Reality Check: Can European AI Achieve 100% Sovereignty?

Introduction

The question of whether European artificial intelligence can achieve complete sovereignty has become one of the most consequential strategic debates shaping the continent’s technological and economic future. As the European Union launches ambitious initiatives like the €200 billion InvestAI program, the Apply AI Strategy and a network of AI Gigafactories, European policymakers increasingly frame AI sovereignty as essential to the bloc’s autonomy, competitiveness, and security. Yet beneath the rhetoric of digital independence lies a complex web of dependencies that spans the entire AI technology stack, from semiconductors and rare earth elements to cloud infrastructure and specialized talent. This analysis examines whether 100% AI sovereignty is achievable for Europe, what the geopolitical and market realities reveal and what forms of strategic autonomy might actually be attainable.

The Sovereignty Imperative and Its Limits

European institutions have explicitly positioned AI sovereignty as a strategic priority. The European Commission’s Apply AI Strategy, launched in October 2025, emphasizes that “it is a priority for the EU to ensure that European models with cutting-edge capabilities reinforce sovereignty and competitiveness in a trustworthy and human-centric manner”. This push reflects genuine vulnerabilities. A European Parliament report estimates that the EU relies on non-EU countries for over 80% of digital products, services, infrastructure and intellectual property. In the AI domain specifically, Europe accounts for just 4% of global computing power deployed for AI, while US cloud providers control 65-72% of the European cloud market. The continent produced only three notable AI models in 2024 compared to 40 from the United States and 15 from China.

The continent produced only three notable AI models in 2024 compared to 40 from the United States and 15 from China.

These statistics underscore a stark reality: Europe begins its sovereignty pursuit from a position of profound dependence across multiple layers of the AI stack. The European approach fundamentally differs from the US model, which combines massive private investment with selective export controls to maintain competitive advantage. It also differs from China’s state-directed strategy that mobilizes resources at scale to achieve technological self-sufficiency despite Western restrictions. Europe’s challenge involves not merely closing a capability gap but doing so while maintaining its commitment to human-centric AI, democratic values, and regulatory leadership – constraints that its competitors do not share.

Europe’s challenge involves not merely closing a capability gap but doing so while maintaining its commitment to human-centric AI, democratic values

The concept of sovereignty itself requires careful definition. As European strategic documents acknowledge, “autonomy is not autarky”. Complete technological self-sufficiency would require Europe to replicate entire global supply chains domestically, an economically irrational and practically impossible undertaking. Instead, the relevant question becomes to what degree of selective sovereignty in critical AI capabilities can Europe realistically achieve? And what irreducible dependencies must be managed through diversification, resilience, and strategic partnerships?

The Hardware Bottleneck

The foundation of any AI system rests on specialized hardware, particularly advanced semiconductors and graphics processing units. Here, Europe faces its most acute sovereignty challenge. The continent holds less than 10% of global semiconductor production, a share that has been declining despite the €43 billion European Chips Act aimed at doubling Europe’s global market share to 20% by 2030. Three years after the Chips Act’s launch, industry observers note that “Europe’s share of global chip production continues to decline”, revealing the immense difficulty of reversing decades of manufacturing migration to Asia and the United States.

NVIDIA commands 92-94% of the discrete GPU market, with AMD holding 5-8% and Intel capturing less than 1% of AI chip share

The GPU dependency presents an even starker picture. NVIDIA commands 92-94% of the discrete GPU market, with AMD holding 5-8% and Intel capturing less than 1% of AI chip share. These GPUs provide the computational muscle for training and running advanced AI models, making them indispensable infrastructure. The problem extends beyond market dominance to geopolitical vulnerability. In January 2025, the outgoing Biden administration imposed export controls that divided EU member states into tiers, with 17 countries facing caps on advanced AI chip imports while only 10 EU nations were designated as “key allies” with unrestricted access. This unilateral US decision effectively fragmented the EU’s single market approach to AI development, treating member states differentially despite their shared economic and political union.European Commissioners Henna Virkkunen and Maroš Šefčovič expressed concern that these restrictions could “derail plans to train AI models using European supercomputers,” arguing that “the EU should be seen as an economic opportunity for the US, not a security risk”. Yet the reality remains that European supercomputers and AI infrastructure depend almost entirely on American GPU suppliers, with five of the nine EU supercomputers under the EuroHPC program located in countries not considered “key allies” by the United States. Even supercomputers that have secured current GPU supplies face obsolescence within three years without access to next-generation chips, creating a perpetual dependency that export controls can weaponize.

