Computing’s Single Point of Failure: Why the Semiconductor Supply Chain Is Tied to a Few Places on the Map

Evidence-first notes on bioscience and deep tech, at the edge of the lab and the market. Information only — not investment advice.

The 30-second version

• What. The ultimate bottleneck for AI and computing is neither chip performance (physics) nor capital (economics), but a geopolitical structure in which advanced logic, EUV lithography, HBM, and upstream materials are extremely concentrated in a handful of actors and geographies — a perspective synthesized from a source asset graded VERIFIED.
• So what. TSMC’s ≤7nm accounts for 74% of wafer revenue (2026 Q1), Taiwan makes roughly 90% of the world’s most-advanced chips, ASML is the sole maker of EUV scanners, and three firms hold about 97% of HBM — a tail-risk structure in which disruption at a single node could halt global compute.
• Now what. Read reshoring and diversification with care: much of it is announced (projected) capacity, not operating capacity. The figures here are a map of structural risk, not a buy/sell or policy signal for any equity or country.

(Note: figures and outlooks below attribute projections/announcements separately from measured values. Diversification is genuinely underway, but on an operating basis production remains concentrated in Taiwan and a few nodes — that is the central caveat.)

The five-minute read

The bottleneck was always “somewhere else”

The through-line of this series is that computing’s bottleneck is usually not where people point (clock speed, transistors). Part 1 named power, Part 2 packaging and memory, Part 3 economics. Part 5 looks at the layer beneath all of them — the single point of failure: advanced logic (TSMC/Taiwan), EUV lithography (ASML/Netherlands), HBM (three firms), and upstream materials (Japan, Ukraine) are each concentrated in a few actors and places, so that if any one node is disrupted, global compute can stop. This is less an engineering problem to solve than a structural and political risk.

The map of concentration

Ranking the concentration the source asset confirmed under three-vote adversarial verification:

• TSMC / Taiwan — TSMC’s ≤7nm “advanced” is 74% of wafer revenue (2026 Q1: 3nm 25% + 5nm 36% + 7nm 13%). Most leading-edge fabs cluster in an approximately 100-mile corridor in western Taiwan, and Taiwan makes roughly 90% of the world’s most-advanced chips. If output halted, competitors’ 2/3nm slots are booked through 2027–28, so the situation is assessed as at least three years without a substitute.
• ASML / EUV — a monopoly inside a monopoly. ASML (Netherlands) is the sole maker of EUV scanners, and ASML itself depends on ZEISS SMT as its sole supplier of (High-NA) EUV optics (2016: 24.9% stake in Zeiss SMT, €1B).
• Three HBM makers — SK Hynix and Samsung (Korea) plus Micron (US) hold about 97%. All sit inside the US-alliance bloc, which makes them effective targets for export controls.
• Upstream materials — neon, critical for DUV excimer lasers, was roughly 50% Ukraine-sourced worldwide and about 90% of US semiconductor-grade imports (2022). Photoresist (a Japanese oligopoly of JSR, TOK, Shin-Etsu) is likewise single-country dependent.

[Single-point-of-failure hierarchy]
(1) TSMC/Taiwan leading edge  (≤7nm = 74% of revenue · Taiwan ~90% · 3-yr no substitute)   ← apex SPOF
        |
(2) ASML → ZEISS EUV  (sole scanner maker + sole optics supplier, no substitute)
        |
(3) Three HBM makers (~97%, US-alliance) · CoWoS packaging (TSMC-concentrated)
        |
(4) Upstream materials (neon ~50% Ukraine · photoresist Japanese oligopoly)

What not to misread

Diversification is real. But announced capacity and operating capacity are different things. Arizona and Kumamoto fabs typically run a node or more behind Taiwan on schedule, and numbers like “~30% 2nm+ after 2030” are full-buildout targets, not operating share. Conversely, the strong narrative that “overseas is only trailing nodes” was REFUTED (0–3) in the source asset — the balanced reading is that diversification is progressing but lagging.

Deep dive

1. Background — the “single point of failure” lens

Discussions of computing power usually headline chip performance (FLOPS, process node). Beneath the performance curve, however, lies the geographic distribution of the physical capacity to actually make those chips. When that distribution is concentrated in a few points, disruption at any one of them halts the whole supply no matter how far ahead performance is. Part 5 surveys four axes through this lens — advanced logic, lithography, memory, materials. The source asset summarizes it thus: computing power’s ultimate bottleneck is neither physics (Part 0) nor economics (Part 3) but concentration in a few geographic chokepoints.

2. What this synthesis newly established

The underlying deep-dive went through three-vote adversarial verification, closing at 106/106 with successful synthesis: 23 items confirmed → 8 synthesized, 2 killed, 0 unverified (attributed to the source asset). Key confirmed items:

• Apex single point (measured, CONFIRMED 3-0): TSMC ≤7nm = 74% of wafer revenue (2026 Q1; 3+5nm alone = 61%), Taiwan ~90% of the world’s most-advanced, replacement lead time 3+ years.
• Nested monopoly (CONFIRMED 3-0): ASML sole EUV scanner maker + ZEISS SMT sole optics supplier. High-NA throughput is capped by Zeiss optics capacity.
• HBM oligopoly (CONFIRMED 3-0): three firms ~97%, entirely US-alliance, HBM included in Dec-2024 controls.
• Upstream materials (CONFIRMED 3-0): neon ~50% Ukraine / ~90% of US imports (2022), photoresist a Japanese oligopoly.

