3.3.1 Trade and Economic Dependencies

Lena Müller works for ASML, the Dutch company that manufactures extreme ultraviolet (EUV) lithography machines—the only machines in the world capable of producing the most advanced semiconductors. ASML sells these machines for about $200 million each. They're the size of a bus, require months to assemble, and are so technologically complex that no other company on Earth can replicate them.

Every advanced AI chip—from Nvidia's GPUs to custom processors used by Google, Amazon, and Microsoft—depends on ASML's machines. If ASML stopped shipping, advanced AI chip production would halt globally within months.

In late 2025, the Dutch government seized control of Nexperia, a semiconductor company with partial Chinese ownership, splitting it into two distinct entities over national security concerns. China retaliated by imposing export controls on Nexperia-originated components. Watching this unfold, Lena recognized what it revealed: semiconductor supply chains have become weapons in a geopolitical contest, and companies like ASML—which make the tools that make the chips—are choke points that governments will control when strategic interests demand it.

The AI economy depends on a global supply chain more fragile, more concentrated, and more geopolitically vulnerable than almost anyone outside the industry realizes. A few companies, in a few countries, control components so critical that their disruption would cascade across the entire AI ecosystem. Those choke points are now targets.

The Semiconductor Bottleneck

In 2026, the AI chip market is estimated at $500 billion, part of a broader semiconductor market approaching $800 billion and on course toward $1 trillion before the decade's end. But this growth is constrained by bottlenecks that money alone cannot resolve, clustered at a surprisingly small number of nodes in the global supply chain.

Critical Input	Key Suppliers	Market Concentration	Primary Strategic Risk
Advanced logic chips	TSMC, Samsung	Taiwan, South Korea	Military conflict, blockade
EUV lithography machines	ASML	Netherlands	Export controls, political pressure
High-bandwidth memory (HBM)	SK Hynix, Samsung, Micron	South Korea, USA	Supply constraints, sanctions
Advanced packaging	TSMC, ASE Group	Taiwan	Conflict, natural disaster
Gallium	Chinese smelters	China (~98%)	Export restrictions
Germanium	Chinese refiners	China (~60%)	Export restrictions

The most acute near-term constraint is not logic chip fabrication itself, but high-bandwidth memory and advanced packaging. Even when leading-edge fabrication capacity exists, HBM availability and packaging throughput determine how many AI accelerators can actually ship. Nvidia and Apple have already secured large blocks of production capacity through the end of 2026, effectively locking out competitors and startups that cannot access equivalent supply. This concentration creates strategic dependencies across every sector relying on AI compute. The semiconductor industry is experiencing what analysts call a "giga cycle"—a structural shift in which AI demand is simultaneously reshaping the economics of compute, memory, networking, and storage at scales larger than any previous technology wave. That giga cycle, however, is built on supply chains that are fragile, concentrated, and increasingly subject to geopolitical interference.

The Export Control Regime

In January 2026, the U.S. Commerce Department's Bureau of Industry and Security revised its export policy for advanced AI chips. Previously, exports of Nvidia H200 and AMD MI325X-equivalent chips to China faced a "presumption of denial." Under the revised framework, applications are evaluated case-by-case, provided exports meet rigorous supply, security, and testing conditions.

The shift reflects competing imperatives. U.S. chip companies want access to China, their largest potential market. National security officials want to deny China the most capable chips, fearing they will accelerate military AI development, surveillance systems, and other technologies that threaten U.S. interests. The resulting compromise—limited exports under conditions, with case-by-case review—generates its own problems. Companies do not know which applications will be approved, how long reviews will take, or what conditions will be imposed. That uncertainty slows investment, complicates multi-year planning, and advantages competitors in countries without equivalent restrictions.

Alongside export controls, the Trump administration announced a 25% tariff on semiconductors meeting certain performance thresholds, following an investigation under Section 232 of the Trade Expansion Act. The tariff targets specific high-performance chips, but its significance lies in what it signals: trade policy is now a formal instrument of national security. For the first time, tariffs and trade policy surpassed talent risk as the leading concern among semiconductor industry leaders, with U.S. tariff policies contributing to inflationary pressures, altering global trade patterns, and generating the kind of cross-border uncertainty that delays long-term capital commitments.

The strategic logic of export controls has faced mounting criticism. During a January 14, 2026, House Foreign Affairs Committee hearing on winning the AI race against China, experts and members of Congress broadly expressed skepticism about the administration's approach, arguing that export controls are strategically incoherent and ultimately unenforceable. The core objection is that restrictions burden U.S. allies who depend on American chips while failing to stop China from acquiring technology through smuggling, reverse engineering, or domestic development. Each restriction also intensifies China's incentive to build indigenous alternatives. Over time, the cumulative effect of such measures risks producing a bifurcated global AI ecosystem—a U.S.-led bloc and a China-led bloc with incompatible technologies, standards, and supply chains—rather than constraining Chinese capability.