The semiconductor manufacturing picture offers marginally more hope but remains constrained by long timelines and limited scope. Taiwan Semiconductor Manufacturing Company is constructing a fabrication facility in Dresden, Germany, while Intel plans two fabs in Magdeburg at a cost exceeding $30 billion. However, these facilities will primarily focus on 10nm to 5nm process nodes rather than the cutting-edge 2nm technology that powers the most advanced AI chips, and full operation remains years away with uncertain timelines. European-headquartered semiconductor firms like ST Microelectronics, Infineon, and NXP collectively account for only about 10% of global semiconductor sales and specialize in automotive, industrial and niche applications rather than the high-performance computing chips essential for AI.

European-headquartered semiconductor firms like ST Microelectronics, Infineon, and NXP collectively account for only about 10% of global semiconductor sales and specialize in automotive, industrial and niche applications rather than the high-performance computing chips essential for AI

Perhaps most critically, Europe faces profound dependency on materials necessary for semiconductor production. The continent relies on China for 85 to 98% of its rare earth elements and rare earth magnets, which are crucial for manufacturing electronics, renewable energy systems and defense equipment. China controls 60 to70% of global rare earth mining and up to 90% of processing capacity, giving it leverage that it has demonstrated willingness to use. Export restrictions China imposed in April and October 2025 caused European rare earth element prices to spike to six times higher, leading to automotive production stoppages across Europe when stockpiles ran critically low. While Europe possesses rare earth deposits in Turkey, Sweden, and Norway, the continent lacks operational mining, refining and processing capabilities that China has built through decades of state-directed investment. Developing this infrastructure faces lengthy approval processes, stringent environmental regulations and public opposition – barriers that do not constrain China’s operations.

The hardware layer also includes a critical European strength that carries its own vulnerabilities. ASML’s monopoly on extreme ultraviolet lithography machines essential for manufacturing advanced semiconductors. While ASML represents genuine European technological leadership, the Netherlands-based company operates under export restrictions that prevent sales of its most advanced equipment to China, reflecting how even European champions become entangled in US-China technological competition. ASML’s deep ultraviolet systems, which are subject to less stringent controls, have been sold to Chinese entities including defense contractors, creating controversy over whether export control frameworks adequately address component-level dependencies. The fact that ASML’s lithography equipment requires specialized maintenance only the company can provide means that China’s access to functional advanced chip-making capability depends significantly on whether Dutch authorities allow ASML to continue servicing Chinese-installed equipment.

This hardware analysis reveals that 100% sovereignty is impossible in the foundational layer of the AI stack

This hardware analysis reveals that 100% sovereignty is impossible in the foundational layer of the AI stack. Europe cannot realistically manufacture advanced AI chips at scale within any relevant timeframe, cannot secure unfettered access to the materials necessary for semiconductor production, and remains subject to export controls imposed by both allied and rival powers. The best achievable outcome involves diversified supply chains, strategic stockpiling of critical components, accelerated but still lengthy development of domestic manufacturing for trailing-edge chips, and diplomatic efforts to secure predictable access to advanced components from allies

Cloud Infrastructure

Moving up the technology stack, cloud computing infrastructure represents the second critical dependency. US hyperscalers – Amazon Web Services, Microsoft Azure and Google Cloud – control approximately 65-72% of the European cloud market, while the largest European provider, OVHcloud, commands only 1-5% market share. This concentration creates multiple sovereignty vulnerabilities that extend well beyond simple market dominance.

The largest European provider, OVHcloud, commands only 1-5% market share

The US CLOUD Act grants American authorities the right to access data stored by US companies even when that data resides in European data centers, creating a fundamental jurisdictional conflict with the EU’s General Data Protection Regulation. European organizations operating on US-controlled cloud platforms theoretically place their data under potential foreign government access regardless of where servers are physically located. This legal vulnerability compounds operational dependencies. European enterprises, having built their digital infrastructure on AWS, Azure, or Google Cloud using proprietary services specific to these platforms, find themselves unable to switch providers without massive migration costs and business disruption. As one European industry observer noted, “European governments and enterprises are bound hand and foot to US cloud service providers. They rarely even manage to switch a service from one US supplier to another US supplier”. The irony intensifies when examining European cloud sovereignty initiatives. The Gaia-X project, launched in 2020 to build an interoperable, secure, European-led cloud infrastructure based on open standards, has struggled with slow progress, complex governance negotiations and controversy over allowing US hyper-scalers to participate. The fundamental tension lies in whether European cloud sovereignty requires exclusion of non-European providers or can be achieved through federated architectures and common standards regardless of provider nationality. Some Gaia-X proponents argue that “the highest level of sovereignty for European end customers can only be provided by providers having their headquarters in Europe,” while others advocate for a more inclusive approach that attracts necessary investment and technical capacity. Three years after launch, Gaia-X has created frameworks and data space specifications but has not yet delivered functional large-scale infrastructure that enables European organizations to meaningfully reduce hyper-scaler dependence.

