Middle Powers’ Bounded Ambition in AI

In material assets, Australia's critical minerals and Brazil’s niobium reserves provide advantages at the resource layer, while the United Arab Emirates (UAE) is positioning itself as a regional compute hub through large-scale data and connectivity investment. In the industrial and knowledge category, South Korea and Taiwan lead in advanced chip manufacturing, while Japan retains strengths in semiconductor materials and equipment, and Türkiye has carved out an application niche in defense AI. Finally, institutional and market assets enable a state’s ability to shape rules, standards, and adoption. The “Brussels Effect”, despite emerging debate on its relevance in the AI era, remains the most prominent example of tech-shaping market power. Middle powers, spearheaded by Singapore, are also writing the rules collectively through a fast-growing web of digital trade agreements, setting baseline norms on how AI is deployed across borders. Domestically, Singapore’s agentic-AI governance framework and South Korea's AI Basic Act show how early regulation can shape emerging standards, while India pursues a path to AI governance through a “techno-legal approach” that builds regulatory oversight into systems by design.

Owning an asset, however, does not necessarily create usable leverage. Its political value depends not only on its intrinsic characteristics but also on factors such as external demand, available substitutes, and broader alliance constraints. Leverage in this context means a state’s ability to expand its own strategic choices, and shape those of others, including through diversification, denial, or conditioning.

Diversifying across suppliers, governance models, infrastructure hubs, and deployment pathways reduces exposure to any single provider or jurisdiction. France, for example, promotes sovereign European AI infrastructure centered on companies such as Mistral, which has moved from selling model access to a “full-stack and open approach” that is anchored by a 2026-2030 framework agreement with the French armed forces. Canada has embedded diversification into its AI strategy, exemplified by its Sovereign Technology Alliance with Germany and the Cohere-Aleph Alpha merger, which has been promoted as “the first transatlantic sovereign AI solution”. India offers a third model, building AI optimized for its linguistic diversity through its IndiaAI Mission rather than directly deploying foreign models.

Yet this diversification remains constrained, largely because many middle powers rely on great-power firms’ core infrastructure to run their own AI stacks. Canada’s compute capacity and India’s AI mission depend on NVIDIA. Germany, despite prioritizing digital sovereignty, must grapple with Chinese firms’ penetration of its 5G infrastructure. Türkiye hedges by hosting Huawei Cloud and a planned Google “sovereign cloud” region in Ankara. Even France’s sovereign offer involves interdependence through partnerships with Microsoft and China’s Z.ai on its GLM-5.2 open-weight model. Ukraine’s defense AI relies partly on Palantir. Diversification, then, is less about escaping great-power dependency than about managing asymmetric interdependence.

A few middle powers host companies that control key chokepoints in the AI value chain, most visibly in chip manufacturing. The Netherlands’ ASML holds a monopoly over extreme ultraviolet (EUV) lithography machines, Taiwan’s TSMC dominates frontier-chip manufacturing, and South Korea’s Samsung and SK Hynix control roughly 90% of high bandwidth memory (HBM) production. In theory, this concentration lets middle powers withhold or restrict access. That is the logic behind Taiwan’s “silicon shield”, which makes a Chinese military move against Taiwan globally costly.

In practice, this leverage is constrained by alliance commitments, supply-chain interdependence, and the economic, security, and reputational costs of restricting access to technology. Dutch export controls, for example, remain heavily influenced by the Netherlands’ alliance with the United States, making those controls unlikely to be used against American interests, as illustrated by the 2024 export controls on advanced semiconductor-manufacturing equipment. Last year’s Nexperia episode also shows the cost of geopolitical alignment. Dutch action against the Dutch-headquartered but Chinese-owned chipmaker, taken under US pressure, triggered a Chinese export block that disrupted European automotive supply chains within weeks. Chokepoint power also invites substitution. Speculation earlier this year of a Chinese alternative to ASML affected its market valuation while TSMC’s global expansion may thin its own silicon shield. Yet Taiwan cannot easily wield its shield as a bargaining chip. Withholding chips would hurt its own economy most. Japan’s 2019 curbs on chip-material exports to South Korea show the risk of actual denial. Seoul turned to other suppliers and expanded domestic production.

