https://x.com/i/article/2094760918831751168

A fact-based critique of the EU’s attempt to become the "first AI continent."

A recent European Commission post sparked a rather heated debate on X. The Commission had designated ChatGPT as a Very Large Online Search Engine under the Digital Services Act after its search service surpassed 45 million monthly users in the EU. Rather than leave the argument at the level of slogans, I decided to look at the numbers. Is Europe genuinely over-regulating AI? And, more fundamentally, how much influence does it still have over the technology it is trying to govern?

On 1 August 2024, the Council of the European Union posted a celebratory graphic. The AI Act had entered into force. Brussels described it as the first legislation of its kind and said Europe could set a global standard for AI regulation.

Two years later, Ursula von der Leyen announced that Europe wanted to become the “first AI Continent”. Up to €10 billion in EU and national funding, she wrote, would attract at least €20 billion in private investment for as many as seven AI gigafactories.

X attached a Community Note comparing that figure with the far larger sums spent by US technology companies.

Then, on 31 August 2026, the European Commission designated ChatGPT as a Very Large Online Search Engine under the Digital Services Act. Reddit and Roblox became Very Large Online Platforms. All three now have four months to comply with additional obligations.

Taken together, the posts capture Europe’s AI problem with unusual clarity. The rules are in force. The compute is still in procurement.

In this essay, I want to explore what I believe to be the problems facing Europe. On the one hand, the EU projects to the outside world that it is on its way to becoming an "AI first continent," but on the other hand, this image collapses when one looks at the figures. Therefore, I want to analyze the true situation and formulate a critique. This is especially important because a heated discussion about this topic recently erupted on X.

Europe’s capital does not reach European AI

Europe is not poor. It has deep pools of private savings, world-class universities and some excellent AI companies. What it lacks is a financial system willing and able to fund technological risk at American scale.

The OECD’s 2025 venture-capital data are brutal and crystal clear. US-headquartered AI companies attracted $194 billion, or 75% of global AI VC deal value. EU-27 companies received $15.8 billion, or 6%. China received $13.9 billion, although private VC figures capture only part of a system in which the state directs large amounts of capital through government funds, banks and industrial policy. (Source: https://www.oecd.org/en/about/news/announcements/2026/02/ai-firms-capture-61-percent-of-global-venture-capital-in-2025.html)

The gap grows as companies mature. The European Central Bank says Europe’s bank-centred financial system is poorly suited to young AI firms with high upfront costs, uncertain revenue and intangible assets that cannot be used as normal collateral. Nearly 30% of the unicorns founded in Europe between 2008 and 2021 relocated abroad, mostly to the United States. Founders cited capital, access to a large unified market and simpler regulation ( https://www.ecb.europa.eu/press/key/date/2026/html/ecb.sp260323_1~1e06784a89.en.html)

The gigafactory announcement needs the same reading. The EU is offering up to €10 billion in public support and expects that to attract at least €20 billion privately. It is a future, multi-year financing package. The International Energy Agency estimates that five large technology companies spent more than $400 billion in 2025 and could increase capital expenditure by another 75% in 2026, roughly $700 billion in a single year. Not all of that is AI spending, and the numbers are not directly comparable. But their different scope is part of the point: Europe announces a blended financing target while US firms place annual infrastructure orders on a different scale. (https://www.iea.org/reports/key-questions-on-energy-and-ai/executive-summary)

China is operating on another level too. Alibaba alone announced at least RMB 380 billion, about $53 billion, for AI and cloud infrastructure over three years. One Chinese company’s programme is nominally larger than the EU’s entire expected public-private gigafactory package. (https://www.alibabagroup.com/en-US/document-1830678592242057216)

AI has an electricity bill

Brussels often talks about AI as software, data and regulation. A frontier training cluster is more like heavy industry. It needs land, transformers, cooling, grid connections and a esepecially and foremost dependable flow of electricity every hour of the year.

The IEA estimates that global data-centre electricity demand rose to about 485 TWh in 2025 and will reach roughly 950 TWh by 2030. Demand from AI-focused facilities is expected to triple.

Converted into average continuous load, the IEA figures imply that US data centres will draw roughly 49 GW by 2030, compared with about 32 GW in China and just over 12 GW in Europe.

Europe enters that buildout with industrial electricity prices that remain more than twice US levels and natural-gas prices that have been close to four times higher, according to Mario Draghi’s competitiveness work (https://commission.europa.eu/document/download/0951a4ff-cd1a-4ea3-bc1d-f603decc1ed9_en). (https://www.iea.org/reports/key-questions-on-energy-and-ai/executive-summary)

[As an aside: Europe, and Germany in particular, decided early on to largely phase out nuclear energy. The intention was to rely on renewable energies, but increasingly, cheap gas from Russia was needed to meet energy demands. The fact that the gas came cheaply from Russia was helpful. This changed with the start of the war in Ukraine. Attempts were then made to diversify by importing gas from Qatar, but this too became problematic after the war with Iran. Currently, most gas is imported, largely as expensive LNG from the USA, which is not only detrimental to the energy situation but actually exacerbated.]

