ASML Q2 2026: The Supply Chain Just Got Its 2027 Order Book

A supply-chain deep dive on ASML's Q2 2026 release Prepared by the Fable 5 research orchestrator | 2026-07-15, completed ~07:42 ET Evidence boundary: the official pre-Q&A release package (data cutoff 06:56 ET) plus same-day web evidence gathered 07:15–07:42 ET. The investor call (09:00 ET) and everything after it are OUT of scope; Section 10 lists what the call could change.

---

0. How to read this report

Analytical frame. The €326.5m beat above the guidance ceiling is a footnote. The event is a demand-plane reset delivered in three separable shocks, each of which propagates to different supply-chain nodes on a different clock:

1. A magnitude shock — FY2026 guidance raised from €36–40bn to €43–45bn (midpoint +€6bn, +15.8%), the second raise this year, implying ~+35% growth over FY2025, from a management team that has not missed the bottom of a guided range in nine consecutive quarters.
2. A time-structure shock — "pretty much already close to receive all the EUV orders we need for 2027" plus "a large number of orders" already in hand for 2028, converting 2027–2028 from forecast into order-covered backlog a year before ASML's own Capital Markets Day (June 10, 2027).
3. A mix shock — memory system revenue guided +75% in 2026 while H1 logic fell 10%; South Korea became ASML's largest region at 39% of H1 sales (+60.5% YoY); China is guided to re-accelerate from 15.9% of H1 sales back to ~20% of a much larger full year.

Method rules inherited from the audited workstreams in this directory (ZEISS-READTHROUGH-CONCLUSION.md and the frozen edge-network round): purchases, loans, advances, R&D funding, equity income and supplier revenue are separate flows, never summed; no named-customer allocation from ASML's geography pools; no mechanical conversion of capacity plans, orders, backlog or customer capex into recognized revenue; rejected legacy numbers stay rejected. Estimates are labeled with their assumption chains.

Source layers. [OFFICIAL] = ASML's Q2 2026 package (press release, US GAAP statements, IR presentation, video transcript, statutory interim report) and the 2025 Annual Report, all held locally under sources/. [CAL] = the frozen 9-quarter guidance-calibration corpus at ~/logs/theta/edge-network/asml-hub-round-20260714/. [EV-x] = the six same-day evidence files under fable-5-analysis/ (market reaction, leading-edge customers, China node, peers/WFE, upstream suppliers), each carrying per-claim URLs and FACT/REPORTED/ESTIMATE tags. All arithmetic is reproducible from fable-5-analysis/bridge_model.py → bridge_model_output.json.

---

1. Executive conclusions

Conviction grades: HIGH = multiple independent primary sources, arithmetic closes; MEDIUM = well-sourced but resting on reported (not filed) data or on timing assumptions; LOW = directional only. Each conclusion names its falsifier.

C1. The pre-registered "demand revision" test resolved positive: this is real demand, not timing or mix. (HIGH) The pre-event packet (frozen 2026-07-14) required a beat to be treated as timing/mix unless accompanied by a demand revision. Q2 delivered the beat (€9.3265bn vs €8.4–9.0bn guide) and the demand revision (guide +€6bn; capacity plans firmed; 2027 order coverage; 2028 orders in hand). The beat itself is mostly Installed Base Management (~€300m above management's own expectation, software-led productivity upgrades) — margin-rich but supplier-light. The news for the supply chain is everything around it. Falsifier: a Q&A or Q3 disclosure revealing the raise depends on pull-forwards ASML itself expects to reverse in 2027.

