Supply chain realities¶
The global semiconductor supply chain has achieved a level of centralisation and fragility that would make any competent risk manager reach for the emergency whisky while updating their CV. The majority of the world’s most advanced chips are manufactured by a single company, in a single country, using equipment from a handful of suppliers, all of which exists in a geopolitical situation that could charitably be described as “tense” and more accurately described as “one diplomatic incident away from catastrophic.”
TSMC, the Taiwan Semiconductor Manufacturing Company, occupies a position in the global technology supply chain comparable to The Patrician’s position in Ankh-Morpork. They’re not officially in charge of everything, but practically nothing happens without their involvement, and if they were suddenly removed, the resulting chaos would be spectacular, expensive, and probably involve at least one small war. Unlike The Patrician, TSMC’s monopoly wasn’t achieved through clever political manoeuvring but through decades of technical excellence, massive capital investment, and competitors deciding that semiconductor manufacturing was too difficult and expensive to pursue independently.
This arrangement worked adequately when demand for advanced chips was predictable and geopolitical tensions were manageable. Neither of these conditions currently holds. AI has created unprecedented demand for cutting-edge semiconductor manufacturing capacity at exactly the moment when the geopolitical situation surrounding Taiwan has become substantially more concerning. The result is a supply chain structure that combines maximum economic efficiency with maximum strategic vulnerability, which is either brilliant or catastrophic depending on whether you’re optimising for peacetime or preparing for disruption.
TSMC’s manufacturing dominance¶
TSMC manufactures approximately 90 percent of the world’s most advanced semiconductors, including the chips that power AI accelerators, high-end processors, and most cutting-edge computing systems. This dominance isn’t accidental or the result of anti-competitive behaviour. It’s the consequence of sustained technical leadership, enormous capital investment, and competitors either failing to keep pace or deciding that competing at the leading edge wasn’t economically viable.
Manufacturing semiconductors at 5 nanometre process nodes and below requires equipment costing hundreds of millions of euros per machine, fabrication facilities costing tens of billions of euros to build, and expertise accumulated over decades. TSMC has made these investments consistently while maintaining manufacturing yields and reliability that competitors struggle to match. When Nvidia, AMD, Apple, and nearly every other major chip designer need cutting-edge manufacturing, they turn to TSMC because the alternatives are inferior, more expensive, or don’t exist.
Intel, historically the world’s leading semiconductor manufacturer, has fallen behind TSMC in process technology after decades of leadership. Samsung manufactures advanced semiconductors but with lower yields and less consistent quality than TSMC. GlobalFoundries abandoned the leading edge entirely, concluding that competing with TSMC was economically unviable. This leaves TSMC in a position of effective monopoly for the most advanced semiconductor manufacturing, which is excellent for TSMC’s business and concerning for everyone who depends on their services.
The concentration creates several problems simultaneously. Manufacturing capacity is finite and allocated among customers through negotiation, long-term contracts, and TSMC’s assessment of strategic priorities. When demand surges, as it has for AI accelerators, not all customers can get the capacity they want when they want it. Lead times extend, prices increase, and customers must either wait, compromise on specifications, or find alternatives that generally don’t exist for cutting-edge processes.
The capital requirements for building additional fabrication capacity are so enormous that TSMC cannot quickly respond to demand surges. A new fabrication facility takes years to build and costs €15 to €20 billion before producing a single chip. TSMC is expanding capacity continuously but cannot match sudden demand increases. This means prolonged periods where demand exceeds supply, customers compete for limited capacity, and TSMC effectively decides which products get manufactured through capacity allocation.
Geopolitical risks and the Taiwan situation¶
Taiwan’s geopolitical situation adds catastrophic risk to the already precarious semiconductor supply chain. China considers Taiwan a breakaway province and has not ruled out military reunification. The United States has strategic interests in Taiwan’s independence and provides security guarantees of varying clarity. This situation has existed for decades but has become more tense as China’s military capabilities have grown and as nationalist rhetoric has intensified on both sides.
If military conflict occurred, or if even the credible threat of conflict disrupted Taiwan’s economy, the global semiconductor supply would experience a crisis of unprecedented severity. TSMC’s fabrication facilities cannot be quickly relocated. They’re enormous physical infrastructure deeply integrated into Taiwan’s economy and dependent on stable power, water, and logistics. Military action, or even heightened tensions that disrupted operations, would halt production of most advanced semiconductors globally.
The economic consequences would be extraordinary. Every industry dependent on advanced semiconductors would face immediate shortages. AI development would halt for lack of GPUs. Smartphone production would stop. Automotive manufacturing would face critical component shortages. Data centres couldn’t expand or refresh hardware. The global economy is sufficiently dependent on semiconductors that disrupting TSMC’s operations would trigger economic crisis measured in trillions of euros of lost output.