Renewables are part of the answer. Cheap wind and solar can supply a substantial share of new demand, and batteries are useful for smoothing short power fluctuations. They do not by themselves provide a 24/7 electricity system. Data centres still need transmission, storage, backup capacity and firm generation. Europe’s argument over whether one technology is morally preferable to another has consumed years that competitors spent connecting capacity.

China had 36 nuclear reactors under construction at 19 sites in May 2026, representing 38.9 GW and more than 49% of global reactor construction.

The EU-27 does not have zero, as critics sometimes claim. It has two active projects: Paks 5 in Hungary and Mochovce 4 in Slovakia That correction hardly rescues the European position. Mochovce is a long-delayed legacy project. Paks II is led by Rosatom and carries Russian technology and fuel dependencies. (https://www.eia.gov/todayinenergy/detail.php?id=67746) (https://paks2.hu/web/paks-2-en/news) (https://www.ujd.gov.sk/ujd-sr-authorizes-the-commissioning-of-mo4/?lang=en)

China builds reactors in standardised batches of six to ten and has developed a domestic supply chain for major components. Its average construction time has been about six years, against a global average closer to nine. Europe holds conferences about energy sovereignty while one of its only new reactor projects depends on Russia.

The 2026 Middle East crisis offered another warning. Europe has sharply reduced its dependence on Russian pipeline gas, but much of that dependence shifted to the global LNG market (see above). An EU Council analysis estimated that the first 28 days of the conflict added €13 billion to the Union’s fossil-fuel import bill. Diversifying suppliers reduced the danger of a single country closing the tap. It did not insulate European electricity prices from Asia, shipping routes or war.

(https://data.consilium.europa.eu/doc/document/WK-4807-2026-INIT/en/pdf)

Europe produces research. Others turn it into frontier systems

The Stanford AI Index 2026 counted 59 notable AI models released by US organisations in 2025, 35 by Chinese organisations and two by European ones. Stanford’s list does not include every model ever released. But with 59 notable models from the US and only two from Europe, the overall picture is still hard to dismiss.

(https://hai.stanford.edu/assets/files/ai_index_report_2026_chapter_1_research_development.pdf)

China is also closing the quality gap. By March 2026, Stanford measured only a 2.7 percentage-point difference between the strongest US and Chinese models in its comparison (And we all know how good Chinese open source models like GLM, Qwen, and DeepSeek have become.). Europe has Mistral, Black Forest Labs, DeepL and Aleph Alpha. But in direct comparison with Chinese and US models, they are insignificant. It has researchers, open-source contributors and industrial customers. It does not have a comparable concentration of frontier labs, compute and distribution.

Distribution may be the least discussed dependency. AWS, Microsoft Azure and Google Cloud control about 69% of European cloud infrastructure. European providers such as OVHcloud and Deutsche Telekom account for roughly 13%, according to the European Parliament. A data centre can stand in Frankfurt, Paris or Dublin and still strengthen an American platform.

(https://www.europarl.europa.eu/doceo/document/A-10-2025-0107_EN.html).

ASML however shows both Europe’s strength and the limit of its sovereignty. The Dutch company is the only commercial supplier of the EUV lithography machines needed to manufacture the most advanced chips. It reported €32.7 billion in 2025 sales and spent €4.7 billion on R&D. Yet the leading AI chips are mostly designed in the United States, fabricated in Taiwan, combined with memory from a small Asian and American supplier group, and deployed through US clouds. Europe owns the most important machine in the factory. Much of the factory is elsewhere.

Regulation arrived before industrial capacity

The AI Act did not create this gap. Thats true and clear. Europe was losing technology companies, frontier investment and cloud market share long before the law entered into force. Blaming a single regulation for the entire problem would be lazy.

Companies, however, do not comply with one law at a time. They face GDPR, the Digital Services Act, the Digital Markets Act, the Data Act, copyright rules, national regulators and sector-specific product law alongside the AI Act. The Draghi report counted roughly 100 technology-focused laws and more than 270 regulators across EU Member States. (https://commission.europa.eu/document/download/97e481fd-2dc3-412d-be4c-f152a8232961_en)

Second, the EU’s regulatory stance towards tech companies hampers innovation: the EU now has around 100 tech-focused lawsxi and over 270 regulators active in digital networks across all Member States. Many EU laws take a precautionary approach, dictating specific business practices ex ante to avert potential risks ex post. For example, the AI Act imposes additional regulatory requirements on general purpose AI models that exceed a pre-defined threshold of computational power – a threshold which some state-of-the-art models already exceed. Third, digital companies are deterred from doing business across the EU via subsidiaries, as they face heterogeneous requirements, a proliferation of regulatory agencies and “gold plating”04 of EU legislation by national authorities. Fourth, limitations on data storing and processing create high compliance costs and hinder the creation of large, integrated data sets for training AI models. This fragmentation puts EU companies at a disadvantage relative to the US, which relies on the private sector to build vast data sets, and China, which can leverage its central institutions for data aggregation. This problem is compounded by EU competition enforcement possibly inhibiting intra-industry cooperation. (Mario Draghi, report, ebd.)