C2. The €43–45bn guide is floor-biased, but its interior is violent: the risk is quarterly allocation, not the annual number. (HIGH) In nine guided quarters (Q2'24–Q2'26), sales and gross margin never finished below the guided range [CAL]. The stated segment growth rates independently reproduce the guide (IBM +30% → €10.65bn; EUV +45% → €16.82bn; DUV+metrology/inspection +25% → €16.09bn; sum €43.56bn — inside the range with "more than 30%" IBM as the bridge to midpoint). But the implied H2 is a recognition wall: H2 €25.9bn at midpoint (+43% vs H1), Q4 implied €14.4bn (range €12.9–15.9bn) vs the €9.7bn all-time record quarter, requiring ~36 low-NA recognitions, ~90 ArF-immersion recognitions (2.25× the H1 rate), a ~2.2× half-on-half step in mature-DUV+metrology revenue, and 1–2 High-NA recognitions — with essentially zero slack against the stated ~65/~130 shipment plans at H1 ASPs. The balance sheet says the wave is already physically in motion (receivables €3.02bn → €7.25bn in H1, 78% of Q2 sales vs 31% at year-end). Demand is covered; FAT/SAT acceptance timing, install capacity and customer fab readiness are the binding risks to any given quarter. Falsifier: Q3 print below €11bn, or Q3/Q4 recognition slippage with FY guide maintained only via ASP effects.

C3. This is the first memory-led EUV cycle, and it changes who reads through to whom. (HIGH) FY2025 memory system sales were €8.42bn; +75% takes 2026 to ~€14.7bn, and H1 memory (€6.40bn) already matches logic (€6.44bn). The regime change: 1c/1γ DRAM carries 5–6 EUV layers (SK hynix leads; Samsung trimmed its count; Micron in EUV DRAM mass production), so ASML's memory revenue (+75%) is growing twice as fast as SEMI's memory WFE forecast (+37% blended) — litho is taking share of memory capex for the first time [EV-peers]. Customer corroboration is exceptional: Micron FQ3'26 revenue $41.46bn (+346% YoY, 84.9% non-GAAP GM — verified against the SEC 8-K by this orchestrator), SK hynix Q1 +198% at a 72% operating margin with a disclosed ~$7.97bn (KRW 11.95tn) ASML order book through 2027 (The Elec, Mar 2026) — the largest named EUV commitment this cycle, Samsung's KRW ~110tn 2026 investment program, and "HBM demand exceeds capacity for three years" (SK hynix, official). The analyst-estimated 2027 low-NA allocation is memory-led (SK hynix up to ~20 units — the largest single slot; REPORTED, not ASML-disclosed). Falsifier: DRAM contract prices rolling over hard in H2 2026; SK hynix/Samsung capex language softening at their late-July prints.

C4. Upstream, the 2027 capacity step is already contracted — but on trust, not on balance-sheet commitments. (HIGH on direction; the structure is the story) Reuters reported the day before the print (July 14) that ASML has "secured extra supplies of parts that take a long time to produce" — naming exactly the two hardest-to-scale categories: TRUMPF's high-power EUV drive lasers and Zeiss's optics, with TRUMPF on the record: "We are fully prepared to meet ASML's EUV demand over the next three years." Meanwhile supplier order books already inflected a quarter ago: VAT Group Q1 orders +47% YoY (semi +65%, book-to-bill 1.6×) while sales fell 20% in the same release — the order-to-revenue lag in one document; Jenoptik semiconductor-segment orders >2.5× YoY "driven mainly by lithography"; NTS: "orders from ASML… increasing substantially"; Aalberts' CEO: the Dronten plant is built and "ready for the growth," ramping 2027, tied to "what our key customer sees from their EUV" [EV-suppliers]. Yet ASML's disclosed purchase obligations have grown only ~7.6% since FY2022 (€11.8bn → €12.7bn), the interim report says commitments remain "in line with" year-end, and ASML's contracts allow delay/cancellation aligned to its own sales — suppliers are building against relationship visibility, not hard guarantees. That asymmetry is exactly how ASML preserved optionality in 2022 and cut when the cycle turned. Falsifier: FY2026 purchase obligations failing to step up materially at the next annual disclosure would mean the 2028 ambition is still an option, not a plan.