Between diversification and denial lies a gray zone of bargaining over strategic niches. The UAE illustrates opportunity and limits. Abu Dhabi sought to become a regional AI hub by leveraging its assets—capital, energy, and location—but this still required advanced US technology. Its leading AI firm, G42, consequently cut its China ties, securing a $1.5 billion investment from Microsoft. In 2024, a US-UAE AI partnership agreement followed. The bargain traded Abu Dhabi’s assets, plus strategic alignment, for American technology, binding the UAE more firmly into the US tech-security ecosystem. Saudi Arabia struck a similar bargain in late 2025. After Crown Prince Mohammed bin Salman pledged $1 trillion in investment in the United States, his country’s AI champion, HUMAIN, won American approval for advanced NVIDIA chips, conditioned on strict security and reporting requirements.

Kazakhstan is experimenting with a different bargain, leveraging its energy and critical mineral assets to pursue a. The country holds a multi-vector foreign policy, holding membership in the US-led Pax Silica initiative and China’s World Artificial Intelligence Cooperation Organization (WAICO). Yet China’s technology footprint in the country and US expectations of Pax Silica membership as a form of strategic alignment demonstrate the possible limits of Astana’s room for maneuver. Malaysia has already met those limits, attempting a similar bargain by leveraging cheap land and electricity to become a data center hub for the United States and China. American pressure to prevent Chinese companies from using Malaysia as a backdoor to restricted AI chips forced Kuala Lumpur to tighten controls, including export permits. These examples show that conditioning access can afford middle powers some leeway, but the window of opportunity is often open for a limited time. Alignment becomes either an opportunity worth taking, as the UAE did, or a pressure too costly to resist, as Kazakhstan and Malaysia found.

Systemic Effects: A (Slightly) Messier Power Map

AI amplifies middle powers’ ability to reshape the balance of power in three ways.

First, AI produces a more differentiated, if not necessarily less hierarchical, global technology ecosystem.

Value-chain assets give middle powers the ability to diversify, deny, or condition the options available to other states and firms. But, as the aforementioned cases show, structural dependencies heavily constrain this agency. Assets do not necessarily provide leverage and can even raise external pressures, as the Netherlands, Malaysia, and Kazakhstan illustrate. Assets can, however, still strengthen bargaining power, as the UAE and Saudi Arabia found. Both negotiated some terms even as they remain dependent on US technology. Rather than full autonomy, middle powers’ asset-based leverage amounts to “managed interdependence” that is asymmetric and constrained but not devoid of agency.

Second, AI incentivizes collective pooling of middle-power technological assets.

Despite structural constraints, middle powers increasingly recognize that their agency counts for more when exercised collectively. Through coordination, minilateralism, and issue-specific partnerships, they can combine complementary strengths to build AI capacity around the great powers. Canada’s AI strategy is the most concrete call for a “coalition of aligned democracies” to pool research, talent, compute, and procurement power. The Franco-German partnership with Mistral AI and SAP to build sovereign AI for public administration shows the same pooling logic on a bilateral basis. Türkiye’s Baykar and Italy’s Leonardo do the same for defense AI, pairing Turkish drones with Italian sensors, though Rome limited exports to EU- and NATO-aligned countries. The recent call on the sidelines of the 2026 UN General Assembly by 20 states, many of them AI middle powers, for global AI regulation is another attempt at collective bargaining. It has frequently been stressed that middle-power coalitions, though forged around a unifying logic, can be eroded by divergent interests. Some may prioritize investment and infrastructure, others stronger regulation. Authoritarian systems may favor state control and surveillance. Others may place acceleration and affordability before safety. Such divergences are not unique to middle-power coalitions but exist even among the closest allies. Success depends on managing such coalitions effectively.

Third, AI creates new dependencies and pressures within and across geopolitical divides.

AI middle powers are unlikely to dilute existing concentrations of power by themselves. But as they build their AI niches, new dependencies can emerge or intensify on energy, compute, connectivity, regulation, locally relevant data, and talent. This may increase great-power pressure on them to pick an AI stack. With Executive Order 14320, the second Trump White House has forcefully promoted US full-stack AI packages and sought to “decrease international dependence on AI technologies developed by [US] adversaries”. Even India, like the Netherlands and the UAE before, may face similar limits given its dependence on the United States for cloud and compute infrastructure despite New Delhi’s sovereign AI ambitions. China, in contrast, is unleashing a global charm offensive, drawing countries into its AI ecosystem with low-cost, powerful models such as Moonshot AI’s Kimi K3 and institutionalizing that pull through bodies such as WAICO.

