In her State of the Union address in September 2025, Ursula von der Leyen identified AI Gigafactories (AIGF) as one of the centrepieces of Europe’s technological sovereignty agenda. In that speech, the president of the European Commission said:
A European AI is essential for our future independence. It will help power our industries and our societies (…). We are massively investing in European AI Gigafactories. They support our innovative start-ups to develop, train, and deploy their next-generation AI models. When we called on the private sector to join forces with us, the response was overwhelming.
With this year’s State of the Union approaching, it is worth asking how that agenda has unfolded over the past year.
Public money as guaranteed demand
At the beginning of 2026, the EU adopted Regulation 2026/150, establishing the legal framework for providing public support to AI Gigafactories. Six months later, just before the summer break, the Commission published the call for the selection of AI Gigafactories Consortia and the Establishment of AI Gigafactories. The Commission estimated that each Gigafactory would require an investment of at least €4–5 billion. Given their size, the tender states that they are to be established under a “predominantly industrial, market-driven model, reflecting the need for long-term financial sustainability.”
Regulation 2026/150 gave the EU and the states participating in the EuroHPC JU (which is the EU’s joint initiative for supercomputing and AI infrastructure) two options for supporting these investments. They could either contribute directly to the capital expenditure of the computing infrastructure, or provide an equivalent contribution by guaranteeing the purchase of compute access time to the AIGF. The AI Gigafactories tender opted for the latter.
Rather than paying upfront to build the infrastructure, the EU and Member States commit in advance to buying compute once it becomes operational. The EU’s purchasing commitment is capped at the equivalent of 17 percent of the cost of the facility’s computing infrastructure and must be at least matched by participating states. In return, public authorities receive a guaranteed share of compute access time. For the initial deployment phase, €1 billion in EU funding has been committed under the current 2021–2027 Multiannual Financial Framework, again to be at least matched by participating states. A further €4 billion in EU funding is envisaged for the subsequent expansion phase, but this remains subject to the adoption of the post-2027 EU multiannual budget. Annex 2 sets out the financial commitments pledged by participating states, showing bidders how much national co-funding is available to satisfy the requirement that state contributions at least match the EU’s share.
The private partners forming the consortium that will establish and operate the AI Gigafactory must therefore finance the facility upfront, but do so knowing that part of their future revenue is guaranteed. In the words of the tender specifications, public authorities act as a “strategic anchor customer,” providing “a de-risking effect through this joint procurement of AI compute access time.” According to the tender, this provides operators with “significant financial security” and “exceptional revenue predictability.”
Where will the demand come from?
AI Gigafactories’ commercial viability will depend on finding customers for the rest of their capacity. The tender specifications pay considerable attention to this question: bidders are required to demonstrate “market traction and customer engagement,” for example by providing letters of intent, memoranda of understanding and pre-committed contracts from potential customers.
However, the demand used to demonstrate commercial viability does not have to come from European entities. Gigafactories are explicitly “open to a global user base,” and the tender specifications on page 9 allow “global service providers and international industrial partners (e.g. hyperscalers)” to act as anchor customers, helping to support the facilities’ operational readiness and commercial viability, particularly during the first phase. These global anchor customers may retain operational autonomy over the capacity and services they provide to their own users, and these services “may not be subject to the overall conditions of sovereignty of the AIGF.”
The tender is not blind to the risk this entails. The bidders must also explain how they will manage the risks associated with customer concentration. If a Gigafactory initially relies on a concentrated group of large customers, bidders must present a “clear, proportionate, and time-bound strategy” to progressively expand and diversify the customer base within the broader European AI ecosystem.
The apparent logic is that European demand alone may not initially be sufficient to make Gigafactories commercially viable. Public authorities provide a guaranteed base of European demand, while large global customers can provide additional demand. The expectation is that, over time, demand from the broader European AI ecosystem will grow.
This creates a clear tension for an initiative intended to strengthen European technological sovereignty. The actors with sufficient demand to help make investments of this scale commercially viable will likely include the same technology companies whose dominance of Europe’s cloud and AI infrastructure the sovereignty agenda is intended to reduce.
But there is another, perhaps more fundamental, assumption behind this strategy. Betting on the build-out of very large computing facilities in anticipation of future European demand is not neutral. It assumes that European AI companies, public institutions, and researchers will require this particular kind and scale of computing infrastructure. The bet is therefore not only that European AI demand will grow, but also that it will develop in a direction that requires infrastructure of this kind and scale. By investing ahead of that demand, the EU is actively steering European AI development in that direction.
