The Chip War Isn't Really About Chips

A chip smaller than a fingernail can hold up a car, a data center, or a weapons system. That sounds like a contradiction: how can something so tiny matter so much? Because the chip is only the visible end of a production system that spans design software, specialized equipment, chemicals, fabrication plants, packaging, and skilled labor. A bottleneck at any one stage can matter more than the number of chips on a loading dock.

That is why the contest over semiconductors is not simply a race to build the fastest processor. It is a struggle over strategic dependence: who can make the most capable chips, who can reach the tools required to make them, and who can withstand a disruption—or impose one. Governments call this national security. Companies call it supply-chain risk. Both are describing the same economic fact from different sides.

One product, many chokepoints

There is no single “chip supply chain.” A semiconductor may be designed in one country, fabricated in another, and assembled and tested somewhere else. The stages require different expertise, companies, and machinery. Some firms specialize in design and outsource manufacturing; others, including Taiwan Semiconductor Manufacturing Company (TSMC), specialize in manufacturing chips designed by customers.

The model is efficient precisely because it is specialized. It also means national self-sufficiency is not a realistic near-term description of how the industry works. The OECD’s 2025 mapping describes a complex, globally distributed chain with concentrated critical inputs, specialized economies, and growing trade dependencies. It argues that resilience depends in part on diversification and international cooperation, not just on counting factories inside national borders. [1]

In a separate analysis using 2018 data, the OECD found that five economies generated about 75% of semiconductor-industry value added, four of them in Asia. It also noted that concentration differs by chip type: in 2021, the top three exporters of memory chips accounted for more than 80% of exports, while the top three for logic and other chips accounted for more than 55%. Those are measures of economic activity and exports, not a claim that all chips are made in one place. [2]

For advanced logic—the chips central to leading-edge computing—the concentration is sharper. The OECD cited 2022 evidence that only TSMC in Taiwan and Samsung in South Korea produced the most advanced logic chips at the time. Its newer capacity study emphasizes why a map of factories is not enough: production plants are not freely interchangeable, and a disruption in one facility cannot necessarily be offset by another making a different chip or using a different process. [2] [3]

Taiwan stands out, but it is not the whole story. A U.S. Commerce Department market guide says Taiwan accounts for more than 60% of global foundry revenue and more than 90% of leading-edge chip manufacturing; it put the island’s semiconductor-industry revenue above $165 billion in 2024. Revenue shares and output shares are not identical measures, and the guide does not mean Taiwan makes every kind of semiconductor. The figures do show why a crisis around the island would be an economic shock well beyond the region. [4]

Why geography became security policy

A factory’s location has become a geopolitical question because chips are both ordinary commercial inputs and building blocks for advanced computing, artificial intelligence, and military systems. The U.S. government’s stated case for restricting certain exports to China is that advanced chips and the equipment used to make them could support military modernization. That is the government’s rationale, not proof that every chip or every Chinese buyer has a military purpose. [5]

China has also made technological independence an explicit policy aim. A Congressional Research Service report describes the State Council’s 2020 framework for supporting the integrated-circuit industry and software, including incentives for capabilities across design, fabrication, equipment, software, packaging, and materials. That documents a state-led industrial strategy; it does not, by itself, establish that every Chinese company or investment acts on government direction. [6]

Export controls work through the chain’s dependencies. They can restrict specified chips, manufacturing equipment, design tools, software, or transactions involving listed entities. In December 2024, the U.S. Bureau of Industry and Security announced controls covering 24 types of semiconductor-manufacturing equipment, three types of software tools, high-bandwidth memory, and additions to the Entity List. BIS said the package aimed to impede China’s ability to produce advanced-node chips for military applications. [7]

But these rules do not make the chain disappear, nor do they amount to a blanket embargo on every semiconductor. CRS noted in 2025 that some advanced items can still be licensed and that China retains access to portions of the chain, including mature-node technologies, some equipment and tools, research, materials, and third-country computing. It also outlines potential workarounds and enforcement gaps. The practical reach of a control depends on technical thresholds, licensing decisions, allied cooperation, and whether firms can route around it. [5]

Policy can change, too. A January 2026 Federal Register rule moved certain exports of advanced-computing chips—including Nvidia’s H200 or equivalents—to China and Macau from a presumption of denial to case-by-case review, if specified conditions are met. Those include certification of adequate U.S. supply, no diversion of foundry capacity from U.S. customers, customer security procedures, and independent testing. Other covered transactions remain subject to a presumption of denial. The change illustrates a live tension: restricting access to capabilities while also weighing commercial supply, U.S. firms’ position, and control over how products are used. [8]

Subsidies are part of the contest

The other side of economic security is building capacity at home. In November 2024, the U.S. Commerce Department announced a final award of up to $6.6 billion in direct CHIPS Act funding and up to $5 billion in loans to TSMC Arizona, supporting a planned investment of more than $65 billion in three Phoenix-area fabs. The grant is milestone-based; the investment and job numbers in the announcement are plans and estimates, not all completed outcomes. [9]

The public case is to create a more reliable domestic source for strategically important chips and attract a cluster of suppliers and workers. There is evidence of progress: TSMC says its first Arizona fab began high-volume production of N4 chips in the fourth quarter of 2024. But Arizona is not an overnight replacement for Taiwan’s manufacturing ecosystem. The U.S. government’s project page gives later target dates for the second and third fabs, and describes advanced packaging as a future capability. Fabrication capacity is only one link; supply reliability also depends on process maturity, yields, materials, workforce, customer qualification, and packaging. [10] [11]

“Reshoring” can therefore be a misleading finish line. Building another fab changes the geography of some production. It does not automatically reproduce the network of suppliers, trained teams, customers, and complementary facilities that took decades to assemble. Nor does it guarantee that a fab can substitute for a different factory’s output during a crisis. The OECD’s latest capacity mapping specifically warns that production capacity is concentrated and that substitutability among plants is limited. [3]

What this means

The chip war is better understood as competition to shape interdependence than as a race toward total independence. Governments want fewer single points of failure and more control over technologies that could carry military consequences. Companies still depend on cross-border expertise and markets to make production viable. The two aims can coexist, but they also collide: duplicating capacity is costly, while concentrating capacity is risky. For consumers, the conflict may surface as delays or higher costs, but not every geopolitical headline translates into an immediate shortage. The effects depend on what is disrupted, how long it lasts, which chips are affected, inventories, and whether another qualified source exists. A foundry’s capacity figure alone cannot answer those questions. For policymakers, the key distinction is between resilience and autarky. More geographic options can reduce exposure. Yet controls that are too broad may encourage rival systems, invite retaliation, or make coordination with equipment-producing allies harder. These are policy trade-offs, not settled outcomes. The OECD calls for better data and cooperation; the U.S. controls and subsidy programs show Washington’s chosen instruments, but their long-term industrial and security results remain uncertain. [1] [3] [5]

The question nobody asks

If a government says it wants “secure supply,” secure for whom—and secure against what? A factory on national soil may still rely on imported chemicals, equipment, intellectual property, or customers abroad. A local chip supply can be less exposed to one risk and more exposed to another. And a company that can manufacture a chip is not necessarily able to design it, package it, or replace another firm’s output at short notice. The less dramatic but more useful question is not “Where are the chips?” It is “Which specific capability would fail first, who else can provide it, and how long would substitution take?” The available public data do not answer that for every process, firm, or crisis. The OECD’s call for better disaggregated capacity data reflects this limitation. A supply chain cannot be made resilient by rhetoric alone; it requires knowing where its irreplaceable steps are, and accepting that diversification has a cost. [3]