Room for Maneuver: Transatlantic Implications

Can AI give middle powers real room for maneuver? The answer is a qualified yes. The hopes of collective middle-power agency are not misplaced but often overstated. AI assets can increase relative bargaining power with high demand, few substitutes, and permissive alliance commitments. Even then, however, leverage tends to erode over time as others diversify. The result is neither inevitable submission to a bipolar US-China contest nor a transition to benign multipolarity but a more diversified and less predictable order in which middle powers exercise meaningful but bounded agency. New sources of leverage are real, but they work within the limits of structural dependence. This plays out differently across the Atlantic. The United States confronts the moment as an established AI power while Europe, home to many middle powers, faces a test: Can its assets collectively provide agency and, ultimately, leverage? 

For Washington, the key question is whether to treat this moment as an opportunity or a threat. Treated as an opportunity, it becomes an exercise in coalition-building. Many middle powers with potential “chokepoint power”, such as the Netherlands, South Korea, and Taiwan, already sit within the US alliance and partner network. The tools Washington deploys against Chinese market power, from export controls to coordination on critical infrastructure cybersecurity, rely on long-standing buy-in by US allies. Coordinating these assets in response to China’s technological rise directly serves American interests. 

Treated as a threat, the moment may cut the other way. Outright hostility toward the agency of middle powers, coupled with expectations that they passively buy the full American stack envisioned by the American AI Exports Program, could push them to cooperate more closely and work around the United States. The more Washington relies on a policy toolkit that exerts regulatory pressure, withholds frontier AI capabilities, or conditions access on alignment with its preferences—especially if capabilities have defence applications—the stronger the incentive to build alternatives. French President Emmanuel Macron’s recent UN General Assembly address captures this sentiment. “I don't want anyone to become the vassal of one of the great powers. ... We must come together to build an open-source frontier model,” he said. Middle-power AI aspirations would remain constrained in this scenario, but “bounded” diversification is still diversification. Once underway, it is hard to reverse. In such a world, the United States may eventually need to accept being not the singular hub of global AI value chains but a participant within them.

For Europe, the moment is another incentive to deepen its geopolitical awakening. The Draghi competitiveness report found the EU relies on foreign countries for over 80% of its digital products, services, and infrastructure. Brussels has since launched a “technological sovereignty” package that it sees as part of the answer to that problem. Yet sovereignty, easy to aspire to, is also difficult and costly to achieve. While some have convincingly argued that Europe will never be an AI superpower, the middle-power moment invites Brussels to expand its partnerships at home and abroad.

Within Europe, seizing the moment requires taking stock of EU member states’ AI assets and of how Brussels and/or European capitals could collectively leverage them. ASML is the obvious case of potential chokepoint power, but unlike Taiwan’s reliance on TSMC, Europe’s strengths are spread across many companies. Sweden's Ericsson, for example, forms much of the digital backbone of US carrier networks, holding major positions at AT&T, T-Mobile, and Verizon. The window for EU member states to effectively leverage their assets by pooling is limited.

Beyond the EU, the moment encourages new tech-centered middle-power partnerships and a rethink of existing ones. The EU and South Korea’s reaffirming a commitment to cooperate on emerging digital technologies in November 2025 is a step in this direction. Coalitions with partners such as Canada, India, Japan, or the United Kingdom could pool capabilities that require scale and, as Washington steps back from global AI governance, help Europe lead on AI rules and safety. Likewise, initiatives such as Made in Europe and technological sovereignty need not be as inward-looking as their names suggest. They should foster supply chain cooperation with fellow-AI middle powers, helping Europe secure and diversify its AI value-chain dependencies. None of this would fundamentally alter Europe’s deep embeddedness in transatlantic tech networks, but it could help shape a more balanced transatlantic tech relationship despite strained ties.

Middle powers, many of them EU members, cannot opt out of the US-China contest over AI, but they can retain agency over its impact on their interests. Whether Washington draws them in or pushes them toward alternatives, and whether Europe can pool what its own middle powers hold, will determine whether their bounded agency adds up to more than the sum of its parts.

The views expressed herein are those solely of the author(s). GMF as an institution does not take positions.

AI tools were used to assist with shortening, restructuring, and language-editing of this article. Substantive content and arguments are entirely the authors’.