Three years after launch, Gaia-X has created frameworks and data space specifications but has not yet delivered functional large-scale infrastructure that enables European organizations to meaningfully reduce hyper-scaler dependence

European cloud providers face structural challenges that transcend mere market share. OVHcloud, Scaleway, and Hetzner – the largest European alternatives – collectively serve less than 5% of the market and invest at a fraction of the scale of their American competitors. US cloud providers invest ten times more than European competitors, creating a widening capability gap. While these European providers emphasize data sovereignty, GDPR compliance, and sustainable infrastructure as differentiators, they struggle to match the breadth of services, global reach, and advanced AI capabilities that hyperscalers offer. For European enterprises deploying AI at scale, choosing European cloud providers often means accepting reduced functionality or investing significantly more to achieve equivalent performance. The AI-specific infrastructure dimension reveals an even starker imbalance. Together.AI announced plans in June 2025 to bring 100,000 NVIDIA Blackwell GPUs and up to 2 gigawatts of AI-dedicated data center capacity to Europe through partnerships, with initial deployments beginning late 2025 and large-scale buildouts through 2028. France separately announced plans to build Europe’s largest AI infrastructure with €15 billion investment targeting 1.2 million GPUs by 2030. These initiatives represent significant progress, yet they also highlight Europe’s starting deficit: the continent currently accounts for only 4% of global AI computing power. The EU’s planned network of 19 AI Factories (each with up to 25,000 H100 GPU equivalents) and five AI Gigafactories (each with at least 100,000 H100 GPU equivalents) would provide research institutions, startups, and SMEs with access to AI compute infrastructure. However, the €20 billion InvestAI fund will cover only approximately one-third of capital expenditures, requiring substantial private investment that remains to be fully mobilized.

The fundamental dependency remains that these supercomputers rely entirely on American GPUs, predominantly from NVIDIA, creating persistent vulnerability to export controls and supply disruptions

The EuroHPC Joint Undertaking has procured twelve supercomputers including JUPITER and Alice Recoque, Europe’s first exascale systems, with these systems interconnected through a federated platform by mid-2026. This represents genuine European capability development in high-performance computing. Yet the fundamental dependency remains that these supercomputers rely entirely on American GPUs, predominantly from NVIDIA, creating persistent vulnerability to export controls and supply disruptions. When US authorities can determine which European countries receive unrestricted access to advanced chips versus which face import caps, the question arises whether Europe truly controls its own computational destiny regardless of who operates the data centers. The cloud sovereignty analysis suggests that Europe can achieve partial independence through scaled investment in European cloud providers, migration of certain workloads to European infrastructure, and hybrid architectures that position critical systems on sovereign platforms while leveraging hyper-scalers for less sensitive operations. Complete independence, however, would require European cloud providers to achieve parity with hyperscalers in scale, service breadth, and AI capabilities – an outcome that seems unlikely absent massive sustained investment and fundamental shifts in market dynamics.

The AI Model Layer

At the AI model layer, Europe has demonstrated meaningful capability through companies like Mistral AI, Aleph Alpha and Velvet AI, yet faces formidable competitive challenges. Mistral AI, founded in April 2023 by former DeepMind and Meta researchers, reached a valuation of €11.7 billion in September 2025 following a €1.7 billion funding round led by ASML, making it Europe’s most valuable AI startup. The company develops open-source language models using efficient mixture-of-experts architectures that achieve GPT-4 comparable performance with drastically fewer parameters, reducing computational requirements by over 95%. Mistral’s Le Chat assistant exceeded 1 million downloads in 13 days following mobile launch, demonstrating European capacity to build consumer-facing AI products that compete directly with ChatGPT.

Mistral’s Le Chat assistant exceeded 1 million downloads in 13 days following mobile launch, demonstrating European capacity to build consumer-facing AI products that compete directly with ChatGPT.

Germany’s Aleph Alpha focuses on sovereign AI models emphasizing multilingualism, explainability and EU AI Act compliance, explicitly targeting public sector and enterprise customers with data sovereignty requirements. Italy’s Velvet AI, trained on the Leonardo supercomputer, emphasizes sustainability and broad European language coverage optimized for healthcare, finance, and public administration. These European models collectively demonstrate technical capability, particularly in multilingual performance, efficiency optimization, and regulatory compliance – areas where European approaches differentiate from US competitors focused primarily on scale and capability maximization. Yet the capability gap remains substantial. The Stanford Human-Centered AI Institute’s 2024 report found that US-based institutions produced 40 notable AI models, China produced 15, and Europe’s combined total was three. This disparity reflects underlying investment imbalances. US private AI investment hit $109.1 billion in 2024, nearly 12 times higher than China’s $9.3 billion and 24 times the UK’s $4.5 billion, with the gap expanding rather than narrowing. European AI startups receive just 6% of global AI funding compared to 61% flowing to the United States. While European AI funding grew 60% from 2023 to 2024, US investment increased 50.7% during the same period from an already dominant base, and grew 78.3% since 2022.

