In 2025, the whole of Luxembourg used 6.53 terawatt-hours of electricity.¹ That includes every home, office, factory, and streetlight in the country. A single one-gigawatt data center running around the clock for a year would use 8.76.
One building would use more power than a whole (small) country.
AI has become remarkably good at producing numbers like these. Billions of dollars for new data centers. Hundreds of thousands of GPUs. Gigawatts of electricity. Valuations that would have looked absurd only a few years ago.
Europe is part of that expansion. And yet something in the European AI story does not quite fit together.
The familiar diagnosis says Europe cannot compete: electricity is too expensive, and regulation makes large projects too difficult. Meanwhile, data centers keep going up across the continent, and demand for AI capacity is rising fast.²
Both things can be true at once. But if energy and regulation are the main explanation for the gap, they should leave a clear trace in the numbers. So we went looking for that trace, starting with electricity.
Electricity is a problem. Yes, but …
The price
The European electricity market has one feature that sounds almost designed to make cheap power expensive. Producers offer electricity at different prices. The cheapest sources are used first, and more expensive ones are added until demand is covered. The last source needed can then set the wholesale price for everyone in that market period. If that last source is a gas plant, gas can shape the price even when most of the electricity came from cheaper nuclear, wind, or solar.³
The mechanism is real. The popular version of it is too simple. Prices vary by country and by hour, large industrial users can sign long-term contracts, and the final bill includes national network charges, taxes, and levies. The EU’s 2024 market reform was partly designed to reduce companies’ exposure to short-term wholesale prices by encouraging more of those long-term contracts.³
The units matter here. A gigawatt is 1,000 megawatts, and a large nuclear reactor produces roughly one gigawatt of electrical power.⁴ CBRE expects Europe’s data-center market to reach about 13 gigawatts of capacity by the end of 2026.² That figure describes capacity, not continuous consumption. In power terms, it is roughly the output of thirteen large nuclear reactors.
So are European electricity prices simply too high? In some countries, yes. Eurostat reported non-household prices in the second half of 2025 ranging from €7.48 per 100 kWh in Finland to €25.52 in Ireland, with Germany also near the top.⁵ These are not hyperscale data-center tariffs. They do make one point very clearly: Europe does not have one electricity price.
Then another number changes the picture. Epoch AI estimates that for a data center with one gigawatt of IT capacity, electricity represents around 7 percent of the annualized cost, while servers account for about 60 percent.⁶ Epoch’s model assumes a US hyperscaler paying US industrial rates of 8.34 cents per kWh, below what businesses pay in much of Europe, so the European share would be higher. Even so, the arithmetic is instructive. At those rates, doubling the electricity bill adds roughly 7 percent to the total cost of compute.
The access
The more serious energy problem may be access. The International Energy Agency estimates that a grid connection in the EU can take between two and ten years, depending on the country. In the five main European hubs of Frankfurt, London, Amsterdam, Paris, and Dublin, queues average seven to ten years.⁷ The data center itself can often be built in one or two.
Energy has not disappeared as a problem. It has become a more complicated one.
What about the regulations?
Some restrictions are very clear. Ireland now requires new data centers seeking a grid connection to provide generation or storage matching their maximum requested demand, and to cover at least 80 percent of their annual electricity use with additional renewable projects in Ireland.⁸
At the EU level, data centers with at least 500 kW of installed IT power must report on energy, water, and other sustainability indicators.⁹ Providers of general-purpose AI models face requirements on technical documentation, copyright policy, and a public summary of training content, with additional obligations for models considered to pose systemic risk.¹⁰ None of this makes an AI company impossible to build. All of it takes time, people, and money.
If regulation were the main explanation, however, we would expect the construction numbers to look worse. At Sines, in Portugal, Nscale is building a second 200 MW facility for Microsoft and plans to deploy more than 66,000 NVIDIA Rubin GPUs from late 2027.¹¹ Across Europe, CBRE expects data-center capacity to grow by about 20 percent in 2026, while new colocation contracts intended for AI more than quadrupled in the first half of the year.²
Regulation can stop a project or send it elsewhere. It can also coexist with very large projects being built.
Other rules sometimes get added to the list, but they fit it less well. EU state-aid law restricts selective government support,¹² yet the EU is using public money to back AI gigafactories.¹³ VAT is normally deductible for VAT-registered businesses, so it creates administration and cash-flow effects rather than a permanent cost at every step.¹⁴ Anti-dumping rules concern goods imported from outside the EU below their normal value, not a cloud company selling compute cheaply.¹⁵
The regulatory burden is real. It simply cannot carry the whole explanation.
The numbers tell an untold story
The figure that changes the direction of the inquiry is market share.
Europe’s cloud infrastructure market reached €61 billion in 2024. European providers held about 15 percent of it. Amazon, Microsoft, and Google together held around 70 percent, while SAP and Deutsche Telekom had roughly 2 percent each.¹⁶ Those are the two largest European providers.
A port can be busy without owning the shipping lines. Europe increasingly looks like that port. The buildings are going up and the customers are buying, but much of the commercial relationship belongs to companies headquartered elsewhere.
Mistral offers a clue to what may be missing. The French company wants to build up to one gigawatt of European compute by 2030. More interesting is how. It has gathered European companies willing to make multi-year commitments for future capacity through what it calls European Compute Units. Those commitments give investors something valuable before a single machine is installed: customers.¹⁷
The EU has arrived at almost the same mechanism. The AI gigafactories announcement speaks of sovereignty and strategic autonomy. The same text also describes the public contribution in far more practical terms: it will act as an “anchor customer,” reducing risk for private investors and helping unlock more than €20 billion of private capital.¹³
Once that piece is visible, the financing problem makes more sense. Amazon, Microsoft, and Google already have customers, cloud platforms, sales organizations, and enormous balance sheets. A European challenger may have to find the customers, raise the capital, and build the infrastructure, all at the same time.
