Denmark Must Not Become Europe’s AI Power Socket

By David A. Williams

Denmark’s decision to join Europe’s proposed AI-gigafactory initiative sounds like the beginning of a serious industrial strategy. The country has indicated that it is prepared to purchase up to DKK 750 million, (equivalent to roughly US$116 million) of computing capacity over five years, giving Danish researchers, companies and public institutions access to machines powerful enough to train and operate advanced artificial-intelligence systems. Europe intends to establish as many as seven gigafactories, each combining enormous quantities of advanced processors with software, cloud infrastructure, high-speed connections and energy-intensive data centres. This is being presented as an answer to Europe’s dependence on American and Asian technology. It might become one. But access to a powerful computer is not the same as possessing technological sovereignty, and Denmark should be careful not to confuse the two.

The word “gigafactory” is politically useful because it makes computation sound like manufacturing. It conjures up an enormous industrial plant turning raw materials into something valuable and distinctly European. Yet an AI gigafactory does not produce intelligence in the way a car factory produces cars. It provides the computational environment in which models can be trained, refined and operated. The more important questions concern who designs those models, who controls the software surrounding them, who owns the resulting intellectual property and who decides which users receive priority when capacity becomes scarce. A country may host the building, provide the electricity and finance part of the machinery while capturing surprisingly little of the value created inside it.

Denmark therefore needs to decide what it is actually buying for its DKK 750 million. Is it purchasing guaranteed computing hours, a strategic share of the infrastructure, preferential access during periods of high demand or merely the right to join a queue? Can unused Danish capacity be withdrawn or reallocated by the consortium operating the facility? Will Danish start-ups obtain affordable long-term access, or will the most powerful companies be able to outbid them? Can researchers inspect the software stack, or will they be working through proprietary systems controlled elsewhere? Unless these matters are settled contractually, Denmark risks acquiring something closer to a very expensive cloud subscription than a sovereign technological capability.

That distinction matters because artificial intelligence is becoming an organising layer of the economy. Compute is no longer simply another business input. It increasingly determines who can conduct advanced medical research, design new materials, automate public administration, develop military systems and build the models through which future knowledge will be organised. Denmark would never describe itself as sovereign in energy merely because it had permission to purchase electricity from a foreign-owned power station. Nor would it claim pharmaceutical independence because Danish hospitals could buy medicines developed and patented elsewhere. Yet the debate about AI infrastructure often applies a much weaker definition of sovereignty, as though the physical presence of processors on European soil were enough.

Real sovereignty is not autarky. Denmark does not need to manufacture every chip, write every program or train a national version of every frontier model. For a country of six million people, that would be wasteful and unrealistic. Sovereignty means possessing enough capacity, knowledge and legal authority to make meaningful choices. Denmark should be able to decide which models its public institutions use, move workloads between providers, protect strategically sensitive data, support domestic research and retain access during an international crisis or commercial dispute. Most importantly, Danish organisations must be capable of building upon the infrastructure rather than remaining permanent customers of systems designed by others.

This is where the European project could become genuinely important. The Commission says the proposed gigafactories will be operated by European-controlled consortia and provide start-ups, established companies, researchers and public authorities with facilities for training, fine-tuning and running advanced models. Up to €10 billion in European and national public support is expected to help mobilise at least €20 billion of private investment. That is serious money. It could give Europe the scale it has lacked while its best researchers, companies and capital have been drawn towards American platforms. It could also create an alternative to a digital economy in which a handful of foreign corporations control the chips, cloud services, models and commercial routes to market.

But public money does not automatically produce public power. If the processors are supplied by American companies, the cloud layer comes from American companies, the most capable models remain under foreign licences and the resulting intellectual property is held by private consortium members, Europe may have built a large machine without escaping the dependency that justified it. The nationality of a data centre’s postal address tells us very little about who controls its technological nervous system. Sovereignty cannot be bolted onto a warehouse after the procurement contracts have been signed.

Denmark is particularly vulnerable to becoming the ideal host and the secondary beneficiary. It offers political stability, excellent digital connections, engineering competence and an electricity system increasingly shaped by renewable power. These are genuine strengths, but they are also things that larger technology companies wish to consume. Land, cooling and electricity are the physical inputs of the AI economy. Models, patents, specialised software and accumulated technical knowledge are where much of its enduring power will reside. If Denmark supplies the former while other countries and corporations retain the latter, the project will reproduce a familiar economic pattern: the peripheral state provides the resources, while the centre owns the intelligence created from them.