DeepSeek achieved performance rivaling OpenAI’s most advanced models while training on dramatically less compute using older chips, demonstrating that efficiency innovations can partially compensate for hardware restrictions

The emergence of China’s DeepSeek R1 model in January 2025 added a disruptive dimension to the competitive landscape. DeepSeek achieved performance rivaling OpenAI’s most advanced models while training on dramatically less compute using older chips, demonstrating that efficiency innovations can partially compensate for hardware restrictions. The model’s open-source release triggered concerns that its architecture and weights provide hostile actors with powerful AI capabilities at minimal cost, while simultaneously proving that export controls on advanced chips slow but do not prevent adversaries from reaching the AI frontier. For Europe, DeepSeek’s breakthrough carries mixed implications. It validates efficiency-focused approaches similar to those Mistral AI pursues, yet demonstrates that open-source model availability reduces the strategic value of developing indigenous models when comparable capabilities become freely accessible worldwide.

The talent dimension intersects critically with model development capacity. Europe boasts a 30% higher per-capita concentration of AI professionals than the United States and nearly triple that of China, reflecting the continent’s strength in technical education through institutions like ETH Zurich, University of Oxford, and France’s Inria. However, Europe suffers from severe brain drain, with only 10% of the world’s top European AI researchers choosing to work within Europe while the remainder migrate to higher-paying positions in the United States. Prominent examples include Yann LeCun leaving France to build his career at Bell Labs, NYU, and Meta; Demis Hassabis building DeepMind in London before Google’s acquisition moved the center of gravity to the US ecosystem; and Łukasz Kaiser, co-creator of the Transformer architecture, leaving Europe for Google Brain and subsequently OpenAI. This talent exodus reflects structural factors beyond compensation alone. European AI engineers describe an environment lacking “upside, transparency, urgency and ecosystem density” compared to Silicon Valley, where “ambition density is insane” and network effects accelerate career growth. The salary differentials are stark enough that one Swiss machine learning engineer noted earning less in Switzerland than from running an Airbnb for two hours weekly in the United States. European initiatives like Germany’s AI Strategy, which funds 100 new AI professorships, aim to stem the brain drain, but retaining top researchers requires competing with American tech giants offering compensation packages that European academic institutions and smaller companies cannot match.

European AI engineers describe an environment lacking “upside, transparency, urgency and ecosystem density” compared to Silicon Valley

The acquisition pattern compounds the sovereignty challenge. Advanced Micro Devices acquired Finland’s Silo AI for $665 million in 2024, Europe’s largest AI deal to date, securing its expertise in custom AI models and enterprise clients. Microsoft paid $650 million to license Inflection AI’s models while hiring the company’s founders and team, exemplifying “acqui-hiring” where US tech giants absorb European researchers to bolster their laboratories. Most major exits involve acquisition by US companies, potentially undermining strategic autonomy goals driving European AI investment. European startups that successfully scale increasingly face the choice between accepting US acquisition offers that provide founders and investors with returns or remaining independent with limited access to the capital and markets necessary for global competition.The AI model analysis reveals that Europe can develop competitive models in specific niches – particularly those emphasizing efficiency, multilingual capability, and regulatory compliance – but cannot achieve complete independence when foundational models are developed primarily in the United States and China with vastly greater investment. European AI sovereignty at the model layer realistically means ensuring the continent possesses credible indigenous capabilities that provide alternatives for sovereignty-sensitive applications while acknowledging that many users will choose frontier models regardless of origin

Innovation-Compliance Tension

Europe’s regulatory approach to AI, embodied in the AI Act that entered force in phases from 2024 to 2027, creates a significant tension with sovereignty ambitions. The Act represents the world’s first comprehensive AI regulation, introducing strict requirements for high-risk AI systems, transparency obligations for general-purpose models, and prohibitions on certain applications like social scoring and facial recognition scraping. While regulation aims to ensure trustworthy AI aligned with European values, it imposes substantial compliance burdens particularly on startups. Research by the German AI Association and General Catalyst found that EU AI Act compliance costs startups €160,000 to 330,000 annually and takes 12+ months to implement. With average seed funding in Europe around €1.3 million providing approximately 18 months of runway, the AI Act requires startups to spend roughly 15% of their cash and 66% of their time on compliance rather than product development. Sixteen percent of surveyed startups indicated they would consider stopping AI development or relocating outside the EU due to compliance burdens. The European Commission has attempted to reduce SME compliance costs through proportional fees and support mechanisms, yet the fundamental tension remains between comprehensive regulation and the rapid iteration necessary for AI innovation.

