commission proposes five AI Gigafactories (AIGFs), each hosting at least 100,000 advanced AI chips. The initiative would be partially funded by the Commission: a dedicated €20 billion fund under InvestAI – part of a broader €200 billion programme to scale AI development and adoption – would cover about one-third of each site’s capital expenditure. The explicit goal as announced by the Commission is for these AIGFs to train and deploy frontier AI models. 

Two operational models for the AI Gigafactories Initiative 

We have analyzed the AIGF initiative in a policy brief and situated it within the broader context of AI compute build-outs.
We argue that the initiative places too much weight on Europe’s past compute limitations – or insufficient supply and overlooks a critical factor: demand. 

Mapping existing data centre build-outs and the landscape of compute suppliers and users, we suggest two plausible operating models for the AIGFs: 

  • Anchor-customer model: Secure one or a few anchor customers with very high AI compute demand, as seen in the United States and China. 
  • Multi-client model: Serve a broader set of clients with low to moderate AI workloads. 

The multi-client model as the more feasible option for the EU 

We argue that the European Commission’s ambition to train and deploy frontier models aligns more closely with the anchor-customer model. However, leading AI labs are the only user group capable of generating high AI workloads and supporting the goal of training and deploying frontier AI models – and Europe currently hosts only one leading lab, Mistral. That makes the conventional anchor-customer model – in which a single high-demand lab guarantees utilisation – less feasible.  

A multi-client AIGF model is likely better suited to Europe’s industrial structure. However, to compete with private providers such as neoclouds, AIGFs would need to offer more than raw compute – acting as a one-stop shop that bundles the ingredients required to train and deploy diverse AI applications: structured onboarding, maturity diagnostics, curated software stacks and ongoing expert support. This approach could catalyse a dynamic ecosystem for SMEs, startups, and established companies, but it would require a recalibration of AIGFs’ current objectives. 

Policy recommendations for the upcoming review 

With the official call for proposals expected by the end of this year, policymakers should prioritise three criteria when assessing AIGF bids: 

  • Demand quantification: Require projected AI workloads and firm user commitments. 
  • Realistic objectives: Align build-outs with EU AI industry dynamics, not just frontier ambitions. 
  • Clear value proposition: Differentiate AIGFs from hyperscalers and neoclouds by offering integrated services – platform, software, and support – in addition to compute. 

This text is licensed under a  Creative Commons Attribution 4.0 International License