EU’s €30B AI Gigafactories Tender Marks a Shift From Regulation to Industrial Power

For much of the past decade, Europe exercised technological influence through regulation. The AI Act became the clearest expression of that model: establish rules, define risks, and impose obligations on companies operating in the European market.

But rules cannot manufacture processors or create electricity capacity.

Europe currently accounts for only about 2 gigawatts of operational AI compute capacity, roughly 5 per cent of the global total. The US has about 35GW, or 78 per cent, while China has around 5GW, or 11 per cent. The imbalance matters because the economics of frontier AI have changed. Access to computing capacity is increasingly as important as software engineering, talent or data.

That leaves European companies dependent on infrastructure controlled largely outside the EU. More than 80 per cent of Europe's critical digital products, cloud services, computing infrastructure and intellectual property are sourced from non-EU entities, according to the figures underpinning the initiative.

The strategic risk is not simply that European firms pay foreign providers for cloud services. It is that access to the underlying technology can become a geopolitical variable.

Recent temporary restrictions on access to advanced US AI models, including Anthropic's, have provided a practical demonstration of that vulnerability. China, meanwhile, has been using state-backed mechanisms to expand domestic access to compute.

Brussels is responding by treating computing infrastructure as strategic industrial capacity rather than merely another commercial service.

That is the significant change.

Public money is being used to make private capital possible

The financial structure of the tender reveals how far the EU's approach has moved from conventional technology policy.

The programme envisages €10 billion in combined EU and national public funding alongside roughly €20 billion in private capital. The public contribution is therefore not intended to finance the entire infrastructure build. Its purpose is to make projects large enough, risky enough and long-term enough to attract investors that might otherwise hesitate.

The two tender lots are designed around that logic.

Lot 1 covers up to four projects. Each can receive up to €100 million in EU funding during the first phase and up to €400 million in the second. Computing capacity must initially match Europe's current leading AI Factory and eventually reach at least three times that level.

Lot 2 covers up to three larger facilities, with up to €200 million available in the first phase and €800 million in the second. Those projects must begin at twice the current leading AI Factory capacity and eventually reach at least four times it.

For infrastructure investors, the important feature is not simply the size of the grants. It is the presence of the state as an anchor customer.

Eighteen EU member states have signed the joint procurement agreement and will participate as public anchor clients, co-financing the facilities and procuring access to computing capacity. Consortia can bring together industrial companies, public bodies, start-ups and investors.

This creates a mechanism that European financial markets have often struggled to provide on their own: guaranteed demand around an asset that requires enormous upfront expenditure before revenue can be generated.

The model is closer to industrial planning than to the traditional European preference for technology-neutral markets.

Sovereignty still depends on American hardware

There is an obvious tension at the centre of the programme.

Europe wants technological sovereignty, but its new AI infrastructure will initially depend heavily on processors supplied by companies such as NVIDIA, AMD and Qualcomm.

The Commission has signed letters of intent with all three to help ensure hardware availability for the gigafactory projects. The facilities will combine advanced processors with high-speed networking, cloud technology and energy-efficient data centres.

That is a pragmatic distinction between sovereignty and autarky. Europe cannot create an independent frontier-AI hardware ecosystem quickly enough to wait for domestic alternatives before expanding compute capacity.

The longer-term answer is supposed to come elsewhere in the technology stack.

The proposed Chips Act 2.0 is intended to reduce strategic dependencies and strengthen advanced semiconductor production in Europe. The Commission's wider technology sovereignty package explicitly links chips, computing infrastructure, cloud services and AI rather than treating them as separate industrial sectors.

The contradiction is therefore temporary by design: buy the hardware needed now while attempting to increase European control over the supply chain later.

Whether that sequence works will depend on whether the infrastructure created by the gigafactories generates enough domestic demand to support the semiconductor ecosystem Europe is trying to build.

€30 billion is large - and still small

The headline figure is substantial, but it needs to be placed against the scale of the problem.

Europe faces a projected infrastructure investment deficit of up to €600 billion over the next decade. Against that number, a €30 billion gigafactory programme is not a solution to the continent's entire compute shortage.

It is an attempt to change the investment equation.

The Commission is building a chain in which one intervention makes another more viable. AI Gigafactories provide compute. The Cloud and AI Development Act is intended to expand cloud and data-centre capacity. The Scaleup Europe Fund is designed to supply late-stage financing to technology companies. Chips Act 2.0 addresses semiconductor production and supply-chain resilience.

The European Commission describes this broader approach as a connected technology-sovereignty strategy spanning chips, infrastructure, software, cloud and AI.

The wager is that infrastructure can become an economic multiplier rather than a standalone expenditure.

That is also why access to the new capacity will matter as much as its construction.

The stated beneficiaries include start-ups, scaleups, small and medium-sized companies, industry, academia and public authorities. The Commission's own AI infrastructure policy envisages these facilities as shared resources rather than infrastructure reserved exclusively for a handful of technology groups.

But a shortage does not disappear merely because new capacity is publicly financed. Decisions over allocation will determine whether the gigafactories actually broaden Europe's AI ecosystem or simply create a new layer of scarce infrastructure around which larger companies compete.

Europe is choosing a different form of state intervention

The comparison with the US and China makes the European experiment more revealing.

The US has built its enormous compute advantage primarily through the scale of private technology investment, with its leading companies operating within a comparatively less restrictive regulatory environment. China has combined large corporate investment with targeted state support, including local compute voucher programmes and substantial public funds.

Europe is constructing a hybrid model.

It does not have the depth of private technology capital that produced the US advantage, nor does it operate China's system of direct state direction. Instead, Brussels is using public money, procurement and regulation to create conditions under which private capital can build infrastructure that the market has not supplied at sufficient scale.

This is demand-side sovereignty: the state becomes not only regulator, but customer and risk-sharing investor.

The approach extends beyond AI. CADA is designed to expand European cloud and data-centre capacity, while the wider sovereignty package includes mechanisms for supply-chain monitoring, risk assessment and intervention in critical sectors. The European state is acquiring tools to influence where critical technology is produced, who supplies it and how quickly infrastructure can be deployed.

That carries a cost as well as an opportunity. Once governments become anchor customers and co-investors, access, procurement rules and industrial priorities inevitably shape which companies receive an advantage.

The deadline may be easier than the infrastructure

The most immediate test will not be whether Europe can announce seven projects. It will be whether those projects can physically be built on schedule.

A gigafactory requires more than processors and financing. It needs power, grid capacity, data-centre infrastructure, high-speed connectivity and permits. The Commission's target of operation within 18 months of contract signature therefore puts pressure on national and regional authorities as well as private operators.

Energy infrastructure may prove particularly important. AI computing capacity is inseparable from electricity availability, and the location of data centres can become constrained by grid connections as much as by land or investment.

The EU has already recognised the connection between AI infrastructure and energy policy through its broader technology sovereignty programme.

There is another timing problem. The gigafactories are supposed to give Europe strategic autonomy while the European semiconductor industry is still being rebuilt. The facilities therefore have to scale rapidly enough to close part of the compute gap while relying, at least initially, on global hardware suppliers.

That makes the programme less a declaration of independence than a race against dependency.