The open-source provisions particularly illustrate the regulatory complexity

The open-source provisions particularly illustrate the regulatory complexity. The AI Act exempts certain open-source general-purpose AI models from key obligations provided they meet stringent conditions. The model’s license must be fully open (i.e. there can be no monetization whatsoever, including technical support or platform fees) and the model’s parameters and architecture must be publicly available. However, “for the purposes of this Regulation, AI components that are provided against a price or otherwise monetized, including through the provision of technical support or other services, including through a software platform, related to the AI component, or the use of personal data for reasons other than exclusively for improving the security, compatibility or interoperability of the software” do not benefit from the exemption. This means that every company with commercial operations immediately falls under strict AI Act rules identical to those applied to proprietary model providers, regardless of whether they use open-source models.

Every company with commercial operations immediately falls under strict AI Act rules identical to those applied to proprietary model providers, regardless of whether they use open-source models.

Critics argue this approach stifles the very innovation Europe needs to compete globally. As one analysis noted, “European companies must also be able to take advantage of this. It must be as easy as possible for them to use open-source AI, without major bureaucratic hurdles. DeepSeek will definitely not be the last open-source model that can compete with the proprietary AI models of the big players”. The regulatory framework essentially treats European startups building on open-source foundations identically to how it treats OpenAI or Google, despite vast differences in resources and market power. Some propose expanding exemptions for commercial use of open-source AI with upper limits to regulate Big Tech more strictly – similar to the Digital Markets Act approach – rather than applying uniform rules regardless of company size. The GDPR intersection with AI training creates additional complexity. As AI models are trained on datasets that may include personal data, GDPR compliance requirements around consent, data minimization, transparency, and explainability directly impact model development. The European Commission has been in advanced talks to formally recognize “legitimate interest” as the legal basis for training AI technologies with personal data under GDPR, representing potential regulatory evolution to reduce friction. However, the fundamental challenge remains that European AI developers must navigate comprehensive data protection requirements that US and Chinese competitors do not face, creating asymmetric regulatory burdens in a global market. The regulatory analysis suggests that Europe faces a critical choice. Prioritize comprehensive AI regulation that may slow indigenous innovation and drive startups to relocate, or streamline compliance burdens particularly for SMEs and open-source usage to create a more permissive environment for European AI development. The current trajectory suggests European authorities recognize the tension, with regulatory simplification proposals and AI Act implementation guidance aimed at reducing burdens. Yet the question remains whether adjustments will prove sufficient to enable European AI champions to compete against rivals operating in less constrained regulatory environments.

Investment Gap

The financial dimension of AI sovereignty reveals persistent structural challenges. European AI funding reached €12.8 billion in 2024, representing steady progress but comprising only a small fraction of the $110 billion in global venture capital flowing to AI-first companies, with the United States claiming 74%. The EU invests in artificial intelligence only 4% of what the United States spends, creating a compounding capability gap. Venture capital access disparities prove particularly acute: firms based in the US attract 52% of venture capital funding, those in China receive 40%, while EU-based startups capture just 5%. The European Union’s €200 billion InvestAI initiative, announced by Commission President Ursula von der Leyen in February 2025, aims to mobilize resources through public-private partnership. The structure envisions €50 billion in public funding with €150 billion from private investors, targeting AI infrastructure development, gigafactories, research, and startups. However, significant uncertainty remains regarding whether this private capital can actually be mobilized. A group called the EU AI Champions Initiative has pledged €150 billion in investment from providers, investors, and industry, yet concrete commitments beyond these pledges remain unclear as EU officials declined to provide specifics on contributor lineup progress. Skepticism toward the InvestAI program focuses on its “highly bureaucratic” nature and lack of urgency. Alexandra Mousavizadeh, CEO of London AI consulting firm Evident, characterized it as “a classic European, ‘We’ve got to have some sort of strategy and then we’ll think about it, we may spend some money on it,'” expressing doubt that European authorities understand the urgency or are deploying resources fast enough. The adoption curve in Europe lags significantly behind the United States across most sectors, reflecting not just capital constraints but also a weaker ecosystem with fewer AI development companies and specialists in business AI integration. The European Tech Champions Initiative represents a more concrete mechanism, with the European Investment Bank and EIF providing €3.75 billion in initial commitments from Germany, France, Italy, Spain, Belgium, and EIB Group resources. This fund-of-funds invests in large-scale venture capital funds that provide growth financing to late-stage European tech companies, addressing the scale-up gap where European startups often lack sufficient capital to compete globally and relocate overseas. Germany separately committed an additional €1.6 billion in January 2026 to support technology-driven startups throughout all development stages. ETCI has supported nine tech scale-ups valued at over $1 billion since 2023, demonstrating tangible impact. Yet the investment gap continues widening despite these initiatives. US private AI investment grew from an already dominant position, with the disparity in generative AI being even more pronounced: US investment exceeded the combined total of China and the European Union plus the UK by $25.4 billion in 2024, expanding from a $21.8 billion gap in 2023. This widening gap reflects not merely public policy differences but fundamental ecosystem advantages: the United States benefits from deeper capital markets, a culture more accepting of risk and failure, networks connecting entrepreneurs with experienced operators, and exit options through acquisition by technology giants or public markets that provide returns enabling venture capital recycling.

