The Sovereignty Paradox

The compute chasm is the sharpest expression of a wider economic divergence. US nominal GDP reached $29.2 trillion in 2024 versus the EU27’s $19.4 trillion; the two economies were within $1 trillion of each other in 2008 ²⁹. Purchasing-power parity narrows the nominal gap considerably, but hyperscaler capex is denominated in dollars, which is what finances the AI buildout.

Europe possesses 5 to 10% of global AI compute capacity. The United States possesses 60 to 75% ³. This is not a gap that policy can close quickly; it reflects decades of underinvestment and structural disadvantages in chip manufacturing, hyperscale data centers, and energy infrastructure.

American hyperscalers (Microsoft, Google, Amazon, Meta) will spend an estimated $700 billion on AI infrastructure in 2026 alone ². The European Union’s various initiatives promise €200 billion, spread over multiple years and multiple priorities ³. OpenAI’s Stargate project is scaling toward roughly 7 gigawatts of committed capacity, the equivalent of 5 to 7 nuclear reactors ²¹. Europe’s largest comparable project would deliver perhaps 20% of that capacity. This is a chasm.

The funding gap at the startup level tells the same story. European AI chip startups raised roughly $800 million in the first months of 2026. Their American counterparts raised $4.7 billion ²². Mistral’s €723 million round ¹⁷ and its 13,800-GPU data center near Paris represent Europe’s most ambitious private effort, and they amount to roughly one month of Microsoft’s AI capex.

Brussels Signal put it plainly in February 2026: “AI is now a much stronger area of European dependency on the US than defense was, and there is little use in trying to pretend this is not the case” ¹.

The dependency runs through every layer of the stack. European AI startups train on American cloud infrastructure; roughly 60 to 65% of the European public cloud market sits with AWS, Microsoft Azure, and Google Cloud ^{15, 16}. They use chips designed in America and manufactured in Taiwan. They license foundational models from OpenAI, Anthropic, and Google. Microsoft’s EU Data Boundary, AWS’s European Sovereign Cloud, and Google’s Sovereign Cloud all exist, but none can promise immunity from the US CLOUD Act; Microsoft itself has acknowledged it cannot guarantee EU data will never be requested by US authorities.

Beneath all of this lies an even more concentrated vulnerability: Taiwan. TSMC controls approximately 70% of global foundry revenue and more than 90% of production at the leading-edge nodes ¹⁸, the 3-nanometer and 5-nanometer processes that power AI training, data centers, and modern defense systems. Volume production on the N2 (2-nanometer) node began in Q4 2025 ²⁰, and Taiwan’s legislature has restricted TSMC from building its most advanced nodes abroad. A single company, on a single island roughly 150 kilometers from mainland China, underpins the entire AI revolution. Bloomberg Economics estimates the global economic damage from a Taiwan conflict at about $10 trillion, roughly 10% of world GDP and exceeding the damage from COVID-19 ¹⁹. China’s PLA is targeting 2027 for Taiwan-scenario readiness, and December 2025’s exercises were the largest Taiwan-focused drills ever conducted ¹⁹. In a Taiwan contingency, AI sovereignty becomes meaningless for everyone.

American policymakers face the sovereignty paradox in mirror image. The rhetoric of trade wars and tariffs suggests that the United States can decouple from partners who do not serve its interests. The reality of economic interdependence suggests otherwise.

Bilateral trade between the United States and Europe amounts to €1.68 trillion annually ⁴, the largest economic relationship in the world. European investors hold nearly half of all US government debt owned abroad. European companies employ millions of American workers; American companies employ millions of Europeans.

When tariff threats peaked in early 2025, market chaos ensued. The S&P 500 fell 21% in weeks, the second-fastest bear market in history ⁵. The reversal was equally swift: when tariffs were paused, markets recovered everything and more. The pattern, aggressive tariff announcement followed by market panic followed by policy reversal, repeated across multiple episodes through 2025 and into 2026. The Turnberry agreement, cleared by the European Parliament in March 2026, locks tariffs on most EU goods at 15% in exchange for EU elimination of duties on most US industrial goods ⁸. The European Central Bank projects EU export growth to the US will slow by roughly 4.6% in 2026 as tariffs converge to 15% across most partners ⁶.

