Semiconductor Supply Chain: Capacity Expansion and Geopolitical Constraints

Executive Summary

Global semiconductor manufacturing investment has entered a structurally new phase: worldwide 300mm fab equipment spending is projected to hit $133 billion in 2026, an 18% annual increase, driven by surging AI chip demand and a growing commitment to regional supply chain self-sufficiency. The expansion is no longer purely demand-driven. Geopolitical restrictions, primarily US-led export controls on advanced chipmaking equipment and chips destined for China, have fractured the industry into two distinct technological trajectories that are widening rather than converging. In 2026, the Trump administration has recalibrated the US stance, downplaying new rule-making while approving higher-tier chip exports and suspending further restrictions, a shift that has provoked sharp congressional backlash and created policy whiplash for companies navigating China strategies. The result is a supply chain simultaneously expanding at record pace and fragmenting along geopolitical fault lines, creating both opportunity and sustained structural uncertainty for corporate planners.

The Ai-Subsidy Feedback Loop Driving Record Investment

The demand for memory has significantly increased due to AI training and inference. AI training has notably driven up demand for High Bandwidth Memory, while model inference has created substantial demand for storage capacity, boosting NAND Flash applications in data centers. This strong demand has led to sustained high levels of investment in the memory supply chain over the near and long term, helping to cushion potential downturns from traditional memory cycle fluctuations.

The interplay between commercial AI demand and state security policy creates a compounding dynamic that makes this cycle unlike previous semiconductor booms. In prior expansions, investment was driven almost entirely by market signals. In 2026, government grant programs in the United States, Japan, Europe, and South Korea are sustaining construction in locations and at scales that commercial returns alone would not justify. These expansions are supported by government subsidies including the US CHIPS Act, Japanese Government grants, and EU Chips Act funding. The broader strategic and economic implications are mutually reinforcing: fabs built today for security reasons become commercial assets over their 20-30 year operating lives, shifting future market share independent of today's demand signals.

The Logic and Micro segment is projected to lead equipment expansion with $175 billion in total investments from 2026 to 2028. SEMI's president and CEO Ajit Manocha described this as "a pivotal era of transformation, driven by unprecedented demand for AI-enabled technologies and a renewed focus on regional self-sufficiency," noting that "strategic global investments and collaboration are driving robust, advanced supply chains and faster deployment of next-generation semiconductor manufacturing technologies."

The broader systemic implications extend beyond chipmakers. Capital equipment companies, Applied Materials, Lam Research, and KLA Corp, are the true intellectual property holders in the AI supply chain, noting that equipment manufacturers are the decisive bottleneck constraining how quickly new capacity can come online. This leads to secondary effects in related domains: energy infrastructure, advanced packaging substrates, and specialist chemicals and gases all face parallel bottleneck constraints that compound the headline equipment story.

The Geographic Fracture: Two Semiconductor Supply Chains Taking Shape

The clearest structural consequence of the export control regime is not the technology gap between US and Chinese chipmakers, that gap is well-understood, but the accelerating geographic bifurcation of the global supply chain into two partially overlapping but increasingly separating ecosystems.

On the Western-aligned side, TSMC's $165 billion Arizona commitment represents the anchor investment. TSMC is transforming Arizona into the epicenter of American semiconductor manufacturing with a $165 billion investment that represents the largest foreign direct investment in United States history. The Taiwanese chipmaker's expansion plans, which now include up to 12 fabrication plants and four advanced packaging facilities near Phoenix, signal a seismic shift in the global semiconductor supply chain that could reshape AI chip production for decades.

In Japan, Rapidus, only founded in 2022, aims to enter 2nm production by 2027, receiving $11.5 billion in financial support, including $5.4 billion from the Japanese government.