Various governments have recognised this risk and are attempting mitigation through domestic semiconductor manufacturing investments. The United States is subsidising construction of fabrication facilities domestically through the CHIPS Act. The European Union has similar programmes encouraging semiconductor manufacturing within Europe. China is investing heavily in domestic semiconductor capability to reduce dependence on foreign suppliers. These efforts are sensible but cannot quickly resolve the concentration risk.
Building semiconductor manufacturing capability takes time measured in years to decades. Constructing fabrication facilities is only part of the challenge. Developing the engineering expertise, supply chains, and equipment manufacturing capacity requires sustained investment and cannot be rushed. The subsidies announced recently will eventually reduce concentration but won’t eliminate dependence on Taiwan for most advanced semiconductors until the 2030s at earliest, and possibly longer if technical challenges prove more difficult than anticipated.
The situation resembles having all of Ankh-Morpork’s food supply dependent on a single farm located in contested territory between Ankh-Morpork and a rival power. Everyone understands this is dangerous. Everyone agrees it should change. Nobody can change it quickly because relocating the farm requires years of infrastructure development. Meanwhile, everyone hopes nothing disrupts the current arrangement while being acutely aware that disruption is plausible.
Competition for fabrication capacity¶
When demand for advanced semiconductors exceeds available manufacturing capacity, customers must compete for limited allocation from TSMC and other manufacturers. This competition takes several forms, all of them expensive and potentially damaging to companies that lose out.
Long-term supply agreements provide guaranteed capacity but require committing to purchase volumes years in advance. This is reasonable for established products with predictable demand but problematic for new products or rapidly growing markets where demand is uncertain. Companies must choose between guaranteeing capacity they might not need or risking insufficient capacity if demand exceeds expectations. Getting this balance wrong is expensive either way.
Premium pricing for priority access allows customers to jump queues by paying more, which TSMC accommodates through various mechanisms that aren’t publicly transparent but definitely exist. This works if you have budget flexibility and desperate need for capacity. It doesn’t work if your margins are thin or if multiple customers are simultaneously willing to pay premiums, which just raises everyone’s costs without increasing total capacity.
Alternative manufacturing partners provide some diversification but generally at lower performance, higher cost, or both. Samsung manufactures advanced semiconductors but with less consistent yields than TSMC. Intel is opening their fabrication facilities to external customers but their process technology lags TSMC’s leading edge. Older process nodes at various manufacturers are available but don’t provide the performance and efficiency that cutting-edge applications require.
Design compromises to use less advanced manufacturing processes allow using capacity that’s more readily available but sacrifice performance. An AI accelerator designed for 5 nanometre manufacturing could be redesigned for 7 nanometre or older processes, which would reduce performance and efficiency but might be manufacturable with less delay. This is strategic retreat from optimal design to achieve manufacturability, which is sensible if the alternative is not being able to manufacture at all.
The competition for capacity affects product development timelines, costs, and capabilities across the industry. Companies designing products requiring cutting-edge manufacturing must secure capacity allocations years in advance, which requires predicting future demand with minimal information. Product announcements increasingly mention manufacturing capacity as a constraint, which was rarely discussed when capacity was more abundant. The entire industry is collectively learning that manufacturing is a strategic constraint requiring attention at executive levels rather than just operational management.
The fragility of specialisation¶
The semiconductor supply chain has optimised for efficiency through extreme specialisation. Companies focus on what they do best and depend on specialists for everything else. This works brilliantly under stable conditions and fails catastrophically when any component of the chain is disrupted.
ASML manufactures the extreme ultraviolet lithography machines required for advanced semiconductor manufacturing. They’re the only company that makes these machines. Each machine costs approximately €150 million and takes years to manufacture. TSMC’s expansion depends on obtaining these machines, which means TSMC’s capacity growth is constrained by ASML’s manufacturing capacity, which creates another single point of failure in the semiconductor supply chain.
Materials and chemical suppliers provide hundreds of specialised inputs required for semiconductor manufacturing. Many of these materials come from single suppliers or have limited alternatives. Disruption to any critical material can halt manufacturing even if fabrication facilities and equipment remain operational. The entire supply chain is deeply interdependent in ways that are invisible until something breaks.
Packaging and testing after fabrication involves additional specialised companies, facilities, and processes. Even after TSMC manufactures a chip, it must be packaged, tested, and validated before shipping to customers. These steps have their own capacity constraints, quality challenges, and geographical concentrations. A complete semiconductor supply chain involves dozens of companies, multiple countries, and hundreds of interdependent processes, all of which must work correctly for the final product to ship.
This specialisation has driven productivity and innovation but has created a system where disrupting any component potentially disrupts the entire chain. The pandemic illustrated this when lockdowns, logistics disruptions, and demand surges combined to create semiconductor shortages that persisted for years and affected industries from automotive to consumer electronics. The supply chain has somewhat recovered but remains vulnerable to similar disruptions in future.