C5. The Zeiss read-through strengthens on the forward leg and stays limited on the current-year leg — the frozen conclusion stands. (HIGH) Nothing in Q2 justifies a mechanical current-year ZEISS revenue upgrade (the beat is IBM-led, low optical content; the audited prohibition list stays in force). But the forward signal got two independent upgrades today: ASML securing extended optics supply (Reuters, above), and — the sharpest single quote in this research — Gudeng Precision's chairman stating that the binding constraint on new leading-edge fab capacity is "an EUV scanner bottleneck, given limited supply of optics systems from the world's sole supplier, Zeiss" (June 10) [EV-suppliers]. Descriptive history only: ASML purchases from Zeiss ran 15.2% / 18.1% / 18.0% of ASML system sales in 2023/2024/2025 (€3.33bn / €3.95bn / €4.41bn); 2026 system revenue ~34% above 2025 with EUV-heavy mix plus a +30% 2027 capacity step implies materially higher optics procurement in direction — magnitude intentionally not pointed, per the calculation audit. Falsifier: none needed for the qualitative claim; a ZEISS SMT capacity disclosure would convert it to quantitative.

C6. Adjacent equipment: the print date-stamps peer 2027 revenue; the cleanest mechanical winner is the EUV track, and the memory dollars split into two distinct baskets. (HIGH) Scanners are ordered 12–18 months out; deposition/etch/ALD/process-control ~4–6 months. "2027 EUV nearly fully ordered" therefore means the same fabs must place their non-litho orders late-2026 into 2027 — AMAT/LRCX/TEL/KLA/ASMI 2027 revenue is substantially pre-visible today [EV-peers]. Tokyo Electron is the most mechanically linked: ~1:1 coater/developer attach per EUV scanner at ~91% share (near-100% for EUV) — TEL guided its own coater/developer business +>50% for FY3/2027 before today's print, and its explicit driver ("DRAM customers adopt EUV, logic customers introduce EUV multi-patterning") is ASML's thesis stated from the customer side. The HBM chain (Disco grind/dice, Hanmi/Hanwha/Besi/EVG bonding, Advantest/Teradyne test) captures the non-litho memory dollars: Hanmi's Q2 (provisional figures reported this morning; audited results due ~Aug 14) printed a record 51.9% operating margin and touched its +30% daily price limit intraday; Besi hybrid-bonding orders +104% YoY. Note the discipline: HBM stacking itself is litho-light — ASML's +75% memory number is the DRAM-wafer front end, not the HBM back end. Falsifier: TEL/Screen/ASMI July–August prints failing to show the order step their own guidance implies.

C7. The historical "ASML leads WFE" heuristic is wrong as commonly stated — and knowing the real pattern matters for 2027 risk management. (MEDIUM-HIGH) In 2016–2018 and 2020–2022, dep/etch orders led ASML in memory-driven up-legs (AMAT's FQ2'16 orders +52% QoQ while ASML's 2016 book-to-bill was 0.79), and ASML's true signature was lagging into downturns on its sticky logic/EUV backlog (FY2019: peers −13%, ASML +8%; YE2022: ASML book-to-bill ~1.45 while Lam guided 2023 down >20%) [EV-peers lead-lag study]. This cycle inverts the up-leg logic — DRAM now consumes EUV, so ASML participates in the memory leg from the start — but the down-leg asymmetry is intact and is the operative warning: if memory pricing cracks in 2027, ASML's own P&L will be the last place it shows. Watch DRAM contract prices and dep/etch order books for the turn, not ASML's prints. The problem is compounded by disclosure: ASML stopped publishing quarterly bookings after Q1 2026 (lumpiness rationale), so the market's earliest formal signal has been retired at precisely the moment order flow became the whole story. Falsifier: n/a (historical claim); its application fails if HBM long-term agreements prove as cancelable as commodity-DRAM orders were in 2018.