Most major exits involve US acquirers rather than European consolidation

European M&A activity has increased, with AI deal value in Europe more than doubling from $480 million across 49 deals in 2023 to $1.1 billion across 45 deals in 2024. However, most major exits involve US acquirers rather than European consolidation, meaning successful European AI innovations frequently exit to American ownership. This pattern creates a self-reinforcing cycle: European investors achieve returns through US acquisitions, which validates the US exit path rather than encouraging patient capital that supports building European champions. The absence of European technology giants comparable to Microsoft, Google or Amazon limits domestic acquisition opportunities and reduces European startups’ negotiating power when US companies make offers. The investment analysis reveals that while Europe is mobilizing significantly more capital for AI than historically, the continent faces a fundamental ecosystem disadvantage that financial commitments alone cannot quickly overcome. Achieving meaningful AI sovereignty requires not just closing the current investment gap but building the patient capital pools, experienced operator networks, and exit pathways that enable venture capital to function as effectively in Europe as it does in Silicon Valley.

Geopolitical Constraints and Strategic Options

The geopolitical dimension imposes constraints on European AI sovereignty that extend beyond technology and markets into the realm of power politics and alliance management. The transatlantic relationship creates fundamental tensions: the United States remains Europe’s primary security guarantor and closest ally, yet simultaneously leverages Europe’s dependence on American technology as an instrument in its global trade confrontation with China. The January 2025 US export controls on AI chips, which divided EU member states into differentiated tiers, exemplified how even allied status does not preclude Washington from using technology access as geopolitical leverage. Europe finds itself caught between the US-China technological rivalry, repeatedly experiencing collateral impact from measures designed to advantage one superpower against the other. When the United States imposed sanctions on Huawei in 2019-2020 and pressured European countries to exclude Chinese telecommunications equipment from 5G networks, European operators faced disruption to planned infrastructure deployments despite their equipment choices posing no direct threat to American security. The semiconductor export control escalation targeting China’s advanced chip capabilities constrains European companies like ASML, which find their commercial relationships with China subject to restrictions imposed by Washington even when technology in question has European rather than American origins.

China’s rare earth export controls, imposed in April and October 2025 in response to US tariffs, demonstrated Beijing’s willingness to weaponize material dependencies against Europe despite the EU’s efforts to maintain amicable relations

China’s rare earth export controls, imposed in April and October 2025 in response to US tariffs, demonstrated Beijing’s willingness to weaponize material dependencies against Europe despite the EU’s efforts to maintain amicable relations. The temporary suspension of controls until November 2026 provides breathing room but highlights vulnerabilities in supply chains where China controls 60-90% of global production. European firms had not stockpiled rare earth elements before restrictions took effect, leading to production stoppages when supplies became scarce and prices spiked. This experience underscores that Europe’s dependencies make it vulnerable not only to deliberate weaponization by rivals but also to becoming collateral damage in Sino-American confrontations.The European response has emphasized diversification through partnerships rather than autarky. The EU’s International Digital Strategy, released in June 2025, states explicitly that “no country or region can tackle the digital and AI revolution alone,” acknowledging that supply and value chains of digital technologies are globally interconnected. The strategy promotes “autonomy through cooperation,” seeking to reduce specific vulnerabilities through diversified partnerships while recognizing that complete independence is neither achievable nor economically rational. This approach contrasts with China’s pursuit of self-sufficiency through massive state investment in indigenous capabilities and differs from America’s strategy of maintaining primacy through technological superiority combined with export controls denying adversaries access to cutting-edge systems. European strategic autonomy doctrine emphasizes selective sovereignty in critical capabilities rather than comprehensive autarky. As scholars analyzing the concept note, it “acknowledges that strategic autonomy is amenable to multiple meanings and diverse policies” rather than implying “independence, unilateralism and even autarky”. The practical application involves identifying which capabilities are genuinely critical for security and economic sovereignty, developing indigenous capacity in those domains, while accepting managed dependencies elsewhere backed by diversification, strategic stockpiling, and diplomatic relationships ensuring reliable access.