On the Chinese-domestic side, the picture is one of genuine but uneven progress. SMIC's strategy in 2025-2026 includes mass production of Huawei's Ascend 910B AI chip and a planned doubling of 7nm capacity, with SMIC also entering pilot runs for its 5nm process, targeting mass production for partners like Huawei and Alibaba. However, the yield and cost constraints of producing near-leading-edge chips without EUV lithography remain significant. SMIC is expanding capacity across its production lines using qualified non-US equipment, including Chinese and Japanese tools, and appears to have secured enough ASML DUV systems to scale advanced-node production to roughly 70,000 wafers per month. SMIC's N+2 and N+3 nodes are nearing 7nm/5nm capabilities but capacity remains limited, estimated at around 10,000 wafers per month, far below market demand.

The resulting spillover affects multiple sectors beyond the semiconductor industry itself. China's mature-node expansion threatens to oversupply legacy markets in automotive, industrial, and consumer electronics, segments where Western firms currently hold competitive positions. The substantial projected increase in China's capacity to produce mid-to-low-tech logic chips over the next decade is projected to establish a pathway toward a self-sufficient ecosystem for strategic industries including autonomous vehicles and smart devices, which is particularly significant because most electronics and vehicles still rely on mature-node chips.

The Policy Tangle: Export Controls Under Negotiating Pressure

The interplay between US national security policy and trade negotiation strategy is creating both economic and operational uncertainty at a scale not previously seen in the semiconductor industry. The system in place as of mid-2026 is best described as layered ambiguity: rules that formally restrict leading-edge exports to China coexist with licensing pathways that allow some sales, a 25% tariff revenue mechanism, and active congressional pressure to tighten enforcement.

On December 8, President Trump announced that the administration would allow sales of Nvidia H200 and AMD MI325X GPUs to China in exchange for a 25% fee for the US government. On January 15, 2026, BIS issued a final rule implementing this policy, revising the export licensing review policy for certain advanced computing items exported to China, shifting from a presumption of denial to a case-by-case review.

Congressional resistance has been fierce. The House Committee on Foreign Affairs chair pushed the AI OVERWATCH Act through committee on January 22, 2026, which would grant Congress veto power over AI chip export licenses, authority that currently belongs to the Department of Commerce. The chair of the House Select Committee on China has written several letters to Commerce Secretary Howard Lutnick opposing the export of H200 chips and calling for restrictions on China's access to advanced semiconductor manufacturing equipment.

The enforcement gap is arguably as consequential as the rules themselves. The United States issued a notice affirming its restrictions on shipments of semiconductors to subsidiaries of Chinese companies located outside China, with the Department of Commerce clarifying that its licensing requirements applied to all businesses with headquarters or a parent company in China, a clarification made necessary because the loophole had been actively exploited. A gray market dynamic exists where distributors in Singapore and Malaysia are diverting restricted hardware toward China through shell companies and falsified documents.

Tokyo Electron's experience illustrates the commercial cost of this policy environment. Tokyo Electron's China sales fell from 279.4 billion yen to 175.5 billion yen in a recent fiscal quarter as Japanese manufacturers pivot toward AI demand to offset geopolitical headwinds, a pattern SCREEN Holdings, Advantest, and Nikon are navigating in parallel. This economic pressure on allied equipment makers translates directly into reduced R&D budgets precisely when the technology competition demands the opposite.

The Mature-Node Oversupply Trap

A risk receiving less attention than the leading-edge race is the structural oversupply building in mature-node capacity. Chinese semiconductor manufacturing prioritizes scaling volume production in mature nodes due to US export restrictions limiting access to technology. Manufacturers are targeting 28nm production, with SMIC's Shanghai and Tianjin fabs, each costing $7-9 billion, targeting 28nm and expected to start operations in 2025-2026.

Advanced nodes account for less than 20% of the global semiconductor market by product volume, while more than 80% of demand comes from mature-node and specialty-process segments. China's strategy of dominating this 80% is rational from a national security standpoint, it secures the chips that power most industrial, automotive, and consumer applications, but it creates a deflationary shock for the global companies that currently supply those markets. With Taiwanese foundries shifting capacity and raising prices, customers in high-voltage and image sensor applications are increasingly turning to Chinese foundries for more stable pricing and capacity, a trend evident since the second half of 2025.