Europe does not lack money in aggregate. On September 14, Christine Lagarde noted that euro-area households save around €1.4 trillion a year and hold about €440 billion in American technology companies. She also argued that the euro area relies far more heavily on banks, while the American AI buildout draws on equity, bonds, private credit, and securitization.¹⁸ That does not prove capital is the single cause. It fits Mistral’s approach closely.
Europe hosts and that’s it
We started with two explanations that still look partly right. Energy can make a project unattractive, and regulation can delay it or add cost. In some European countries, both problems are serious, and grid access may matter even more than the headline electricity price.
But neither explanation accounts for the commercial result on its own. Europe is building data centers. European companies are signing up for AI capacity. The cloud market is worth €61 billion a year. What Europe has not yet built at comparable scale is a company that controls enough of the customer relationship, the capital, and the infrastructure at the same time.
If the test is whether Europe can host AI infrastructure, Europe has not failed. If the test is whether Europe has created AI and cloud businesses at the scale of the American leaders, the conclusion is harder to avoid. Because, so far, Europe has failed that second test.
Europe is now running two experiments in parallel. Public money is acting as an anchor customer through the gigafactories, while companies like Mistral are trying to build the same confidence through commercial commitments.
By 2028, the most revealing number may not be how many gigawatts Europe has installed but how many of those gigawatts are sold by a European company, to customers who chose it because it was the best supplier.
References
- Institut Luxembourgeois de Régulation, “Données statistiques nationales du secteur de l’électricité de l’année 2025,” June 30, 2026. https://www.ilr.lu/lb/publications/donnees-statistiques-nationales-du-secteur-de-lelectricite-de-lannee-2025/
- CBRE, “European Data Centres Figures Q2 2026,” August 17, 2026. https://www.cbre.com/insights/figures/european-data-centres-figures-q2-2026
- European Commission, “Electricity Market Design,” accessed September 28, 2026. https://energy.ec.europa.eu/topics/markets-and-consumers/electricity-market-design_en
- International Atomic Energy Agency, IAEA Nuclear Energy Series, representative net electrical output of 1,000 MW for a large light-water reactor. [Publication year to confirm] https://www-pub.iaea.org/MTCD/Publications/PDF/Pub1598_web.pdf
- Eurostat, “Non-household electricity prices in 2nd half of 2025: -3.5%,” May 8, 2026. https://ec.europa.eu/eurostat/en/web/products-eurostat-news/w/ddn-20260508-2
- Amelia Michael and Ben Cottier, “Servers account for 60% of the total cost of ownership of a one-gigawatt AI data center,” Epoch AI, May 14, 2026. https://epoch.ai/data-insights/ai-datacenter-cost-breakdown
- International Energy Agency, “Overcoming energy constraints is key to delivering on Europe’s data centre goals,” November 16, 2025. https://www.iea.org/commentaries/overcoming-energy-constraints-is-key-to-delivering-on-europe-s-data-centre-goals
- Commission for Regulation of Utilities, “The CRU Publishes its Decision on New Electricity Connection Policy for Data Centres,” December 12, 2025. https://www.cru.ie/about-us/news/the-cru-publishes-its-decision-on-new-electricity-connection-policy-for-data-centres/
- European Commission, Commission Delegated Regulation (EU) 2024/1364 on the first phase of the establishment of a common Union rating scheme for data centres, March 14, 2024. https://eur-lex.europa.eu/eli/reg_del/2024/1364/oj/eng
- European Commission, “General-purpose AI obligations under the AI Act,” accessed September 28, 2026. https://digital-strategy.ec.europa.eu/en/factpages/general-purpose-ai-obligations-under-ai-act
- Nscale, “Nscale to Deliver 66,000+ NVIDIA Rubin GPUs to Microsoft at Start Campus’ Site in Portugal,” May 5, 2026. https://www.nscale.com/press-releases/nscale-start-campus
- European Commission, “State Aid Overview,” accessed September 28, 2026. https://competition-policy.ec.europa.eu/state-aid/overview_en
- EuroHPC Joint Undertaking, “The EuroHPC Joint Undertaking launches the AI Gigafactories Call,” July 30, 2026. https://www.eurohpc-ju.europa.eu/eurohpc-joint-undertaking-launches-ai-gigafactories-call-2026-07-30_en
- European Commission, “VAT for businesses,” accessed September 28, 2026. https://taxation-customs.ec.europa.eu/taxation/vat/vat-businesses_en
- European Commission, “Anti-dumping measures,” accessed September 28, 2026. https://policy.trade.ec.europa.eu/enforcement-and-protection/trade-defence/anti-dumping-measures_en
- Synergy Research Group, “European Cloud Providers’ Local Market Share Now Holds Steady at 15%,” July 24, 2025. https://www.srgresearch.com/articles/european-cloud-providers-local-market-share-now-holds-steady-at-15
- Mistral AI, “In-region inference, open models, and new European infrastructure for sovereign AI,” August 11, 2026. https://mistral.ai/news/regional-inference-open-models-new-compute/
- Christine Lagarde, “A new age of capital: growth, sovereignty and AI,” European Central Bank, September 14, 2026. https://www.ecb.europa.eu/press/key/date/2026/html/ecb.sp260914_2~a3f0efbee4.en.html
