Does Europe Need Its Own Artificial Intelligence?
By Matthew Parish
Monday 14 September 2026
There is an increasingly fashionable argument in European political circles that the continent has made a grave strategic mistake. Europe invented much of the intellectual architecture upon which modern computing rests, possesses some of the finest universities in the world and remains one of the richest economic regions on Earth. Yet when Europeans wish to use the most sophisticated artificial intelligence systems, they overwhelmingly turn to technologies developed elsewhere, and principally in the United States. The leading frontier systems, cloud platforms and much of the hardware and software infrastructure upon which artificial intelligence depends are American. The European Commission itself now describes dependence upon foreign cloud and AI suppliers as a matter of economic security and strategic autonomy. Its proposed Cloud and AI Development Act observes that three non-EU hyperscalers control more than 70 per cent of the European cloud market.
At first sight this looks alarming. Artificial intelligence is rapidly becoming something more consequential than another category of computer software. It is becoming an intellectual infrastructure through which businesses make decisions, governments analyse information, scientists undertake research and citizens obtain knowledge. If artificial intelligence develops in the direction that currently seems probable, AI systems will increasingly write software, design machines, analyse intelligence, operate financial systems and assist in military planning. A continent that imports this infrastructure might therefore appear analogous to a country that imports all her electricity, telecommunications equipment or military aircraft. Dependence upon another power for something so fundamental inevitably acquires geopolitical significance.
Nevertheless, the argument is more complicated than the rhetoric of “AI sovereignty” sometimes suggests. Europe certainly has an AI problem. It does not necessarily follow that Europe has an American AI problem. Indeed, an attempt to disentangle these two propositions reveals a much broader question about what technological sovereignty actually means in a world in which almost every advanced technology is constructed through extraordinarily complicated international supply chains.
The first difficulty with the European anxiety is that reliance upon American technology is hardly new. European businesses have depended upon American operating systems, databases, microprocessors, cloud services, search engines and enterprise software for decades. Microsoft, Google, Amazon, Apple, Oracle and innumerable other American technology companies became deeply embedded within European economic life without thereby transforming Europe into an American colony. Europeans purchased these products because they were good, and American companies became rich because hundreds of millions of Europeans voluntarily decided that purchasing American technology was preferable to purchasing inferior domestic alternatives.
Artificial intelligence may simply be another manifestation of the same phenomenon. The United States possesses an extraordinary ecosystem for financing extremely speculative technology companies. American universities attract international scientific talent, Silicon Valley has immense concentrations of venture capital and American capital markets permit successful technology companies to raise sums almost unimaginable elsewhere. Frontier artificial intelligence is exceptionally expensive. It requires enormous quantities of computing power, specialised chips, electricity, data centres, engineers and capital. Europe has traditionally been much less successful at combining these things quickly.
The difference in scale remains striking even as European AI companies grow. France’s has emerged as Europe’s most conspicuous frontier-model company and in September 2026 raised €3 billion at a valuation of approximately €21 billion. That is an extraordinary European technological success. Yet the amounts of capital being deployed by the largest American AI companies remain dramatically greater. Europe is therefore not technologically barren. Rather, she has difficulty converting scientific excellence into gigantic corporations capable of sustaining the capital expenditures required at the technological frontier.
This distinction matters because technological dependence has several different meanings that are too often bundled together. Europe might worry that she cannot produce frontier AI models. She might worry that European data are processed on foreign infrastructure. She might worry that European companies pay economic rents to American corporations. She might worry that the United States government could compel American companies to provide information about European customers. Or she might worry that in some future geopolitical confrontation Washington could simply prevent Europeans from obtaining advanced artificial intelligence at all. These are different problems and they require different solutions.
The last is the genuinely geopolitical concern. Europe and the United States are allies, but alliances are not immutable facts of nature. American foreign policy can change dramatically between administrations and European governments have increasingly discovered that economic interdependence may become political leverage. Washington controls important financial, technological and military systems upon which European states depend. Artificial intelligence might eventually join this list. The European Commission now explicitly argues that asymmetric technological dependencies can become geopolitical vulnerabilities when export controls or extraterritorial legal powers are employed.
