AI Gigafactory

An AI Gigafactory is a very large facility full of computers that train artificial intelligence. The name is modeled on the gigafactories of the auto industry, that is, on huge factories for batteries. Only here, nothing is built that you can touch. The product is computing power: the ability to perform enormous amounts of computational steps in a short time. Such a facility consists of tens of thousands of specialized computers, plus cooling, backup power, and its own connection to the power grid. The European Union has used the term since 2025 for a funding program with which it intends to build such sites in Europe.

Why Europe wants its own data centers of this size

Anyone who wants to train a very large AI model today needs access to enormous computing power. This capacity is predominantly located in the US and China. European companies and universities therefore often have to rent capacity from American providers. This costs money, creates dependency, and raises questions about data protection.

There is also an economic argument. Computing power has become a scarce resource, similar to oil or natural gas. Whoever possesses it has a say in which projects can even come into being. A start-up without access to large data centers may be able to come up with good ideas, but it cannot train a competitive model.

The EU has pledged around 20 billion euros for its program, combining public and private funds. Critics consider this too little, since individual US corporations spend similar sums on their own. Supporters counter that Europe would otherwise have no base of its own at all. The dispute over this regularly appears in business news.

What actually stands inside such a hall

The core components are graphics processing units, or GPUs for short. These are chips originally developed for video games that can perform very many simple calculations simultaneously. That is exactly what AI training needs. According to the EU's definition, an AI Gigafactory should contain at least around 100,000 such chips. For comparison: a well-equipped gaming PC has a single one.

The chips need to be able to communicate with each other very quickly, otherwise they wait for one another. That is why they are connected via extremely fast network cables and are placed close together. You can think of the whole thing less as many individual computers and more as a single machine spread across multiple halls.

The limiting factor is usually not the money for the chips, but the electricity. A facility of this size draws several hundred megawatts, as much as a medium-sized city. The resulting waste heat has to be removed, often using water cooling. Sites are therefore sought where electricity is cheap, for example in Scandinavia or Spain.

AI Gigafactories in the news

You mainly encounter the term in political reports. When the EU Commission selects sites or countries apply, it appears in the headlines. Germany, France, Spain, and several smaller countries have also submitted applications.

An important distinction: not every large data center is an AI Gigafactory. A normal cloud data center stores photos, runs websites, and delivers videos. An AI Gigafactory is tailored to a single task, namely training large models. Related, but smaller, are the so-called AI Factories, which the EU also funds in parallel and which serve more research-oriented purposes.

Another common misconception is that these facilities run the finished AI services. When you use a chatbot, your request usually runs on entirely different servers. Training happens once and takes months. Ongoing operation is a separate task, for which different hardware and different locations are used.