The economic logic driving specialisation is sound. Companies focusing on specific capabilities and achieving scale efficiency produce better products at lower costs than vertically integrated alternatives. The strategic vulnerability is equally sound. Highly specialised, geographically concentrated supply chains optimised for efficiency are fragile under disruption. These aren’t contradictory observations. They’re trade-offs between efficiency and resilience that were optimised for the former because disruptions were historically rare enough that the efficiency gains justified the risk.
Attempts at diversification and their limitations¶
Various efforts are underway to reduce dependence on TSMC and diversify semiconductor manufacturing geographically. These efforts are necessary, expensive, and won’t quickly resolve the concentration that accumulated over decades.
Intel’s foundry services aim to become a viable alternative to TSMC by opening their manufacturing to external customers. This could reduce concentration if Intel achieves process technology and manufacturing quality competitive with TSMC. Current evidence suggests this will take years of sustained investment and technical development. Intel’s recent manufacturing challenges indicate the difficulty of competing with TSMC’s technical leadership.
Samsung’s foundry business provides some alternative to TSMC but hasn’t captured market share commensurate with their capabilities because yield and reliability concerns lead many customers to prefer TSMC despite capacity constraints. Improving this requires sustained technical excellence and quality consistency, which Samsung is pursuing but which takes time to achieve and demonstrate to risk-averse customers.
Government subsidies for domestic semiconductor manufacturing in the United States, Europe, and China will eventually increase capacity and geographical diversity. The CHIPS Act in the United States allocated €52 billion for semiconductor manufacturing and research. The European Chips Act has similar goals with comparable funding. These investments will build additional fabrication facilities outside Taiwan over the next decade.
However, these subsidies face several limitations. Building fabrication facilities is necessary but insufficient for competitive semiconductor manufacturing. The facilities require skilled workforce, equipment supply chains, materials suppliers, and accumulated expertise that cannot be quickly replicated. TSMC’s advantage comes partly from their facilities but largely from decades of manufacturing experience and continuous improvement. New fabrication facilities will take years to achieve comparable expertise.
The economics of semiconductor manufacturing benefit from scale and specialisation. Distributing manufacturing across multiple smaller facilities for strategic resilience increases costs compared to concentrating in a few large facilities for efficiency. Governments are effectively subsidising strategic resilience at the expense of economic efficiency, which is defensible for critical industries but still means higher costs and potentially reduced innovation as resources are spread across multiple efforts rather than concentrated.
The timeline for meaningful diversification is measured in decades rather than years. Fabrication facilities being built now will begin operations in the mid to late 2020s. Achieving the manufacturing expertise and ecosystem development that makes them competitive with TSMC will take additional years. Reducing dependence on Taiwan-based manufacturing from 90 percent to perhaps 60 or 70 percent might be achievable by 2030 if current investments proceed successfully. Eliminating the dependence entirely is implausible for the foreseeable future.
Living with concentrated supply chains¶
The realistic outlook is that semiconductor manufacturing will remain geographically concentrated and dominated by a few companies for the next decade at minimum. This concentration represents strategic vulnerability that cannot be quickly resolved given the technical challenges and capital requirements of semiconductor manufacturing. Companies, governments, and industries must navigate this reality rather than assuming it will change quickly.
Risk management requires acknowledging that semiconductor supply disruptions are plausible and preparing accordingly. Maintaining buffer inventory where possible, designing products with manufacturing flexibility, and developing contingency plans for shortage scenarios are sensible precautions. These mitigations are expensive and reduce efficiency but provide insurance against supply chain disruptions.
Strategic relationships with semiconductor manufacturers matter more when capacity is constrained. Companies with strong relationships, predictable demand, and long-term commitments receive priority when capacity is limited. Building these relationships requires lead time and commercial commitment but provides resilience when supply tightens.
The geopolitical situation surrounding Taiwan demands attention from anyone dependent on TSMC’s manufacturing. Monitoring tensions, understanding potential disruption scenarios, and planning responses isn’t paranoia but prudent risk management given the concentration of critical manufacturing in a geopolitically contested region.
The semiconductor supply chain will gradually diversify as government subsidies enable new fabrication facilities and as companies invest in alternative manufacturing partners. This diversification will partially mitigate concentration risk over the next decade but won’t eliminate it. Planning should assume that TSMC remains the dominant manufacturer of advanced semiconductors and that Taiwan’s geopolitical situation remains a potential source of catastrophic supply disruption.
The situation is uncomfortable for everyone except TSMC, who are profiting handsomely from their dominant position while being quite aware that their geopolitical situation creates risks that even extraordinary manufacturing expertise cannot entirely mitigate. It’s a supply chain structure that worked adequately when times were calm and demand was stable but reveals its fragility under the combined pressures of surging AI demand and increasing geopolitical tensions. Whether this fragility results in actual crisis or merely persistent anxiety depends on geopolitical developments that are outside the control of chip designers, manufacturers, and the many industries that depend on their continued ability to produce the semiconductors on which modern technology depends.