The Mint Magazine’s analysis was direct: neither side can decouple without inflicting severe damage on itself. European creditors cannot dump US Treasuries without crashing the value of their own holdings. American manufacturers cannot relocate supply chains without years of disruption and higher costs. The sinews of integration, financial flows, supply chains, talent mobility, are too deep to sever without mutual harm.

The starkest illustration of the sovereignty paradox involves defense. European leaders speak of strategic autonomy in security, particularly following uncertainties about American commitment to NATO. Yet AI is central to modern defense capabilities, from autonomous systems to intelligence analysis to cyber operations.

If Europe cannot achieve AI sovereignty in commercial applications, how will it achieve AI sovereignty in defense? The likely answer is that it will not. European militaries will depend on American AI systems just as European businesses depend on American cloud infrastructure. The pursuit of defense autonomy will, paradoxically, deepen technological dependence.

The numbers make the dependency concrete. The US Department of Defense requested roughly $1.8 billion for AI-specific funding in fiscal 2024 ²³, with DARPA alone reporting that approximately 70% of its programs now use AI and machine learning ²⁴. Britain’s defense AI investment, by comparison, is measured in hundreds of millions of pounds across a multi-year program. France and Germany together allocate less to AI defense development annually than a single major American prime contractor invests in AI research. American defense AI spending is accelerating precisely as European governments debate the regulatory frameworks within which any future sovereign AI capability would operate.

The dependency is structural at every layer. Autonomous drone systems, whose decisive role in contemporary conflict was demonstrated in Ukraine, are trained on American platforms using American data. Battlefield intelligence analysis at alliance level flows through American infrastructure. The satellite communications European militaries now rely on (Starlink’s role in Ukraine was not a policy decision but an infrastructure dependency) sit on American commercial networks that neither the US government controls nor European governments can guarantee. Increased NATO contributions buy more platforms and more personnel. They do not, without a decade of semiconductor and compute investment, buy freedom from American AI.

The sovereignty paradox extends to the international system itself. For decades, leaders spoke of a rules-based international order, a system of norms and institutions that constrained state behavior and provided predictability. Recent events have supposedly disrupted this order, with great power competition replacing cooperative multilateralism.

The historical record suggests a different interpretation. A rules-based order requires credible enforcement, and the structure of international institutions has never provided it. The UN Security Council, with its veto-wielding permanent members, was designed to prevent great power conflict, not to enforce universal rules. Every major power has acted on its own interests when it had the capability to do so. The Security Council has proven most effective at issuing statements that change nothing.

Consider the evidence. The 1990s Balkans conflicts demonstrated that intervention depends on great power will, not legal principles. In 2003, the United States invaded Iraq without UN Security Council authorization. In 2014, Russia annexed Crimea while the international community issued condemnations that achieved nothing. China has steadily expanded its territorial claims in the South China Sea despite international tribunal rulings. In March 2026, the United States and Israel launched Operation Epic Fury against Iran’s military capabilities, closing the Strait of Hormuz and triggering the largest oil supply disruption in the history of the global market; no UN Security Council authorization was sought. In each case, nations acted according to capability and consequence.

What has changed is not the existence of rules but the perception of them. The unipolar moment created the illusion of universal norms. As power becomes more distributed, the fiction of universal rules becomes harder to maintain.

Institutions still matter. Norms constrain behavior at the margin. Multilateral coordination solves real problems. But investors should be clear-eyed: there was never a stable rules-based equilibrium to which the world will return. International politics has always been shaped by power, interest, and capability. Those who bet on a return to normal are betting on a normal that never existed.

The sovereignty paradox looks different outside the transatlantic corridor. While Europe debates regulatory frameworks, other regions have been building.