The geopolitical and economic implications are mutually reinforcing in a destabilizing way. Western chip producers invested heavily in mature-node capacity during the post-COVID shortage cycle. If Chinese foundries displace them in volume markets through subsidized pricing, a pattern the G7 has explicitly flagged as a concern in the context of what the Turing Institute describes as the risk of "dumping" — the revenue base sustaining Western firms' leading-edge R&D erodes precisely as the technology competition intensifies.

Securitization Theory Analysis

Securitizing Actor: The United States government, acting through the Bureau of Industry and Security, the White House, and supporting congressional committees, is the primary securitizing actor. Japan and the Netherlands function as aligned secondary actors who have adopted parallel restrictions under US diplomatic pressure.

Referent Object: The referent object is US technological leadership and, more specifically, the AI computing stack, defined by policymakers as foundational to future military, economic, and strategic power. The Congressional Research Service framing captures this: semiconductors are described as "strategic and uniquely important," "fundamental to most industrial and national security activities," and "essential building blocks of artificial intelligence."

Existential Threat Construction: The threat is framed not as an immediate military challenge but as a structural capability transfer, the idea that allowing China to acquire leading-edge chipmaking tools would permanently and irreversibly shift the balance of AI development capacity. CSIS analysis frames the stakes as: "The outcome of the chip innovation race will determine which country leads in the development and application of AI, with major strategic and economic security implications." The word "irreversible" does specific rhetorical work: it moves the issue beyond normal trade policy into emergency-register language justifying anticipatory restrictions.

Target Audience: The primary audience is the domestic US legislative and executive branch. Allied governments (Netherlands, Japan, South Korea) are a secondary audience whose alignment is sought through diplomatic negotiation rather than unilateral mandate. The export control framework also implicitly addresses US technology firms, whose commercial behavior is being constrained by the security framing.

Extraordinary Measures: The extraordinary measures in place include unilateral extraterritorial controls applied to foreign-produced goods containing US technology (the Foreign Direct Product Rule), entity-list designations bypassing normal trade dispute processes, pressure on allied governments to align their export regimes, and in 2026, a proposed congressional veto over executive branch licensing decisions, an unusual intrusion on executive trade authority.

Classification: SECURITIZED

The semiconductor-AI nexus has moved beyond politicization. Emergency measures, extraterritorial controls, entity lists, allied alignment negotiations, congressional veto proposals, are fully operational, not debated. The US domestic political consensus across both parties supports the security framing even where they disagree on implementation details.

Process Tracing Analysis

Cause and Outcome: Cause: the US October 2022 export control package restricting advanced chip and equipment exports to China. Outcome: accelerated Chinese domestic semiconductor investment and the bifurcation of global supply chains into two partially separating ecosystems.

Causal Mechanism Chain:

• Step 1: BIS imposes controls on advanced chips and semiconductor manufacturing equipment exports to China in October 2022, tightened in October 2023 and December 2024.
• Step 2: Chinese access to leading-edge ASML EUV tools is cut off; advanced chip imports for AI and supercomputing become restricted.
• Step 3: Beijing responds with "Big Fund III" ($47.5 billion, May 2024) and accelerated domestic equipment substitution mandates, the Chinese government requiring new fab installations to contain 50% domestic equipment.
• Step 4: SMIC, Hua Hong, YMTC, and CXMT redirect investment toward capacity China can build without foreign tool dependency, particularly mature-node volumes.
• Step 5: Allied firms (Tokyo Electron, ASML) face revenue loss from China business restrictions, creating commercial pressure to lobby for carve-outs.
• Step 6: US allies partly comply but implementation varies; ASML DUV sales to China reportedly continued at elevated levels through 2024 before Dutch restrictions tightened.
• Step 7: Global supply chain splits: leading-edge AI chip production concentrates in the US-allied orbit; mature-node production shifts toward China.