Imagine, for example, a future European foreign policy strongly opposed by Washington. If essential European government departments, defence contractors, banks and industrial companies depended upon American AI services, the theoretical possibility of their restriction would become a source of American influence even if no American President ever actually threatened to switch them off. Dependency creates leverage because the possibility of interruption alters calculations before interruption occurs.
There is also the question of intelligence and data. AI systems will increasingly process extraordinarily sensitive information. Governments may use them to analyse diplomatic cables, military logistics or intelligence reports. Companies may provide them with trade secrets, legal strategies, engineering designs and financial information. Individuals already provide conversational AI systems with intimate details about their professional and private lives. Where models, computing infrastructure and corporate ownership all fall under another country’s jurisdiction, legitimate questions arise about the ultimate location and control of this information. The proposed European Cloud and AI Development Act specifically identifies the extraterritorial reach of third-country laws as one of the risks associated with foreign technological dependence.
Yet from this it does not follow that Europe must reproduce the entire American AI industry behind a digital Maginot Line. That would be extraordinarily expensive and probably impossible. Modern technology does not respect such neat national boundaries. Even an ostensibly European AI model may depend upon American-designed chips fabricated in Taiwan using Dutch lithography equipment, Asian memory, American software tools, European electricity infrastructure and training data drawn from every corner of the internet. Sovereignty understood as complete technological autarky is therefore largely an illusion.
There is another danger. If Europe becomes excessively preoccupied with ensuring that Europeans use European AI, it may inadvertently impose inferior technology upon her it citizens and companies. This would be a peculiar method of improving European competitiveness. If an American model is substantially better at designing pharmaceuticals, writing software or optimising industrial production than a European equivalent, requiring a European company to use the inferior system would make Europe poorer rather than more sovereign. The geopolitical benefits of technological independence would have been purchased by imposing a productivity tax upon the entire European economy.
This dilemma is already visible in debates about European defence technology. Some European defence officials have reportedly resisted proposals that might restrict reliance upon American cloud and AI suppliers because European alternatives are not yet sufficiently capable for certain sensitive applications and because interoperability with NATO systems matters. This is an important warning against transforming technological sovereignty into technological protectionism. A sovereign Europe that deliberately equips herself with worse technology may discover that she has achieved sovereignty principally in the right to become weaker.
There is nevertheless a sensible middle course between complete dependence and impossible autarky. Europe does not need to possess the world’s largest AI company. She needs credible alternatives. The distinction resembles military deterrence. A country does not necessarily need to dominate every category of weapon; she needs sufficient independent capability that nobody can credibly threaten to deprive her of something indispensable.
This is where Europe’s emerging AI strategy makes considerably more sense. The European Commission’s AI Continent Action Plan envisages large public and private investments in computing infrastructure, including AI Factories and up to five much larger AI Gigafactories. The Commission says that 19 AI factories are now deployed and its broader InvestAI programme is intended to mobilise €200 billion of investment, including €20 billion for the proposed Gigafactories. Whether all these ambitions will be realised efficiently is another question, but the strategic objective is intelligible: Europe needs enough computing infrastructure that European researchers and companies are not permanently dependent upon foreign hyperscalers merely to experiment with advanced models.
European companies are also beginning to develop the other half of this insurance policy. Mistral is expanding European computing infrastructure and regional inference capabilities, presenting the ability of customers to control where their models and data operate as a component of technological sovereignty. Smaller national initiatives are appearing as well; Portugal, for example, launched an open-source foundation model in July 2026 partly as an exercise in reducing dependence upon foreign AI technologies.
Open models may ultimately prove especially important for Europe. If governments and companies can download, modify and operate capable models upon infrastructure under their own control, the geopolitical significance of the nationality of the original developer diminishes. The strategic question becomes less “Who invented this model?” and more “Can we operate, inspect, modify and replace it without somebody else’s permission?” This is a considerably more sophisticated definition of technological sovereignty.
Europe’s regulatory instincts are more ambiguous. The European Union was extraordinarily quick to regulate artificial intelligence through the AI Act, including specific obligations governing general-purpose models and additional requirements for models deemed to present systemic risks. There are respectable reasons for these rules. Europeans are entitled to insist that powerful technologies operating within their societies comply with European conceptions of privacy, safety, transparency and fundamental rights. Regulation is itself an expression of sovereignty.