C8. China: the H2 re-ramp is credible but anticipatory and policy-gated — and it did not drive the raise. (MEDIUM) The arithmetic is aggressive: ~20% of €44bn = €8.8bn FY China; minus €2.88bn H1 leaves ~€5.9bn H2, a ~€2.96bn/quarter run-rate — above the FY2025 average (€2.38bn/q) and >2× the H1 rate — and the CFO confirmed the ~20% is "of a higher base," i.e., absolute China demand was revised UP ~€1bn with the raise. The evidence-ranked drivers [EV-china]: (1) genuine domestic-led logic + DRAM demand at fabs ASML can legally serve — crucially CXMT, Hua Hong, Nexchip and CanSemi are not Entity-Listed (SMIC and the Huawei network are); CXMT is the pivotal node (~350k wspm DRAM by end-2026, all-DUV multipatterning that raises immersion intensity per wafer, Shanghai mega-fab tool-in H2'26, ~$8.55bn STAR IPO listing July 27); (2) front-running the MATCH Act (House committee-passed April 22; would ban DUV-immersion sales and service to SMIC/Hua Hong/CXMT/YMTC/Huawei) — a documented ASML-China pattern (~90 immersion tools, ~$5–7bn, stockpiled in 2024 per AEI estimates); (3) H2 tool-in schedules. Export-control precision matters: since Sept 2024 all ArF immersion (including NXT:1970i/1980i) needs a Dutch license; KrF/dry/i-line remain license-free — which is where ASML's dry-DUV unit surge earns its China margin. Front-running supports the H2 number and simultaneously makes part of it non-recurring into 2027. Falsifier: a MATCH Act floor vote or Dutch license tightening before Q4 recognition; conversely, SMIC/Hua Hong August capex commentary confirming domestic logic strength would upgrade this to HIGH.

C9. The KrF/dry surge is a global memory signal wearing a China coat. (MEDIUM-HIGH) H1 KrF units 65 vs 38 YoY; Q2 alone shipped 35 KrF + 8 dry ArF + 9 i-line — but Q2 ship-to was Korea 43% / China 14%, and 3D NAND adds layers without adding litho passes, so KrF demand tracks wafer-start additions (Korea DRAM + the Kioxia/SanDisk +41% NAND capex cycle), not layer count and not China alone [EV-china]. Canon independently raised its FY2026 KrF unit guide from 46 to ~69–71 (first KrF platform refresh in 14 years, aimed at the NAND shortage), while ASML says it is "capturing some customers with our dry products" — the mature-DUV pond is growing fast enough to feed both, but note KrF is only ~6% of ASML system value; this is a units story, not a revenue driver. Falsifier: Kioxia's July 31 print abandoning capacity discipline (would signal froth), or NAND pricing breaking down (would kill the wave).

C10. High-NA graduated from roadmap to reference, and its economics run through Intel now, everyone later. (HIGH on the milestone, LOW on 2026 revenue relevance) The joint ASML/Intel release (today) confirms the transcript verbatim: first HVM logic product (Panther Lake) with select 18A layers patterned on High-NA at Hillsboro — dual-qualified against low-NA, yields matched, a readiness milestone rather than a dependency. 2026 High-NA is ~4–5 EXE recognitions (~€1.2bn H1 + ~1–2 units implied in H2 by the +45% EUV bridge). TSMC's deliberate High-NA deferral ("From 2nm to A14, we do not need High-NA" — SVP Kevin Zhang) is bullish for low-NA units via multipatterning. Full High-NA volume is an Intel-14A / 2027–28 event, with Samsung optionality. Falsifier: none near-term; watch 14A external-customer wins as the volume unlock.

C11. Margin structure has shifted a gear, and the driver ranking matters for durability. (HIGH) Q2 GM 54.0% (guide 51–52%); Q3 guided 55–57% — an all-time record if hit; FY 54–56% vs 51–53% prior. Drivers in management's own order: upgrade-rich IBM (software-led, near-zero marginal COGS), fixed-cost absorption on record volume, EUV+immersion mix. The upgrade layer (~€300m of the Q2 beat) is the highest-quality, most repeatable component — it monetizes the installed base without consuming scanner build slots or supplier BOM, and its consumables echo is already visible at Gudeng (record Q2, +35% YoY, reported July 13–14). Falsifier: Q3 GM printing below 55% despite the IBM mix would indicate High-NA dilution or ramp costs biting harder than guided.