The Critical Minerals Chokepoint

Advanced semiconductors depend not only on fabrication equipment and manufacturing capacity but on a set of specialized minerals that underpin their performance. Gallium, germanium, indium, palladium, and tantalum are among the materials essential to modern AI chips. Without them, advanced chip manufacturing becomes impossible—yet supply is heavily concentrated in a single country.

China controls approximately 98% of global primary gallium production and roughly 60% of germanium refining. These are not marginal inputs. As of October 2025, Beijing added five more rare earth elements to its export control list and signaled that export licenses would be withheld from arms manufacturers and select semiconductor firms—moves widely interpreted as calculated steps to retain strategic leverage in ongoing U.S.-China negotiations.

The concentration of rare earth refining in China is a product of deliberate industrial policy over several decades. Western countries largely exited rare earth refining as China scaled up, drawn by lower environmental compliance costs and cheaper labor. The result is that China dominates not just extraction but the refining and processing stages—the most technically demanding and economically valuable parts of the value chain. Restarting this capacity in the United States, Europe, or allied nations would require years of investment, specialized facilities, and waste management infrastructure that no longer exists at scale outside China.

This creates a vulnerability structurally comparable to Taiwan's semiconductor concentration. If China restricted gallium or germanium exports, Western AI chip production could be severely constrained within months. Existing stockpiles provide a buffer, not a solution. A sustained embargo would force a fundamental reorganization of AI supply chains at enormous cost and with uncertain outcomes. Rare earth export controls are, in effect, leverage—a reminder to Western governments that economic interdependence is not a one-way relationship. The U.S. can restrict chip exports to China; China can restrict mineral exports to the West. Mutual vulnerability produces mutual leverage.

The Taiwan Problem

TSMC—Taiwan Semiconductor Manufacturing Company—produces more than 90% of the world's most advanced chips. If TSMC's operations were disrupted, the global AI industry would face collapse within months. There are no backup suppliers at TSMC's scale and technical capability. Samsung and Intel are investing to close the gap, but they remain years behind. If Taiwan's fabrication facilities went offline—due to natural disaster, cyberattack, or military conflict—there is no credible Plan B.

This dependency makes Taiwan the most strategically consequential island on the planet. A Chinese invasion or blockade would not be merely a regional crisis; it would be a global economic catastrophe, disrupting supply chains for consumer electronics, automotive, aerospace, defense, and AI simultaneously. Western governments understand this clearly. The U.S. CHIPS Act, passed in 2022, allocated $52 billion to incentivize domestic semiconductor manufacturing. TSMC is building fabs in Arizona, Samsung is expanding in Texas, Intel is investing in Ohio, and Europe has launched parallel semiconductor manufacturing initiatives.

These programs are necessary but insufficient as near-term hedges. The Arizona TSMC fab is not expected to reach full production until the late 2020s, and even at capacity it will not match TSMC's Taiwan operations in scale or technological depth. Domestic production represents insurance against catastrophic disruption, not a functional replacement for Taiwan's fabrication ecosystem. Until that ecosystem is genuinely replicated elsewhere—a task measured in decades—the entire AI economy depends on Taiwan remaining accessible, stable, and politically aligned with the countries that depend on its output.

The New Dependency Calculus

Dependencies are no longer purely economic. They are strategic. Governments are intervening in markets at scales not seen since the Cold War: industrial policy measures motivated by national and economic security—tariffs, direct equity stakes, export controls—rose more than sixfold between 2021 and 2026. Security is now understood to encompass not only military capability but broad swathes of the civilian economy, and AI supply chains are increasingly treated as national security infrastructure requiring protection, redundancy, and, when necessary, state control.

This reframing imposes a new calculus on companies. Efficiency once meant sourcing components from wherever they were cheapest and best suited to purpose. The new logic demands that efficiency be weighed against resilience, geographic diversification, and geopolitical alignment. Companies cannot simply ignore trade restrictions on the grounds that they are economically counterproductive. They must comply at significant cost, and prepare for future restrictions that do not yet exist but could materialize quickly.

The Dutch government's intervention in Nexperia illustrates the shift. Nexperia was not an acute or obvious national security threat; its profile was that of a commercially oriented semiconductor firm. But the Dutch government judged that Chinese ownership of any semiconductor assets—even partial ownership—was strategically unacceptable, and intervened unilaterally, overriding market logic. This is now standard operating procedure in many jurisdictions. Governments intervene when they perceive strategic risks, regardless of whether those interventions disrupt markets, violate commercial expectations, or complicate relations with trading partners. For companies like ASML—dependent on components from Japan, China, the United States, and Europe—export controls, sanctions, and political decisions can disrupt supply chains overnight, leaving corporate strategy perpetually hostage to overlapping and often contradictory government demands.

The AI Dependency Trap

The strategic tensions within AI supply chains are compounded by a structural fragility that runs deeper than any single chokepoint: the global economy has become increasingly dependent on the continuation of the AI investment cycle itself, and disruptions to that cycle propagate in ways that extend far beyond the technology sector.