But regulation cannot substitute for innovation. Europe cannot regulate herself into technological leadership any more than the Soviet Union could decree herself into possessing Silicon Valley. If compliance costs disproportionately burden small European companies while enormous American corporations can employ armies of lawyers, regulators may perversely reinforce the dominance they hope eventually to diminish. The correct European objective must therefore be to regulate catastrophic or socially unacceptable risks while making it dramatically easier to finance, build and expand AI companies within Europe.
There is also an uncomfortable historical irony. Europe possesses many of the ingredients necessary for technological leadership. Britain, France, Germany, Switzerland, the Netherlands and the Nordic countries contain outstanding universities and scientific institutions. European engineers are highly skilled. The Netherlands possesses one of the most strategically important companies in the semiconductor industry. Europe has enormous pools of private savings and some of the world’s largest industrial companies. What Europe frequently lacks is not intelligence or money but institutional willingness to tolerate the disorder associated with innovation.
Silicon Valley produces spectacular successes partly because it produces spectacular failures. Venture capitalists finance improbable ideas. Engineers move between companies. Founders become extraordinarily wealthy. Businesses collapse. Technologies appear before governments have decided precisely how they ought to be regulated. European political culture tends to prefer stability, predictability and precaution. These are admirable qualities for administering mature societies, but they are not necessarily the qualities that create revolutionary industries.
Hence Europe’s real geopolitical disadvantage may not be that Europeans use American artificial intelligence. It may be that Europeans have constructed economic institutions under which the easiest path for a brilliant European AI researcher is to move to California, the easiest path for a successful European technology company is to raise American capital and the easiest path for a European corporation seeking sophisticated AI is simply to purchase it from an American supplier.
If that is the problem, banning or discouraging American AI would address the symptom while worsening the disease. Europe should instead make herself a place in which European AI companies can grow to enormous scale. That means abundant computing capacity, cheaper energy, deeper capital markets, easier movement of technical talent, sensible immigration rules, competitive taxation, research universities connected to commercial capital and regulation sufficiently predictable that entrepreneurs can understand the rules before spending billions of euros.
Even then Europe should continue using American artificial intelligence. The objective should not be separation but redundancy. European companies should be able to choose between American, European and perhaps other trustworthy systems. Governments should maintain independent capabilities for particularly sensitive functions. Critical infrastructure should not depend upon a single foreign supplier. European data should be capable of remaining within European jurisdiction where national security requires it. Open models and interoperable systems should make switching providers technically possible rather than prohibitively expensive.
This is analogous to energy security. A country does not become insecure merely because she imports energy. She becomes insecure when she imports an indispensable resource from a supplier that cannot readily be replaced. Diversification, reserves, domestic production and alternative infrastructure transform dependency into ordinary trade. Europe should think about artificial intelligence in much the same way.
There is therefore a genuine geopolitical issue here, but it is sometimes overstated. Europe is not becoming an American technological protectorate simply because Europeans use ChatGPT, American cloud computing or American-designed semiconductors. Voluntary technological exchange between allied economies is ordinarily a source of wealth rather than weakness. Indeed, excluding the world’s best technology merely because it is foreign would represent an extraordinary act of economic self-harm.
The danger arises when reliance becomes irreplaceable. Artificial intelligence may eventually become sufficiently important that no major political power can prudently allow herself to possess no independent capacity whatsoever. Europe therefore needs models, computing infrastructure, technical expertise and companies sufficiently capable to provide strategic alternatives should relations with the United States deteriorate or American technology become temporarily unavailable.
But Europe does not need to win the artificial intelligence race in order to achieve this. Perhaps there is not even a single race to win. Artificial intelligence will probably become an ecosystem containing frontier models, specialised models, open models, autonomous agents, enormous data centres and millions of applications embedded throughout ordinary economic life. Europe can occupy important parts of this ecosystem without reproducing Silicon Valley in its entirety.
The sensible European ambition is consequently neither technological nationalism nor complacent dependence. Europe should continue buying the best American AI where it makes Europeans richer and more productive while investing aggressively in sufficient European capability that it always has another choice. Genuine sovereignty does not mean making everything oneself. It means retaining the practical freedom to make one’s own decisions.
On that definition, Europe presently has an AI sovereignty problem, but not yet an AI sovereignty crisis. The distinction is important. The solution is not to build walls around European technology. It is to ensure that if one day somebody else builds a wall, Europe is capable of functioning on its own side of it.
