By Paolo Licandro
1st July, 2026


At the beginning of the 1980s, when I started working as a political adviser at the European Parliament, I followed the work of the committees dealing with industry, energy and research. It was a formative period. The European Community was beginning to understand that research policy could no longer be treated as a marginal academic matter. It had to become part of industrial strategy.

I remember a young German senior official in the Commission services dealing with the emerging field of information technology — at the time when Europe was preparing what would later become the ESPRIT programme — warning us that Europe was already missing the major trains of technological innovation. Consumer electronics, video recorders, television screens, personal computers and early digital technologies were moving elsewhere. The United States and Japan were turning innovation into industrial power. Europe was discussing programmes, procedures and coordination.

The European Commissioner for Industry at the time was Étienne Davignon. He understood the strategic importance of that warning, supported the young official’s work, and helped transmit the message across the European institutions: unless Europe acted at scale, it would become dependent on technologies developed, manufactured and commercialised elsewhere.

That warning has stayed with me for more than forty years because it was not wrong. It was early.

Many other voices have been added to that warning over the past forty years. The latest, and perhaps one of the most authoritative, is Mario Draghi, former President of the European Central Bank and former Prime Minister of Italy. In his report to the European Commission on the future of European competitiveness, Draghi denounced Europe’s growing innovation gap, its weak productivity growth, its excessive fragmentation, and its dependence on external suppliers in key strategic technologies.

The message is strikingly similar to the warning we heard in the early 1980s. Europe does not lack intelligence, scientific talent or regulatory capacity. What it lacks is the ability to transform technological opportunities into large-scale industrial power. Draghi’s report makes the point with unusual clarity: without faster innovation, deeper capital markets, lower energy costs, stronger industrial coordination and much larger investment, Europe risks falling further behind the United States and China in the technologies that will define the next economic cycle.

Europe has had important successes, but the structural pattern has too often remained the same. We have been excellent at science, standards, regulation and engineering niches. We have been much weaker at transforming technological waves into large-scale industrial ecosystems.

GSM was a European success as a standard. But Europe did not retain control of the next layers of value: smartphones, mobile operating systems, app stores, digital platforms, cloud infrastructure and data ecosystems. European companies that once mattered in phones, consumer electronics, displays and digital equipment progressively disappeared, were absorbed, or became marginal in the most profitable parts of the value chain.

Airbus succeeded because it was not only a research project or a regulatory problem. It was a political-industrial project backed by states, procurement, engineering, financing and a clear strategic objective: Europe needed a civil aircraft champion capable of competing with the United States.

Now the same question returns with artificial intelligence: Will Europe once again regulate a market that others are building?

The European Union has moved faster than most jurisdictions in regulating artificial intelligence. The AI Act is an important legal instrument. It creates a risk-based framework, defines prohibited practices, sets obligations for high-risk systems, introduces transparency rules and addresses general-purpose AI models.

This is not irrelevant. Trust, safety, accountability and fundamental rights matter. Europe is right to insist that AI should not be deployed without legal and ethical safeguards. But regulation is not industrial policy. A market cannot be built only by defining the conditions under which foreign technologies may operate inside it.

The uncomfortable truth is that Europe risks becoming the most sophisticated regulator of an AI economy whose critical infrastructure is produced elsewhere. The foundation models are largely non-European. The cloud platforms are largely non-European. The most advanced AI chips are largely designed outside Europe. The data-centre giants are largely non-European. The scaling capital is largely non-European. The commercial platforms through which AI reaches users and businesses are largely non-European.

Europe is not absent from the technology chain. The duch ASML is one of the most strategic companies in the world for Photolithography systems for the semiconductor industry. It provide technology for producing chips, not to produce them.  European research is strong. Several European AI startups are promising. The EU has launched ambitious initiatives such as InvestAI, AI factories and the European Chips Act. But the aggregate picture remains worrying: Europe has components of excellence, not yet a sovereign AI ecosystem.

One of the most dangerous misunderstandings in the European debate is the idea that AI is mainly an application layer. It is not.