Evidence Assessment:

• The causal link between October 2022 controls and China's accelerated domestic investment is supported by a smoking gun: China launched Big Fund III in May 2024 at $47.5 billion, explicitly framed as a response to export controls, and Yole Group and TrendForce both document the acceleration in domestic equipment adoption.
• The allied compliance dynamic carries hoop test evidence: Japan imposed restrictions on 23 categories of semiconductor manufacturing equipment in July 2023, aligning with US and Dutch efforts, a necessary step without which Chinese access via third parties would have remained largely intact.
• The bifurcation outcome is a straw-in-the-wind observation consistent with the mechanism but also consistent with commercial forces predating 2022.

CAUSAL_MECHANISM_STRENGTH: MODERATE

The core causal pathway from US controls to Chinese domestic investment acceleration is well-evidenced. The longer-term outcome, permanent supply chain bifurcation, rests on assumptions about political durability that remain contested and dependent on future policy choices.

Constructivism Lens Analysis

Actor Identities: The United States projects the identity of "defender of the rules-based technological order" — a state that sets the terms of access to advanced technology to preserve a security architecture it anchors. China projects the identity of a rising power entitled to technological sovereignty, systematically denied tools it needs. This identity contest is not merely rhetorical: it structures the domestic politics of both states and constrains the diplomatic space for compromise.

Operative Norms: Two competing norms are in direct tension. The norm of open technology markets and WTO-consistent trade, which historically governed semiconductor commerce, is being displaced by an emerging norm of "technology security zones" in which access to critical technologies is conditioned on alliance status. CSIS analysts note that China's response involves building its own norm: the idea that technology self-sufficiency is a legitimate national security objective requiring state coordination rather than market reliance.

Intersubjective Meaning: The same event, a US export control, is understood entirely differently by each actor. Washington reads it as a proportionate, targetted restriction on dual-use technology with military implications. Beijing reads it as an attempt to permanently subordinate China's technological development to US permission. These competing readings are not resolvable by evidence alone, because they reflect foundational disagreements about the legitimate scope of US technology leadership. This divergence is precisely why neither side treats the other's stated intentions as reliable signals.

Norm Lifecycle Stage: The "technology security zone" norm is in active cascade. A tipping point occurred with the October 2022 controls: before that date, unilateral extraterritorial chip controls of this scope were essentially unprecedented. Since 2022, Japan, the Netherlands, and to varying degrees South Korea and Taiwan have adopted parallel frameworks. The G7 June 2026 coordination commitment on critical minerals, drawing on the same logic, signals the norm is spreading beyond semiconductors into adjacent material domains.

Ideational vs. Material: A purely material analysis would predict that US firms would resist export controls because China represents their largest market, and the commercial lobbying against controls has been persistent. That the controls were implemented and sustained despite this pressure reflects ideational factors: the national security frame had captured sufficient institutional consensus that commercial objections were overridden. The material analysis also misses why China cannot simply buy its way out by offering higher prices for restricted tools: the norm of allied alignment means the tools are politically inaccessible, not just commercially priced.

Norm Lifecycle: CASCADE

Counterarguments

1. The bifurcation narrative overstates the clean break between supply chains. TSMC maintains a Validated End-User authorized fab in Nanjing, China. Samsung and SK Hynix retain China operations under special exemptions. ASML DUV tools flowed to China at elevated rates through at least 2024. The evidence base, drawing heavily on US government and allied government sources, may overstate the coherence of the restriction regime and understate the degree to which commercial interdependence persists. A supply chain that is "bifurcating" is not the same as one that has bifurcated, and the transition costs of full decoupling, which no government has formally committed to, remain unknown and potentially prohibitive.