European strategic autonomy doctrine emphasizes selective sovereignty in critical capabilities rather than comprehensive autarky.

The challenge lies in European member states reaching consensus on which capabilities require sovereignty investment versus which can be sourced globally. Countries with strong technology industries like France and Germany may prioritize indigenous capability development, while smaller member states might prefer leveraging partnerships to access advanced systems without bearing development costs. The US export controls that differentiated between EU member states, designating some as “key allies” while imposing restrictions on others, revealed how external actors can exploit this fragmentation to Europe’s disadvantage. The geopolitical analysis suggests Europe must accept that 100% AI sovereignty is impossible in a deeply interdependent global technology system where hostile actors can weaponize dependencies while even allies can impose conditional access. The realistic goal involves achieving sufficient indigenous capability in genuinely critical domains  – such as AI systems supporting national security functions, critical infrastructure protection, and sensitive government operations – while accepting market-based solutions for commercial applications. This requires sustained investment in European champions, diversified supply chains reducing concentration risk, strategic stockpiles of critical components, and diplomatic initiatives ensuring European interests receive consideration in allied decision-making.

The geopolitical analysis suggests Europe must accept that 100% AI sovereignty is impossible in a deeply interdependent global technology system

Pathways to Pragmatic Sovereignty

If 100% AI sovereignty remains unachievable, what forms of pragmatic sovereignty can Europe realistically pursue? The evidence suggests several pathways that balance ambition with constraints.

1. Layered sovereignty recognizes that different applications require different degrees of autonomy. National security AI systems, critical infrastructure control systems, and government functions processing highly sensitive data demand maximum sovereignty achievable, justifying premium costs and reduced functionality relative to foreign alternatives. Commercial applications with lower security implications can leverage global solutions, including US cloud infrastructure and frontier models, provided contracts include appropriate data protection guarantees and exit provisions preventing vendor lock-in. This tiered approach allows Europe to concentrate limited resources on genuinely critical capabilities rather than attempting comprehensive self-reliance.

2. Capability sovereignty focuses on maintaining indigenous expertise and industrial base even when not seeking complete market dominance. Mistral AI’s success – reaching €11.7 billion valuation with viable products competing against OpenAI and Google – demonstrates European capacity to develop world-class AI models. The existence of credible European alternatives provides negotiating leverage with US providers, creates options for sovereignty-sensitive deployments, and ensures Europe retains the specialized talent and operational experience necessary to assess, integrate, and potentially modify foreign systems. Capability sovereignty does not require capturing majority market share but demands sufficient scale to sustain ongoing development and attract top talent.

3. Infrastructure sovereignty involves building physical computing infrastructure and data center capacity within European jurisdiction subject to European law. The EuroHPC supercomputers, AI Factories, and AI Gigafactories provide research institutions, startups, and public sector entities with computational resources not subject to foreign access requests. Investment in European cloud providers like OVHcloud, Scaleway, and Hetzner, though not eliminating hyperscaler dependency, creates alternatives for organizations prioritizing data sovereignty. France’s €15 billion AI infrastructure investment targeting 1.2 million GPUs by 2030 represents meaningful capability development even if not achieving parity with US infrastructure.

4. Supply chain resilience through diversification reduces concentration risk without requiring autarky. Europe cannot manufacture leading-edge semiconductors domestically in relevant timeframes but can secure commitments from multiple international suppliers, maintain strategic stockpiles, develop domestic capacity in trailing-edge nodes sufficient for many applications and cultivate diplomatic relationships ensuring predictable access. Rare earth dependencies can be partially addressed through European mining development, diversification to Australian and Malaysian sources, and development of recycling technologies reducing primary material demand. Complete independence proves impossible, but diversification transforms existential dependencies into manageable risks.​​​

5. Regulatory sovereignty involves using Europe’s market power to shape global AI development through standards and requirements that reflect European values. The AI Act, despite its compliance burdens, establishes norms around transparency, explainability and risk management that become de facto global standards for companies seeking European market access. GDPR precedent showed that European regulation can achieve global reach when multinational companies find compliance more efficient than maintaining separate regional practices. Regulatory sovereignty allows Europe to project influence even when not achieving technological leadership, though this approach requires balancing regulatory ambition against innovation requirements.