The EU data centres market
The role assigned to hyperscalers in the tender reflects a broader feature of the European data centre market: operators rely on a small number of global technology companies as large, long-term customers. Such contracts provide the predictable demand that helps make major infrastructure investments commercially viable.
The Commission’s impact assessment accompanying the Cloud and AI Development Act points to this dynamic. New colocation capacity is often leased to Amazon Web Services, Microsoft and Google, largely because contracts with these hyperscalers allow data centre operators to secure a faster return on investment. By 2028, these three companies are expected to account for around 65% of the demand for data centres in Europe. Recent research by SOMO sheds more light on how this dynamic is playing out across Europe. The report finds that planned data centre development in the EU remains heavily dependent on demand from US technology companies. It provides several examples of how this dependence plays out in practice. While this does not necessarily mean that AI Gigafactories will reproduce this pattern, the structure of the existing data centre market suggests that it may be difficult to find sufficient demand without relying heavily on US technology companies.
This expected dependence on global actors may not be limited to demand, which is only one side of the equation. Gigafactories will require significant amounts of private capital, with the Commission hoping to “unlock” more than €20 billion in private investment. It is not yet known who will provide this capital, but the existing data centre market provides a warning here, too. SOMO found that data centre expansion in Europe is already financed to a significant extent by US investors, including large asset managers and infrastructure funds.
This creates two related dependencies. First, hyperscalers may provide the long-term demand that makes Gigafactories commercially viable. Second, global investors may provide the capital needed to build them. The shaky bet behind the EU’s AI Gigafactories initiative is that public procurement can alter this dynamic. By guaranteeing part of the demand, public authorities seek not only to reduce investment risk, but to steer private capital towards infrastructure designed to strengthen European technological sovereignty. In effect, the Commission is hoping to reverse the market dynamic that currently reinforces Europe’s dependencies.
Can de-risking deliver sovereignty?
The AI Gigafactories exemplify a broader approach the EU has adopted to financing its technological sovereignty agenda: using public resources to mobilise private investment and seeking to steer that investment towards public objectives. Daniela Gabor, who coined the term “de-risking state,” points to a tension in this kind of state-capital relationship where capital dominates: when public policy is structured around the risk-return requirements of private investors, the state’s ability to direct investment towards its own objectives may be constrained.
In the case of AI Gigafactories, the Commission tries to manage this tension by attaching sovereignty conditions to the private investment it seeks to attract. The tender stipulates that the Gigafactory coordinator must be established and effectively controlled in the EU, that the hosting entity and physical infrastructure must be located in the EU, and that core technical operations must be performed within the Union. It prohibits subcontracting core technical operations and data processing to entities established in non-EU countries and subjects potentially sensitive third-country access to public oversight. The tender also seeks to reduce technological dependencies: bidders are expected to present a concrete plan for a Europe-based and -controlled software stack, including cloud and AI layers, to become operational by the fourth year, and are encouraged to progressively integrate Europe-designed hardware.
These requirements are intended to ensure that private investment is channelled towards infrastructure over which Europe retains a degree of strategic and technological control. However, this control is not absolute. As noted in the previous section, global anchor customers, including hyperscalers, may retain operational autonomy over the capacity and services they provide to their own users. These services “may not be subject to the overall conditions of sovereignty of the AIGF,” though infrastructure-level safeguards and end-user screening will still apply.
What happens next
Consortia have until 12 November to submit their bids. The first Gigafactories are expected to be selected in early 2027. It will be important to see who their anchor customers are and where their financing comes from. Will the selected consortia reproduce the current pattern of infrastructure being physically located in Europe, but financed by global capital and reliant on demand from global technology companies? Or will the prospect of public de-risking be attractive enough to mobilise a different coalition of investors, operators and customers?
The Commission’s approach is an attempt to use public procurement both to build infrastructure that serves European sovereignty objectives and to steer private investment towards those objectives. This is an approach that many, including us, have called for. At the same time, that building and steering takes place within the limits that the EU has already set for itself. By committing to very large facilities before European demand for them has materialised, the EU has already made a strategic choice about the kind of AI ecosystem it wants to foster. Therefore, the bigger question might not be whether the EU can make Gigafactories sovereign, but whether the Gigafactories model is the technological trajectory for AI development that Europe should be supporting with public resources in the first place.
Ahead of this year’s State of the Union, Europe needs a clearer plan for its AI future. Rather than treating the current direction of AI development as given, asking how Europe can compete within it, and allocating public resources to mitigate the risks of a Silicon Valley-driven trajectory, Europe should take greater control of what it wants from AI. This must extend beyond policy strategies and action plans to encompass the technological and infrastructural choices that underpin them.