The energy question cannot be separated from this calculation. Researchers at the University of Copenhagen’s National Centre for AI in Society have estimated that a single gigafactory could increase Denmark’s total electricity consumption by around five per cent, depending on its design and use. That is not an argument against building one. It is an argument for demanding a proportionate return. Electricity allocated to an AI facility cannot simultaneously be used by an industrial plant, a transport system, a district-heating project or another data centre. Grid capacity is finite, even in a country with substantial wind generation. Once computation begins competing directly with homes and industries for connections, the decision about who may compute becomes a decision about how national resources are distributed.

The right bargain would connect energy access to technological contribution. A consortium receiving favourable grid treatment or public financial support should be required to invest in Danish research, train Danish specialists, establish meaningful development teams in the country and make capacity available to universities and smaller companies at predictable prices. Intellectual property created with heavily subsidised Danish computation should not be allowed to disappear automatically into a foreign corporate balance sheet. Where public institutions contribute data, expertise or finance, Denmark should receive licences, equity, access rights or other durable benefits. Otherwise, the public sector will absorb part of the risk while private actors capture most of the upside.

There is also a democratic issue hiding beneath the engineering language. Decisions about access will shape which kinds of artificial intelligence Europe develops. A gigafactory optimised for training enormous general-purpose models serves a different political and economic purpose from one focused on medical research, climate modelling, industrial systems or secure public-sector applications. Capacity could be concentrated among a few companies capable of consuming it at scale, or deliberately distributed to universities, start-ups and public institutions. Neither approach is neutral. The scheduling system may ultimately exercise as much industrial influence as a ministry because it will determine which experiments proceed and which businesses never get the chance to grow.

Denmark should consequently negotiate for more than a national allocation of processor time. It needs representation in the governance of the facilities, transparent criteria for distributing capacity and enforceable rights during shortages. It should insist on portability, open technical standards and access to the tools required to fine-tune, evaluate and audit models. Danish companies using the infrastructure must be able to retain ownership of what they create. Public-sector data should remain under European legal control, while strategically important workloads should be protected against unilateral interruption by vendors or consortium partners. These conditions may sound technical, but they define whether Denmark enters the AI economy as a participant or a tenant.

A national programme should also surround the European investment. Computing capacity without skilled people is simply expensive machinery. Denmark needs researchers who understand model architecture, engineers capable of operating large distributed systems, specialists in semiconductor and energy integration, and public officials who can negotiate with technology companies without being intellectually dependent upon them. Access to a gigafactory should be linked to programmes that help Danish organisations turn computing hours into products, scientific discoveries and institutional knowledge. The objective must be to build a domestic ecosystem that becomes more capable each time it uses the infrastructure.

There is an opportunity here for Denmark to exercise the small-state advantage. It does not possess the money to imitate the United States or China, but it can concentrate resources around sectors in which it already has scientific knowledge, trusted data and industrial depth. Pharmaceutical research, biotechnology, energy-system management, shipping, agriculture, robotics and public administration all offer areas in which Danish institutions might build specialised models more useful than another general chatbot. Sovereignty will not come from creating a Danish imitation of every American product. It will come from ensuring that Denmark controls critical applications where its economy and public services cannot safely depend upon the priorities of somebody else.

The government should publish a clear test of success before the contracts are settled. How much guaranteed capacity will Denmark receive? Who will own the models and discoveries produced with publicly financed computation? How many Danish companies and researchers will gain access? What happens during a shortage? Which parts of the technology stack will remain proprietary? How much electricity will be required, and what economic value must the facility produce in return? Without measurable answers, almost any outcome can later be declared a triumph simply because a large building has been constructed and its processors are running.

Europe certainly needs more computing power. Denmark is right to seek a place inside the project rather than watching another technological system become established around it. But the success of the initiative should not be measured in chips installed, megawatts consumed or ministerial photographs taken beside humming machines. It should be measured by the choices Denmark gains. Can Danish researchers train important models without asking permission from a foreign platform? Can domestic companies turn national expertise into intellectual property they retain? Can public institutions use powerful AI without surrendering control of their data and essential services?

If the answer is yes, the gigafactory initiative could become part of Europe’s long-delayed technological recovery. If the answer is no, Denmark will have helped finance an impressive piece of infrastructure while remaining dependent on the companies that control the most valuable layers above it. The country will provide land, electricity and political legitimacy, then pay to access the intelligence produced from them. That would not be computational sovereignty. It would be digital tenancy with a Danish power cable.

Sources

Danish Ministry-sourced announcement on Denmark’s participation, reporting a possible Danish commitment of up to DKK 750 million over five years.

European Commission: AI Gigafactories call, outlining plans for up to seven facilities, public support of up to €10 billion and at least €20 billion in anticipated private investment.

University of Copenhagen and CAISA analysis, examining governance, access, business models and the potential effect of a gigafactory on Danish electricity consumption.

European Commission overview of AI research and frontier-model policy, describing the role of computing capacity, data, talent and frontier models in European technological sovereignty.