3. Strengths and limits of the method

Strengths — measured values (TSMC 6-K revenue mix, ASML/ZEISS disclosures) and schedules (SiliconCanals, TrendForce) were cross-checked across multiple sources, and two exaggerated narratives were explicitly killed. Limits — (a) many diversification/reshoring figures are announced (projected) capacity and must be distinguished from operating capacity; (b) the “probability” of a geopolitical tail risk is inherently uncertain, and the source asset itself insists on exaggerating it in neither direction; (c) share and capacity figures are as of announcement and shift with node transitions.

[Skeptic mandatory caveat, inherited] Diversification and reshoring are mostly announced (projected) capacity, not operating capacity — conflating the two overstates any “easing of concentration.” Geopolitical tail risk likewise must not be exaggerated in either direction.

4. Neighbouring domains

The source asset deliberately does not force a link to CKM (cardio-renal-metabolic) — stated honestly. As a non-forced meta-connection, however, the single-point-of-failure / concentration-risk structure is isomorphic to pharmaceutical supply chains (concentration of active pharmaceutical ingredient, API, in a few countries). From a general-investor vantage, “tail risk of a concentrated supply chain” is a lens common to semiconductors and pharma. We record only the structural similarity and do not graft a CKM axis on by force.

5. Commercialization and market context (as structural risk)

• TSMC (TSM): apex of advanced logic. Overseas fabs (Arizona Fab21 4nm late-2024, Fab2 3nm 2H2027, Kumamoto 3nm 2028) typically trail Taiwan (N2 2025 Q4, A16 2026 H2) by a node or more.
• ASML (ASML) · ZEISS SMT: nested monopoly of EUV and optics. Concentration deepens rather than eases in the High-NA generation.
• Three HBM makers (005930.KS · 000660.KS · MU): ~97% oligopoly, an export-control target, supply tight.
• Chinese alternatives (SMIC 0981.HK · Huawei): SMIC 7nm (2022) and 5nm are attempted via DUV multi-patterning, but at ~40–50% higher cost and ~20% yield, and Huawei Ascend lags NVIDIA by several generations (attributed to the source asset). Absence of EUV is the core constraint.
• Export controls: Dec-2024 China-wide HBM/DRAM/advanced-packaging SME controls and an SMIC <10nm license line. Yet visible loopholes exist — the immersion-DUV gap (China pre-purchased ~70% of ASML DUVi revenue, ~$5–7B, in 2024) and CXMT’s exclusion from the Entity List. The source asset’s read: controls raised the cost curve but did not fully cut off supply.

(The company references above are factual statements of position on a structural-risk map, not a recommendation to buy or sell any security.)

6. The opposing view (skeptic block, quoted)

The core of the source asset’s §9 skeptic gate:

“Diversification and reshoring are mostly announced (projected) capacity, not operating (distinguish required). Do not exaggerate geopolitical tail risk in either direction.”

Two exaggerations were explicitly killed: (1) “China’s DUVi fleet will almost certainly print Huawei’s 2026 target of 1.6 million logic dies” → REFUTED (0–3); (2) the strong frame that “overseas is only trailing nodes” → REFUTED (0–3; diversification is progressing but lagging). Both the “self-sufficiency imminent” and the “reshoring complete” narratives warrant caution.

7. What to watch (falsifiable predictions from the source asset)

1. By 2028, if leading-edge (≤3nm) operating capacity outside Taiwan reaches a meaningful share (say >20%), “easing concentration” is confirmed; if announcements abound but operation stays in Taiwan, “persistent concentration” is confirmed.
2. If China reaches high-volume leading-edge (≤5nm) production without EUV, “weakening control efficacy” is confirmed.
3. Whether the actual reach of a (hypothetical) disruption at any single chokepoint on compute supply matches the prediction.

References

(Inherited directly from the source asset’s sources field. URLs/DOIs not provided by the source asset are not fabricated; unprovided items are labeled by source name and publication context.)

• TSMC. 2026. Q1 2026 6-K (≤7nm = 74% of wafer revenue mix).
• SiliconCanals; TrendForce. Arizona / Kumamoto fab schedules.
• ZEISS SMT; ASML. 6-K / disclosures (High-NA EUV optics monopoly, Zeiss SMT 24.9% stake).
• AEI (American Enterprise Institute). Immersion-DUV export-control loophole.
• ChinaTalk; CSET (Center for Security and Emerging Technology). HBM and export controls.
• SMIC / Huawei 2026 status (SMIC 5nm cost +40–50%, yield ~20%).
• Neon supply (2022 basis, Ukraine ~50%, US imports ~90%).

Disclosure

This post is for information only and is not investment advice. The author holds no position and no financial interest in the listed companies mentioned (TSMC, ASML, Samsung, SK Hynix, Micron, SMIC, Intel and others).

COI note. This post describes listed companies and US/China export-control policy in a structural and technical context, factually and neutrally. It carries no buy/sell implication and no advocacy or criticism of policy. Share and capacity figures distinguish “measured values vs. announced/projected,” and some quantitative claims — Chinese localization cost and yield among them — are attributed to secondary sources.