6. Talent sovereignty focuses on retaining and developing human capital that ultimately determines AI capability. While Europe cannot match Silicon Valley compensation, it can leverage strengths in work-life balance, social systems, geographic proximity to family, and mission-driven opportunities to retain researchers who prioritize factors beyond salary maximization. Initiatives funding AI professorships, supporting research institutes, facilitating industry-academia partnerships and streamlining immigration for international AI talent can help offset the brain drain. The fundamental requirement involves creating an ecosystem where ambitious AI researchers can build globally significant careers without relocating to the United States.​

These pathways collectively define a sovereignty strategy that European institutions increasingly adopt: strategic autonomy rather than autarky, diversified dependencies rather than complete independence, selective indigenous capability rather than comprehensive self-sufficiency. The European approach emphasizes partnerships and cooperation as sovereignty instruments rather than obstacles to sovereignty. Success requires sustained political commitment, substantial financial investment beyond current levels, regulatory frameworks that enable rather than constrain innovation, and realistic expectations about what sovereignty actually means in a deeply interdependent global technology system.

The Verdict: Strategic Autonomy, Not Complete Sovereignty

The accumulated evidence leads to an unambiguous conclusion: European AI cannot be 100% sovereign within any realistic timeframe or reasonable resource commitment. The dependencies span too many layers of the technology stack, the investment gaps have grown too large, the supply chains prove too globally distributed, and the geopolitical constraints remain too powerful for complete independence to be achievable. Europe lacks indigenous GPU manufacturing and will not develop competitive alternatives to NVIDIA in the foreseeable future. The continent depends structurally on US cloud infrastructure and will not displace hyperscalers from market dominance despite scaled investment in European alternatives. Critical material dependencies, particularly rare earths, cannot be eliminated through domestic production given geological constraints and decades-long infrastructure development timelines. The brain drain of top AI talent continues despite retention efforts, reflecting ecosystem advantages that policies alone cannot quickly overcome. Yet acknowledging impossibility of complete sovereignty does not condemn Europe to technological vassalage. The pragmatic sovereignty pathways outlined above—layered sovereignty, capability sovereignty, infrastructure sovereignty, supply chain resilience, regulatory sovereignty, and talent sovereignty—collectively enable Europe to achieve meaningful autonomy in critical domains while accepting managed dependencies elsewhere. Mistral AI’s success proves European capability to develop competitive AI models. The EuroHPC supercomputers demonstrate European capacity to build world-class computational infrastructure. ASML’s lithography monopoly shows European industrial strength in specific technological domains remains globally unmatched. The AI Act and GDPR exemplify regulatory power that shapes global technology development through market access requirements. The strategic autonomy framework differs fundamentally from self-sufficiency. Strategic autonomy means ensuring Europe possesses sufficient indigenous capabilities, diversified options, and resilient systems that no single external actor can compromise European security or coerce European policy through technology denial or conditional access. It means Europe can pursue its interests and values even when those diverge from allies or adversaries. It means European organizations have genuine alternatives—perhaps not perfect substitutes, but viable options – when sovereignty concerns preclude using foreign systems. It means Europe retains the specialized talent, operational experience, and industrial base to independently assess technological developments, make informed procurement decisions, and potentially indigenise critical capabilities when circumstances demand. The path forward requires European institutions to clearly articulate what sovereignty actually means operationally, which specific capabilities require indigenous development versus which accept managed foreign dependencies, and what trade-offs between sovereignty ambition and economic efficiency or capability access European societies are willing to accept. It demands sustained investment at levels dramatically exceeding current commitments – the €200 billion InvestAI target likely represents a floor rather than a ceiling for what achieving meaningful autonomy requires. It necessitates regulatory evolution that reduces compliance burdens on European startups while maintaining commitments to trustworthy AI, creating asymmetries that constrain foreign giants more than indigenous innovators. Most critically, achieving pragmatic sovereignty demands that European decision-makers resist both triumphalist rhetoric suggesting complete independence is attainable and defeatist resignation accepting perpetual dependency as inevitable. The realistic middle path—building selective indigenous capabilities, diversifying supply chains, investing in European champions, retaining critical talent, leveraging regulatory power, and cultivating strategic partnerships – offers Europe meaningful autonomy without the impossible goal of comprehensive autarky. In a world where technology has become a primary domain of great power competition, even partial sovereignty represents a substantial achievement worth the considerable investment it requires.

The question is not whether European AI can be 100% sovereign – the evidence clearly demonstrates it cannot. The relevant questions are what degree of sovereignty can Europe achieve, what will it cost to get there and what governance structures will ensure investments actually deliver the strategic autonomy they promise rather than merely funding industrial policy that fails to reduce dependencies?These questions demand continued attention as Europe navigates the treacherous intersection of technological ambition, market reality, and geopolitical constraint that defines the contemporary landscape of artificial intelligence sovereignty