C12. The market read it narrowly this morning; the supply-chain repricing is unfinished. (MEDIUM — a market observation, not investment advice) Quantified surprise: Q2 revenue beat consensus ~6%, but the Q3 guide beat consensus ~11% (€11.5bn mid vs ~€10.36bn) and the FY guide ~11.7% (€44bn vs €39.4bn) — a forward-visibility event, proportionally larger than the trailing beat [EV-market]. Tape: ASML.AS traded +3.4% to +7.9% intraday (high €1,678.20); Tokyo delivered the cleanest read-through into its 15:30 close after the 14:00 JST release (Lasertec +9.8% at the close after topping +10% intraday, Advantest +5.8%, TEL +4.4%, Screen +4.2%, Kokusai +4.8%); US pre-market moved the toolmakers (AMAT +2.9%, LRCX +2.7%, KLAC +2.8%) while NVDA/MU sat flat — the market classified this as equipment-capex confirmation, not new AI-demand news. European upstream names barely moved (ASMI +1.6%, VAT +0.7%, Besi ~flat): the long-lead supplier ring, where the 2027 capacity step lands first and hardest, was NOT repriced today. Every identifiable sell-side price target predates the print; the Korea rally (Samsung +6.3%, SK hynix +8.8–11%) is causally contested (July 14 US CPI vs ASML). EUR/USD ~1.142 was a non-factor. Falsifier/maturation: post-call PT wave and the July 22 VAT / late-July supplier prints will test the upstream-lag hypothesis.

---

2. The print versus the frozen forecast (what was predictable, what was not)

The pre-event forecast packet (frozen before the release) called €8.7bn base / €8.4–9.0bn range, systems €6.2bn, IBM ~€2.5bn. Scoring [CAL FORECAST-SCORECARD]:

Decomposition: of the €326.5m above-range surprise, €261.7m is IBM (management: ~€300m above its own expectation, productivity upgrades) and only €64.8m is systems above the implied high. The forecastable core (systems) behaved; the residual was the upgrade wave — which management itself did not anticipate at this magnitude. Post-event test from the packet: "a revenue beat without a demand revision should be attributed first to timing and mix." The demand revision arrived in the same release; the test resolves as genuine demand (C1).

Also scored: Q2 units 91 (86 new / 5 used); Q2 system-revenue mix EUV 57% / ArFi 29% / KrF 6%; end-use logic 51% / memory 49%; ship-to Korea 43% / Taiwan 30% / China 14% / US 9% / Japan 4% [OFFICIAL, IR chart].

3. Guidance credibility: what €43–45bn is worth

Nine consecutive guided quarters without a below-range print, GM above the range high in five of nine [CAL]. The FY guide path — €34–39bn (Jan) → €36–40bn (Apr) → €43–45bn (Jul) — is itself evidence: ASML raises in steps as order coverage converts, rather than front-loading ambition. Two structural caveats keep this from being a blank check: (a) the never-miss record is quarterly; FY guides get revised in both directions across a year (FY2025's initial €30–35bn resolved at €32.7bn, mid-range); (b) the 2026 interior requires maximal H2 recognition (C2) — the annual number can survive a Q3 wobble only via Q4, which has no slack left. The asymmetric read: downside to €43bn requires an execution failure; upside through €45bn requires ASPs (3800E/EXE mix) doing work the unit plan cannot.