Modern AI is an industrial system. It requires chips, cloud infrastructure, data centres, energy, water, advanced packaging, high-bandwidth memory, specialised engineering, large procurement contracts and massive capital expenditure. The chatbot on the screen is only the visible surface. Behind it there is a physical machine: servers, GPUs, cooling systems, electricity grids, fibre networks, data pipelines and software stacks.

This is where Europe is structurally weak.

The AI race is increasingly a race for compute. Whoever controls compute controls the speed of model development, the cost of deployment, the capacity to serve users, the ability to build industrial applications and the bargaining power over the rest of the economy.

AI chips are therefore not a technical detail. They are strategic infrastructure. NVIDIA H100, H200 and Blackwell systems are not simply “better processors”; they are the engines of the new AI economy. They determine who can train frontier models, who can run large-scale inference, who can offer AI services at competitive prices and who can accumulate the operational data needed to improve future systems.

Europe cannot seriously claim AI sovereignty if it does not address the hardware and infrastructure base of AI.

Artificial intelligence rewards scale. This is true in several ways.

• First, there is scale in capital. The largest AI companies and cloud providers are investing at levels that no ordinary European programme can match. The United States benefits from a combination of hyperscalers, venture capital, defence and federal procurement, deep capital markets and a culture of rapid deployment. China combines state direction, industrial policy, domestic demand and strategic mobilisation.
• Second, there is scale in infrastructure. Training and deploying frontier models requires vast clusters of accelerators, advanced networking and reliable access to energy. Data centres are no longer peripheral facilities. They are becoming central nodes of economic power.
• Third, there is scale in procurement. Large orders create industrial learning. They justify investment in production capacity. They reduce unit costs. They anchor supply chains. Without large public and private procurement, Europe cannot create the demand pull needed to build competitive AI infrastructure.

This is where European fragmentation becomes costly. Twenty-seven national strategies do not make a continental AI strategy. Small national projects do not create scale. Announcements do not create industrial capacity. Pilot programmes do not replace procurement.

Airbus succeeded because it was not only a research project. It was a political-industrial project backed by states, procurement, engineering, financing and a clear strategic objective: Europe needed a civil aircraft champion capable of competing with the United States.

AI requires an Airbus logic, not another collection of dispersed initiatives.

The AI debate is often conducted as if compute capacity could be expanded simply by buying chips. But chips need data centres, and data centres need electricity, cooling and water.

This is a major strategic problem for Europe.

The International Energy Agency estimates that global electricity consumption by data centres could more than double by 2030, reaching around 945 TWh. That would make data centres one of the most dynamic new sources of electricity demand in the world economy.

For Europe, this matters because energy is already a competitiveness problem. Industrial electricity prices, grid constraints, permitting delays and the slow expansion of firm low-carbon generation all affect the ability to host large-scale AI infrastructure. A region that struggles to provide competitive energy for industry will struggle to provide competitive energy for AI.

Water is the other neglected issue. Large data centres require cooling, and cooling requires either water, electricity-intensive alternatives, or location strategies designed around climate, hydrology and grid capacity. Europe has no serious continental discussion about where AI data centres should be located, how they should be powered, how water use should be managed, and how infrastructure planning should be coordinated across borders.

This is not a secondary environmental issue. It is an industrial planning issue.

If Europe wants sovereign AI capacity, it must decide where the compute will physically sit, who will power it, who will cool it, who will finance it and who will use it.

Europe’s position in semiconductors is paradoxical. It has world-class assets in semiconductor equipment, research and selected industrial applications. But it does not control the most important AI accelerator ecosystem.

The advanced AI chip market is dominated by non-European actors. NVIDIA leads in data-centre GPUs. AMD is the main alternative in merchant accelerators. Google, Amazon and other hyperscalers are building custom AI chips for their own clouds. Huawei and other Chinese players are developing domestic alternatives under geopolitical pressure. Taiwan Semiconductor Manufacturing Company (TSMC)  remains central to advanced manufacturing, while high-bandwidth memory and advanced packaging are concentrated in a small number of suppliers. TSMC produces NVIDIA chips.

Europe’s Chips Act aims to strengthen the semiconductor ecosystem and reduce dependencies. That is welcome. But the question is whether the scale, speed and focus are sufficient for the AI era.

