The expansion of artificial intelligence is accelerating the construction of data centers around the world and transforming the geographic distribution of this strategic infrastructure. According to the study The Global Build-out Race: AI Data Center Study | Part 1, prepared by Roland Berger, global installed capacity will multiply fourfold in a decade, going from 88 gigawatts (GW) in 2025 to 341 GW in 2035. During that period, the weight of AI in total power demand will grow from 20% to almost 60%.
Forecasts indicate that global capacity will reach 224 GW in 2030, after registering an annual growth of 21%. By then, artificial intelligence will account for about half of data center energy demand. However, this development will not be homogeneous. Europe will lose relative weight and will go from concentrating 13% of global installed capacity to 10%, while North America will maintain a dominant position, with around 58%.
Investment in hyperscalar doubles
The increase in computing needs of new generations of large language models, together with the rapid adoption of these technologies, explains the drive for digital infrastructure projects. In this context, the investment of large cloud operators, known as hyperscalers, will be around $750 billion in 2026, almost double that of 2025.
Roland Berger believes that the sector’s long-term growth prospects remain strong. However, it warns that the current investment pace, which exceeds the income generated by AI at present, and the pressure on the cash flow of some operators introduce vulnerabilities in the market and could cause a temporary moderation of activity.
«AI data centers are very capital-intensive infrastructures with long development times, and those who invest need stable rules and predictable processing times. The study shows that financial aid is not a sufficient incentive and that location decisions largely depend on the regulatory framework and network access. In addition to the United States, there is competition between European markets, so countries that offer more predictability will take advantage,” says Juan Luis Vílchez, senior partner of Roland Berger in Iberia.
North America increases its advantage over Europe
The differences between both regions are already evident. In 2025, North America had 4,322 data centers, of which 4,032 were in the United States. This figure exceeds the sum of the facilities in Western Europe and Asia. Additionally, North American capacity will grow from 49 GW in 2025 to 130 GW in 2030.
Europe will also expand its infrastructure, but at a slower rate. Its capacity will increase from about 11 GW to approximately 23 GW during the same period. Although this means practically doubling the installed power, the annual growth will be 15%, compared to the 22% expected for North America and 19% for China.
The gap is also reflected in the size of the large projects announced. Among the developments that exceed one gigawatt of capacity, only one is located outside the United States: the SINES campus, in Portugal, which plans to reach approximately 1.2 GW in 2030. In Europe, Cambois, in the United Kingdom, also stands out, a project whose planned capacity is around 700 megawatts.
Added to this difference in scale is the profile of European activity in artificial intelligence. The region is mainly oriented towards inference, that is, the execution of already trained models, which usually requires smaller facilities than those intended for training frontier models. This specialization limits Europe’s ability to organically attract large investments in data centers.
Up to seven years to connect a data center in Europe
The report identifies three major obstacles that explain Europe’s difficulties in competing in this market. The main one, according to the professionals surveyed, is access to the electrical grid. Connecting a new data center can take up to seven years in Europe, compared to one to three years in the United States.
Contributing to this difference is the possibility for US operators to resort to their own on-site electricity generation systems, an alternative that European environmental standards considerably restrict.
The second factor is the cost of energy. An AI data center with a power of one gigawatt consumes as much electricity as a city of around one million homes. Europe bears some of the highest industrial electricity prices among the major markets for this infrastructure, with energy accounting for approximately 40% of the operating costs of its facilities.
“Power has become the limiting factor for AI data centers. Currently it is not enough to have demand and capital, we must have power and connection to the grid on time. In Iberia we have attractive conditions, such as renewable generation and a strategic geographical position, but they will only translate into investment if the network and permits reach the pace required by the market,” says Pedro Galhardas, senior partner of Roland Berger in Iberia.
Having data centers within European borders makes it possible to process sensitive information and deploy artificial intelligence applications under community jurisdiction
The third obstacle is regulatory complexity. Requirements related to data protection, cybersecurity, and planning and environmental permits add difficulties to the launch of new data centers. Survey participants rank this administrative burden among the main barriers to expanding European capacity.
An opportunity for local suppliers
Despite the difficulties, slower development and the specific requirements of each market can also open opportunities for companies specialized in infrastructure, engineering and construction contractors and suppliers with a local presence. These actors can benefit from the need to adapt projects to the technical, energy and regulatory conditions of each country.
This is a scenario that may be more complex for operators accustomed to working mainly in the United States, where connection times and conditions for deploying additional capacity are different.
Roland Berger believes that political initiatives aimed at expanding digital infrastructure represent an important step forward in strengthening European strategic autonomy, although he warns that there is still some way to go. Having data centers within European borders makes it possible to process sensitive information and deploy artificial intelligence applications under community jurisdiction, while providing the basis for developing a competitive technological ecosystem.
Computing capacity will, therefore, be a determining factor for Europe’s digital future. If it fails to accelerate the construction of infrastructure, facilitate access to energy and reduce administrative obstacles, the region risks losing some of the economic growth associated with AI and increasing its dependence on external suppliers.
