When we talk about artificial intelligence we use apparently weightless words: cloud, model, token, inference. In reality modern AI is one of the most physical infrastructures built by the digital economy. Every response travels through data centers, chips, cooling systems, high-speed networks and power grids. As models grow, competition becomes less about algorithms alone and more about the ability to own, finance and power the machines on which those algorithms run.
This is the context for the European Union’s AI Gigafactory initiative. The Commission describes these facilities as large-scale computing centers dedicated to training, fine-tuning, inference and the development of next-generation models. InvestAI aims to mobilize around €20 billion to support several gigafactories across Europe. The ambition is larger than adding a few more conventional data centers: Europe wants computing capacity capable of competing with infrastructure concentrated in the United States and Asia.
The scale of interest is significant. A preliminary European process attracted dozens of proposals across multiple member states and potential sites. Each bid involves far more than a technology company. It requires energy, land, permits, high-speed networks, universities, research centers, capital and local infrastructure. AI therefore becomes a territorial project. Like railways, power plants and highways before it, computing capacity starts reshaping economic geography.
Europe sees these facilities as strategic because access to compute has become a bottleneck. A country can have strong researchers, datasets and startups, but if frontier GPUs are scarce or prohibitively expensive, advanced experimentation remains limited. Compute is becoming a raw material. Whoever controls the infrastructure influences which experiments are economically possible, at what scale and how often they can be repeated.
Energy is consequently central. An AI Gigafactory cannot be planned as if electricity were a secondary technical detail. Large GPU clusters demand enormous continuous power, stable grids and sophisticated cooling. Site selection will be tied to energy availability, transmission capacity, climate, water and the ability to integrate renewables or other reliable generation. The real cost of a model is not only the chip bill; it includes the physical ecosystem that keeps the chips alive.
Data-center architecture is changing as well. AI accelerators concentrate far more heat than traditional servers and increasingly require liquid cooling. Internal networks must move huge volumes of data with minimal latency. Storage, security and redundancy become parts of a single organism. The word factory is therefore surprisingly accurate: inputs, energy, machinery and flows are orchestrated to manufacture a new strategic commodity—compute.
Europe also wants these infrastructures to serve research and industry beyond the largest foundation-model labs. The stated goal is to give companies, startups, universities and public institutions access to resources they could not independently finance. This matters. A gigafactory that merely imitates Big Tech has limited strategic value. It becomes transformative when it supports medicine, manufacturing, robotics, climate research, European languages and creative industries.
There is a danger, however, in confusing scale with innovation. More GPUs do not automatically produce better models, just as a giant film studio does not automatically produce a great film. Data quality, research, algorithmic efficiency and the ability to convert AI into useful applications remain decisive. Infrastructure policy must therefore avoid becoming purely quantitative.
For image makers the subject may appear distant, but it is not. Generative video, neural rendering, 3D models, audiovisual translation and synthetic production all depend on compute. Creative AI services feel almost magical because the physical infrastructure is hidden behind a browser. Gigafactories expose what exists behind the interface: buildings, servers, energy and capital.
There is also a geopolitical layer behind the technical language. The most advanced chips are produced by a small number of companies and countries, while hyperscale cloud infrastructure is dominated by American groups. European capacity will not create complete independence overnight, but it can reduce strategic vulnerability. Compute infrastructure is therefore also bargaining power: the more capacity Europe controls, the more freedom it has in industrial and scientific policy.
Local impact matters too. A major computing center can bring investment and high-skilled employment, but it can also place pressure on electricity networks, water resources and surrounding communities. Public debate will need to become more transparent: how much energy will the site consume, where will that energy come from, can waste heat be reused, and how much capacity will actually be available to universities and smaller companies?
Then there is time. Physical infrastructure takes years to authorize and build, while AI hardware changes in months. A gigafactory designed today may need very different accelerators and cooling systems when it opens. The architectural challenge is therefore unusual: build concrete, power and networking systems flexible enough to host a technology that evolves faster than the building containing it.
FACTORIES OF INTELLIGENCE is therefore more literal than it sounds. AI does not live in the cloud; it lives in very concrete places. Europe is trying to build those places before the infrastructure gap with global technology powers becomes irreversible.