2. The US policy incoherence finding may be anchored on a transitional moment rather than a stable equilibrium. The H200 reversal, the 25% tariff mechanism, and the licensing framework were introduced within a specific negotiating context in late 2025 and early 2026. Characterizing the overall control regime as "incoherent" risks missing the strategic logic: managed access at lower chip tiers may be a deliberate choice to maintain US leverage while preserving some commercial relationship. Enforcement capacity, not new rulemaking, is now the primary instrument of US leverage, which is a coherent strategy even if it looks disorderly from the outside.

3. China's self-sufficiency timeline is anchored on industry-friendly projections that have a track record of overoptimism. The "Made in China 2025" plan targeted 70% semiconductor self-sufficiency by 2025; China achieved approximately 33% by 2024, less than half the target. Electronics Weekly and TrendForce note that the new 80% by 2030 target is being set by industry executives with commercial incentives to attract state funding. The Turing Institute finds that while mature-node progress is real, full autonomy in advanced logic and lithography remains "structurally out of reach." Analysts should apply a significant discount to China's headline self-sufficiency targets when constructing base-case scenarios.

Indicators To Watch

Decision Relevance

Scenario A (~55%): Managed bifurcation continues, controls hold at the equipment level, chip sales to China remain partially accessible under licensing, and TSMC-led Western-aligned capacity expands as planned. Recommended: Accelerate supply chain audits to map exposure to Chinese mature-node suppliers. Engage CHIPS Act incentive programs for facilities qualifying for domestic content provisions. For technology-dependent manufacturers, model a dual-sourcing posture that maintains legacy-node flexibility through both Chinese and non-Chinese channels. Do not assume current chip access to China is permanent, build product lines that can absorb a licensing restriction tightening cycle within 12 months.

Scenario B (~30%): US-China diplomatic deal partially relaxes chip controls in exchange for other concessions, creating a window of broader China market access. Recommended: Prepare contingency commercial plans for re-entry into Chinese AI infrastructure markets, but treat any re-entry as short-cycle. A deal-based relaxation is moderate-to-high confidence to face both congressional challenge and future reversal; long-duration capital committed on the basis of restored China access carries asymmetric downside risk. Monitor the AI OVERWATCH Act's legislative progress as a leading indicator of how quickly congressional override could apply.

Scenario C (~15%): Chinese domestic lithography breakthrough or third-country diversion materially narrows the leading-edge technology gap faster than the 2030 baseline. Recommended: Immediately revalue the competitive moat of advanced-node capacity investments in allied geographies. If the SMIC-Huawei ecosystem achieves commercial-scale 5nm production by 2028, the pricing and strategic differentiation assumptions underlying current fab investment decisions in Arizona, Japan, and Germany require recalibration. Establish early-warning monitoring protocols tied to SMIC yield disclosures and Huawei Ascend chip shipment volumes.

Analytical Limitations

• China's advanced-node yield rates and effective wafer capacity are not publicly disclosed and are assessed primarily from inference, export data, and secondary industry sources. The picture could be materially more advanced or more constrained than available evidence suggests.
• The Trump administration's export control policy is in active negotiation posture, and the balance between enforcement and diplomatic accommodation can shift rapidly. Analysis current as of June 2026 may not reflect decisions taken in subsequent weeks.
• SEMI and SEMI China projections for regional capacity share and equipment spending are based on disclosed capex plans and fab construction announcements, not verified operational output. Delays in TSMC Arizona, Intel's 18A ramp, and Rapidus' 2nm timeline, all of which carry documented risk, could significantly alter the Americas capacity share trajectory.
• The analysis draws on trade press, think tank, and government sources concentrated in the US, Japan, and allied jurisdictions. Chinese domestic semiconductor industry reporting carries participation bias: both state sources that overstate progress and commercial sources that understate it to attract investment create a noisy signal environment.
• Potential anchoring bias toward the bifurcation framing is acknowledged: the evidence base consulted is heavily weighted toward US-aligned analytical institutions whose framing treats decoupling as directionally inevitable. Alternative framings that treat current restrictions as a temporary negotiating posture with high reversal probability deserve greater analytical weight than the consensus view allows.