The AI chip market is not simply about producing “more chips”. It is about producing the right chips, at the right nodes, with the right memory, packaging, software ecosystem, cloud integration and customer base. A fab without demand is not sovereignty. A research programme without industrial deployment is not sovereignty. A subsidy without scale is not sovereignty.

Europe often describes its approach as human-centric, trustworthy and rights-based. These are legitimate objectives. 

But they do not answer the industrial question.

• Who owns the models?
• Who owns the cloud?
• Who owns the chips?
• Who owns the data-centre infrastructure?
• Who controls the software stack?
• Who captures the margins?
• Who sets the de facto standards through market dominance?

If the answers are mostly non-European, then Europe’s role is reduced to regulating access to its internal market. That may provide legal influence, but it does not create industrial power.

The Brussels effect can shape global rules, but it cannot substitute for production capacity. Legal influence is not the same as technological leadership. A continent that regulates platforms owned by others, chips designed by others, clouds operated by others and models trained by others is not sovereign. It is administratively sophisticated, but strategically dependent.

The crisis of the European automotive sector should be read as a warning. Europe spent years regulating emissions, defining targets and debating transition pathways. But the industrial centre of gravity in batteries, electric platforms, software-defined vehicles and supply chains shifted rapidly. China built scale. The United States mobilised industrial policy. Europe moved, but too slowly and too defensively.

Artificial intelligence could repeat the pattern, but faster.

In automotive, Europe at least had powerful incumbent companies. In AI infrastructure, Europe starts from a weaker position. It has no European equivalent of NVIDIA, Microsoft, Amazon Web Services, Google Cloud or OpenAI at global scale. It has no single capital market comparable to the United States. It has no unified procurement machine comparable to a large federal state. It has no integrated industrial command structure comparable to China.

This does not mean Europe cannot act. It means incrementalism is not enough.

Europe needs a serious AI industrial strategy based on five pillars.

1. First, compute capacity. Europe should build genuinely large AI compute facilities, not symbolic projects. These facilities should be accessible to European companies, researchers and public administrations under transparent conditions, but they must be large enough to matter.
2. Second, chip strategy. Europe should use the Chips Act and future instruments to focus not only on generic semiconductor capacity, but on AI-relevant accelerators, advanced packaging, high-bandwidth memory integration, photonics, edge AI and strategic supply-chain bottlenecks.
3. Third, energy and water planning. AI infrastructure must be integrated into energy policy, grid planning, renewable deployment, nuclear discussions where relevant, water management and regional development. Data-centre policy cannot be left to local permitting and corporate site selection alone.
4. Fourth, procurement. The EU and Member States should create demand. Public administrations, defence, healthcare, research, education, energy systems and industrial policy can all become anchor customers for European AI solutions. Without procurement, Europe will continue financing innovation that others commercialise.
5. Fifth, governance at European scale. National AI strategies are insufficient. The relevant market is continental. The relevant competitors are global. Europe needs common projects, common financing and common infrastructure. Otherwise, internal divisions will again convert ambition into fragmentation.

A final choice

Europe is not condemned to decline in artificial intelligence. It has talent, scientific capacity, engineering culture, industrial users, regulatory sophistication and some strategic assets. But it must stop confusing coordination with strategy and regulation with industrial power.
The AI race will not be won by declarations. It will be won by those who build the machines, finance the infrastructure, secure the energy, control the chips, train the models, deploy the applications and capture the markets.

Europe can still choose to be more than a regulator. But that choice requires money, scale, speed and political will.
Otherwise, the result is predictable. Europe will regulate AI. Others will build it. European companies and public administrations will buy it. European citizens will use it. European policymakers will debate it. But the strategic value will be captured elsewhere.

That is why the debate on artificial intelligence must go beyond the AI Act. It must include chips, data centres, energy, water, procurement, industrial policy and digital sovereignty.

Because the central question is no longer whether Europe can regulate artificial intelligence.
The central question is whether Europe can still build the future it wants to regulate.

@ eEuropa Belgium, eEuropa.org - 2020-2026