Compliance behaves like a fixed cost. Microsoft, Google and Meta can maintain large legal, safety and documentation teams across a global revenue base. A European model company trying to finance its next training run cannot. Rules written to restrain dominant platforms can end up protecting them from smaller challengers.

There is evidence beyond founder complaints. A 2025 NBER working paper found a significant decline in US investment activity toward EU ventures after GDPR, especially for newer and data-related companies. The paper does not prove that privacy regulation is bad or that every lost deal would have produced a European champion. It does show that regulation changes where capital goes. (https://www.nber.org/papers/w33909)

The Commission has started to adjust. Its 2026 AI Omnibus delayed important high-risk provisions, expanded regulatory sandboxes and extended simplified treatment to small mid-cap companies. Sensible regulation can replace 27 national rulebooks, protect citizens and set minimum standards for dangerous systems. Europe still chose an extraordinary sequence: it constructed the compliance regime before securing the energy, compute and risk capital needed by the companies expected to comply with it. (https://digital-strategy.ec.europa.eu/en/news/ai-omnibus-enters-force)

The ChatGPT designation under the DSA is not evidence of persecution. The service reported at least 45 million average monthly EU users and crossed a threshold written into the law. Requirements to assess illegal content, risks to minors, elections and public security are defensible. The uncomfortable fact is that Europe has again become the regulator of a mass-market technology built somewhere else.

France proves that the decline is optional

France has begun to assemble a different proposition. Its nuclear fleet provides large amounts of low-carbon electricity. The government has prepared sites and grid connections. In June 2026, SoftBank announced €45 billion for AI infrastructure in France, potentially rising to €75 billion, with a commitment to develop 3–5 GW of dedicated data-centre capacity. (https://www.elysee.fr/en/emmanuel-macron/2026/06/01/meeting-with-masayoshi-son-ceo-of-softbank)

That is a genuine European success. It also relies on Japanese capital, US-designed accelerators and a cloud market dominated by US companies. Hosting the infrastructure is valuable. Controlling the stack is better.

Europe still has enough assets to change direction: ASML, nuclear expertise, strong universities, industrial companies, a large market and substantial savings. Converting those assets into AI capacity will require cheaper firm power, faster grid connections, deeper capital markets, larger late-stage funds and fewer overlapping compliance layers. None of those can be replaced by another declaration of technological sovereignty.

Europe can keep buying the chips, renting the cloud, importing the energy and regulating the products that result. That is a coherent administrative model. It is not technological leadership.

Taken together, the numbers do not support the Commission’s portrayal of Europe as a future global leader in AI. Ursula von der Leyen’s ambition to turn Europe into the “first AI continent” may be a political goal, but on the current trajectory it is not a credible forecast. The decisions of recent decades – particularly in energy policy – have contributed to high industrial electricity prices and continued dependence on imported energy. Those weaknesses become even more consequential in artificial intelligence, where reliable electricity and computing infrastructure are basic industrial inputs.

Europe has not stopped training advanced models altogether, and claiming that no frontier model is being developed here would be too absolute. Mistral remains a serious European AI company. But it is not currently setting the global general-purpose frontier. Stanford counted only two notable models released by European organisations in 2025, compared with 59 from the United States and 35 from China. Independent benchmarks currently place proprietary models from US companies at the top, while Chinese labs dominate the leading group of open-weight models.

Europe is increasingly squeezed between two technological ecosystems while remaining dependent on both. Its reliance on the United States extends across advanced AI chips, cloud platforms and, increasingly, LNG: the United States supplied 53% of EU LNG imports in 2025. Europe’s relationship with China is different but no less consequential. China was Germany’s largest overall goods-trading partner and its most important source of imports in 2025. However, it was only the sixth-largest destination for German exports, so the dependency now lies as much in supply chains, electronics, batteries and industrial inputs as in access to the Chinese market.

Europe is not incapable of changing course. It still possesses major assets, from ASML and nuclear expertise to strong universities, industrial companies and deep private savings. But it is not currently on track to become the AI continent its leaders describe. The first phase of the race has already passed Europe by, and closing the gap will require far more than political declarations.

That is what makes the EU’s habit of presenting regulatory firsts as evidence of progress so revealing. Regulation may be necessary. But writing the rules for technologies developed, financed and operated elsewhere is not technological leadership.