If AI investment slows—whether from market saturation, regulatory action, or a loss of investor confidence—semiconductor demand contracts sharply. Because chip fabrication, lithography equipment, and materials supply are all capital-intensive industries with long investment horizons, even a moderate slowdown triggers layoffs, project cancellations, and potentially bankruptcies across multiple supply chain tiers. The feedback runs in reverse as well: supply chain disruptions that raise chip costs or constrain availability erode AI investment returns, cooling the demand that justifies continued semiconductor expansion. Neither dynamic can be easily interrupted once it begins.

Uncertainty itself functions as a damaging input. When companies cannot reliably project whether cross-border shipments will clear export controls, whether tariffs will change, or whether licensing reviews will conclude favorably, many pause new investment until conditions stabilize. Paused investment defers innovation, delays deployment, and creates gaps that better-positioned competitors can exploit. The practical consequence is that geopolitical interventions designed to secure the AI supply chain often introduce precisely the instability they are meant to prevent.

The costs of this fragmentation also fall unevenly. Large hyperscalers with established supplier relationships, stockpile capacity, and significant lobbying influence can absorb supply disruptions that would be fatal to smaller AI firms, research institutions, or national AI programs in lower-income countries. Every episode of supply chain tightening therefore compounds existing inequalities in who can access and develop AI capability—a dynamic explored further in the chapters on economic distribution and global development.

Strategic Scenarios and Long-Term Trajectories

The trajectory of AI trade dependencies is not predetermined. Three broad futures are identifiable, each with distinct implications for the industries, governments, and regions involved.

The first is managed interdependence: governments negotiate frameworks—bilateral or multilateral—permitting continued trade in critical AI inputs under defined conditions, with agreed limits on military applications and reciprocal verification mechanisms. Under this scenario, supply chains remain globally integrated but subject to layered oversight. The outcome preserves economic efficiency and reduces the risk of catastrophic disruption, but requires sustained diplomatic cooperation that current geopolitical conditions make difficult to sustain.

The second is bifurcated fragmentation: the U.S.-aligned and China-aligned blocs each develop parallel, largely self-sufficient AI supply chains, with minimal overlap. This produces incompatible technology standards, divergent AI architectures, and a global ecosystem divided along geopolitical lines. The scenario reduces mutual vulnerability but at enormous cost—duplicated capital investment, reduced innovation from the absence of global knowledge exchange, and constrained access for countries forced to choose between blocs.

The third, and most disruptive, is escalation and abrupt decoupling: a major geopolitical event—most plausibly a conflict involving Taiwan—collapses the existing supply chain architecture rapidly. Near-term consequences for global AI development would be severe: fabrication capacity disrupted, shipments halted, and years of investment stranded. Recovery would require restructuring AI supply chains under crisis conditions at timelines and costs that are difficult to project with confidence.

For the many companies, governments, and industries that depend on the current supply chain architecture, none of these trajectories offers clean resolution. Managing the resulting exposure requires constant navigation of competing demands, and no strategy eliminates the underlying risk. The era of efficiency-driven globalization in AI hardware, in which cost and capability were the dominant considerations, has ended. What has replaced it is a regime of strategic dependencies, managed vulnerabilities, and economic interdependence deployed as leverage in an ongoing geopolitical contest.

Key Takeaways

The global AI supply chain is highly concentrated at a small number of critical nodes—TSMC in Taiwan, ASML in the Netherlands, memory manufacturers in South Korea, rare earth refiners in China—and disruption at any of these points would cascade across the entire AI economy. This concentration reflects decades of efficiency-driven specialization, but it now represents a fundamental strategic vulnerability that no amount of short-term investment can quickly resolve.

Export controls, tariffs, and industrial policy interventions have become primary instruments of AI competition. The strategic logic is genuinely contested: restrictions may constrain adversaries in the short term while accelerating their motivation and capacity to build independent alternatives, risking a long-term bifurcation of the global technology ecosystem into incompatible blocs.

Rare earth minerals—particularly gallium and germanium, where China controls the overwhelming majority of global refining capacity—represent an underappreciated and structurally durable chokepoint. Rebuilding Western refining capacity would require years and substantial investment even under favorable political conditions.

Supply chain fragility and the AI investment cycle are mutually reinforcing sources of economic risk. Uncertainty introduced by trade policy interventions deters investment, and the costs of fragmentation fall disproportionately on smaller firms, research institutions, and countries in the developing world—compounding existing inequalities in AI access and capability.

No single trajectory for AI trade dependencies is assured, but three plausible futures—managed interdependence, bifurcated fragmentation, and abrupt decoupling—differ enormously in their consequences. The central challenge for governments, companies, and international institutions is to manage strategic dependencies without triggering the disruptions that supply chain security policies are designed to prevent.

Sources:

Last updated: 2026-02-25