India offers the most striking counterexample. The Unified Payments Interface processed 21.7 billion transactions in January 2026 alone, worth ₹28.33 lakh crore ⁹, handling roughly 81% of retail digital transactions in the country. UPI now runs across 691 participating banks, up from 21 in 2016 ²⁷. Aadhaar covers 1.4 billion people ¹⁰. From April 1, 2026, two-factor authentication became mandatory for all digital payments, and the government’s new MEITY cloud framework locks Aadhaar, UPI, and PAN data to sovereign cloud providers ¹¹. UPI has also crossed borders, live or operational in Singapore, the UAE, Bhutan, Nepal, Sri Lanka, Qatar, France, and Mauritius ²⁶. India’s digital public infrastructure was not achieved through sovereignty rhetoric or regulatory directives. It was built by a government that treated technology as infrastructure rather than a threat to be regulated.

The Gulf offers a different model entirely: sovereignty through capital deployment. Saudi Arabia’s Public Investment Fund now manages roughly $930 billion in assets and is targeting $2 trillion by 2030 ¹². In 2025, PIF launched HUMAIN, its national AI company; in January 2026, Saudi Arabia awarded a $2.7 billion contract for the 480-megawatt Hexagon data center in Riyadh, targeting three to six gigawatts of domestic AI computing capacity ²⁵. Abu Dhabi’s Mubadala invested $12.9 billion in AI and digitalization in 2025, including $15 billion deployed directly into GlobalFoundries chip manufacturing ¹⁴. The two funds together manage more than $1.5 trillion. Their strategy is not sovereignty in the traditional sense. It is optionality purchased at scale: financing whoever wins, across geographies and political alignments, simultaneously.

The investment implications diverge. European technology exposure carries regulatory risk and uncertain execution, though the pivot from regulation to investment is real if belated: Nebius is targeting a $7 to $9 billion annual run rate by the end of 2026, Mistral is building its Paris data center, and the EU Cloud and AI Development Act (CADA) is attempting to unlock permitting for new data centers. The European Central Bank even restricted digital euro eligibility to EU-based providers only, a concrete step toward sovereign critical infrastructure. Indian digital infrastructure presents a growth story with sovereign backing and, increasingly, international adoption. Gulf sovereign wealth deployment creates co-investment opportunities but carries concentration risk in entities whose mandates serve geopolitical as well as financial objectives.

Mistral’s leadership and backers have argued that the CLOUD Act asymmetry is itself the European thesis ²⁸. Mission-critical workloads in defense, finance, and logistics cannot legally or politically sit on US-operated cloud, regardless of the data-residency marketing from US hyperscalers. That makes regulatory divergence a structural market wedge for European sovereign infrastructure, not a compliance problem to be engineered around.

Sovereignty is not achieved through declarations or directives. It is achieved through capability, whether built domestically or acquired globally. Investors who remain anchored to the transatlantic axis may find that the most consequential sovereignty plays of the next decade originate in Mumbai and Riyadh, not Brussels and Washington.

For the portfolio constructor, AI sovereignty translates into two exposures worth pricing explicitly.

The first is concentration risk in technology supply chains. Any holding with deep dependency on a single jurisdiction for compute, semiconductors, or data infrastructure carries geopolitical risk that standard financial models do not capture. The risk is invisible until a disruption makes it undeniable. TSMC’s dominance at leading-edge nodes, the US hyperscalers’ grip on European cloud, the concentration of AI talent in a handful of firms: each of these is a bet on political stability dressed as an efficient supply chain, and most portfolios do not price them as bets.

The second is the emerging premium on jurisdictional resilience. Companies building genuinely multi-source capabilities (redundant cloud providers, diversified chip sourcing, data localization) will command a resilience premium that increases, not decreases, as decoupling matures. Meta’s MTIA program to reduce Nvidia dependency, Nvidia’s US manufacturing push, and the European chip startups that raised $800 million in early 2026 are all early expressions of this trade.

The infrastructure of the AI era is being built along geopolitical fault lines. Portfolios that ignore these fault lines are carrying risk they have not measured.