The International Energy Agency estimates that data centers used about 415 terawatt-hours of electricity worldwide in 2024, and that Europe accounted for around 15 percent of that, behind the United States at 45 percent and China at 25 percent. Europe is therefore a significant but smaller market for AI infrastructure than the United States. [1]

Demand is nonetheless rising fast. Energy think tank Ember projected in June 2025 that European data-center electricity demand would grow from 96 terawatt-hours in 2024 to 236 terawatt-hours by 2035, and that growth to 2030 would exceed the additional demand from electric vehicles over the same period. [3]

FLAP-D is industry shorthand for Frankfurt, London, Amsterdam, Paris and Dublin. According to Ember, these five markets hold about 62 percent of Europe's data-center capacity, a share it expects to fall to 55 percent by 2030 and 51 percent by 2035. The main reason is the grid: Ember reports that new facilities in the leading hubs wait on average seven to ten years for a connection, and up to 13 years in some cases. [3]

Commercial data points in the same direction. CBRE's June 2026 global report put colocation vacancy at 5 percent in Frankfurt, 6.7 percent in Paris, 8.6 percent in London and 9.4 percent in Amsterdam, and said grid upgrades in Frankfurt and London were delayed until the early 2030s. [4]

Ireland shows how policy is adapting. In December 2025 its energy regulator, the CRU, decided that new data centers must bring generation or storage, on site or nearby, matching their requested import capacity, and must source at least 80 percent of annual electricity from new Irish renewable projects. The CRU noted that data centers used 22 percent of national electricity demand in 2024, projected to reach 31 percent by 2034. [5]

Ember expects data-center capacity in countries such as Sweden, Denmark and Norway to triple by 2030, helped by uncongested grids and proactive planning by transmission operators such as Statnett and Energinet. Outside FLAP-D as a whole, it projects demand growth of 110 percent to 2030, double the rate in the established hubs. [3]

Large GPU projects illustrate the shift. In July 2025 OpenAI, Nscale and Aker announced Stargate Norway in Narvik, targeting 100,000 Nvidia GPUs by the end of 2026 on 230 megawatts of initial capacity powered by renewable energy, with closed-loop direct-to-chip liquid cooling. In March 2026 Nebius announced a 310-megawatt AI facility in Lappeenranta, Finland, with first capacity for customers in 2027. [6][7]

Cool climates and district heating networks help. Nebius says its Mäntsälä site in Finland feeds server waste heat into the town's heating network, supplying over 30 gigawatt-hours in two years and covering up to 72 percent of local annual heating demand, and reports a PUE of 1.13 at full load. [8]

For buyers, the practical consequence is that the newest and largest European GPU clusters are increasingly located far from the traditional metro hubs. That matters little for training, which is rarely latency-sensitive, but can matter for inference serving users in western European cities and for data that must stay in a particular country.

The IEA expects renewables and nuclear to supply about 85 percent of the additional electricity used by European data centers to 2030, a cleaner mix than in the United States, where it expects natural gas to play the largest role. [2]

Data centers in the EU also face disclosure duties. Under the revised Energy Efficiency Directive and Delegated Regulation 2024/1364, facilities with installed IT power of at least 500 kilowatts must report metrics such as energy consumption, PUE, water use, waste-heat reuse and renewable energy share to a European database; in Sweden, for example, reports on 2025 data were due by 15 May 2026. [9]

Where GPUs sit legally matters as much as where they sit physically. Transfers of personal data from the EU to the United States currently rely on mechanisms such as the EU-US Data Privacy Framework, adopted in July 2023, and standard contractual clauses. The EU General Court upheld the framework in September 2025, but the decision was appealed to the Court of Justice in October 2025, which previously struck down the Safe Harbor and Privacy Shield arrangements. [10]

The EU AI Act entered into force in August 2024. Obligations for general-purpose AI models have applied since 2 August 2025, and according to the European Commission, the AI Omnibus amendments that entered into force in July 2026 set 2 December 2027 as the date for high-risk systems in areas such as critical infrastructure and employment, and August 2028 for AI embedded in regulated products. [11]

Providers have responded with sovereignty-focused offerings. AWS launched its European Sovereign Cloud in Brandenburg in January 2026, run by EU-resident staff and separate from its other regions. Deutsche Telekom and Nvidia announced an Industrial AI Cloud in Munich with up to 10,000 Blackwell GPUs, launching in the first quarter of 2026. Mistral AI raised 830 million dollars of debt in March 2026 for a data center near Paris and targets 200 megawatts of compute across Europe by 2027. [12][13][14]

The EuroHPC Joint Undertaking, which pools EU and national funding for supercomputing, selected its first seven AI Factories in December 2024, in Spain, Italy, Finland, Luxembourg, Sweden, Germany and Greece, backed by about 1.5 billion euros. They are intended to give start-ups, researchers and industry access to AI-optimised supercomputers. [15]

By July 2026 EuroHPC counted 19 operational AI Factories and 13 AI Factory Antennas. On 30 July 2026 it launched a call for up to seven AI Gigafactories in at least seven member states, with public co-funding acting as an anchor customer and more than 20 billion euros of private investment expected. Submissions close on 12 November 2026, with selected projects expected to begin operating within 18 months of selection in early 2027. [16]

Public facilities are allocated through access calls and partnerships rather than hourly self-service, so they complement commercial GPU clouds rather than compete on the same terms. For European start-ups, though, they can be a meaningful additional source of capacity.

Kovara tracks published GPU prices and availability signals across roughly 100 providers, including European clouds and the European regions of global providers. The regions view shows where capacity is listed across European countries, the GPU prices page compares the same accelerator across locations, and Kova can help narrow options for a given data-residency requirement. Buyers who need capacity in a specific country or under specific contractual terms can request a quote to confirm what is actually available.