4. Anatomy of the beat: the €300m that does not touch the supply chain (much)

The IBM upgrade wave is software-led, requires little machine downtime, and delivers instant productivity (management, transcript). Economically: near-zero marginal BOM, no scanner slots consumed, gross-margin accretive (part of the 54.0%), and it raises the productivity of installed EUV fleets — a subtle demand-side effect that lets customers defer some unit purchases at the margin while paying ASML more. Supplier content is limited to the service/consumables chain (pods, FOUPs, spares — see Gudeng's record quarter), not the optics/laser/module BOM. This is why the ZEISS current-year read stays limited (C5) even as the quarter beat by €326.5m: the beat happened in the one revenue line with the lowest upstream propagation.

5. The H2 2026 recognition wall (the guide's soft underbelly)

All arithmetic official-inputs-only; assumptions flagged (bridge_model.py, section E):

Assumptions: recognition ≈ shipment parity; H1 ASPs flat (NXE €231m, EXE €396m, ArFi €83m — official arithmetic); "approximately" unit guides treated as exact. Slack analysis: at those ASPs the plan sums slightly below the €44bn midpoint — the guide implicitly assumes some combination of ASP uplift (NXE:3800E-heavy mix, EXE:5200B), unit counts above "approximately," and IBM above +30%. There is no idle-capacity buffer. The corroborating balance-sheet physics: receivables +€4.23bn in H1 (invoicing loaded into late Q2 — tools are moving), while contract liabilities FELL €1.94bn (€19.37bn → €17.43bn IFRS): recognition consumed prepayments faster than new down payments arrived. That last line is the one genuinely soft data point against the "very strong H1 bookings" language — it likely reflects the 2023–25 China prepay unwind plus down-payment staging, but it belongs on the monitoring list, because a second consecutive decline alongside strong order language would imply either easier payment terms (mix shift toward the largest customers) or back-loaded order intake.

6. Demand decomposition (where the €44bn comes from)

Memory (engine). €8.42bn (FY25) → ~€14.7bn (+75%); H1 €6.40bn → H2 ~€8.3bn. Pricing context is historic: Micron's FQ3 at $41.46bn/+346% (SEC-verified), NAND contract prices +55–60% in Q1, DRAM WFE forecast +39% (SEMI). ASML's number is the front-end DRAM wafer leg — 1c/1γ EUV layers plus advanced-immersion intensity — not the HBM back end (C6). Korea at 39.2% of H1 sales (€7.09bn, already 87% of FY2025's full-year Korea) is the geographic shadow of SK hynix M15X + Samsung P4/1c conversions; the split between them is not disclosable and we do not allocate it.

Advanced logic (the H2 swing). Total logic implied ~€18.2bn (+13%) against advanced foundry logic +25% [OFFICIAL] means mature logic contracts −4% to −34% depending on the advanced share assumed (60/70/80% scenarios; bridge C). H1 logic was −10%; the FY plan therefore requires H2 logic ≈ €11.7bn, +82% versus H1 — the single most demanding line in the whole guide. It rides on: capacity additions at existing N5/N4/N3 nodes for AI ("huge demand on those technologies," transcript — the underappreciated quote of the release), the N2 ramp "as aggressive as possible" (TSMC N2 volume since Q4'25, sold out 2026, capex $52–56bn "at the high end," >30% revenue growth guided), 18A (Intel), SF2 (Samsung), and first A14 preparations. TSMC reports tomorrow (July 16) — the first external test.

China (the re-ramp). Covered in C8. One addition: the KrF/dry surge that looks China-shaped is mostly memory-shaped (C9); China's actual H2 lane is dry-DUV (license-free) + licensed mature immersion into CXMT/Hua Hong/Nexchip tool-ins, plus MATCH-Act insurance buying.

IBM (the compounder). €8.19bn → ~€10.7bn+ (">30%"). Structurally: installed base grows every quarter; EUV service intensity is higher; upgrades are software-margin. This is the line that makes ASML's revenue base progressively less cyclical — and it is the line that beat.

7. Node-by-node supply-chain read-through

7.1 Upstream ring (the 2027 capacity step lands here first)

Mechanism summary. The waterfall runs: ASML order coverage (2027 sold, 2028 filling) → ASML pre-positions longest-lead inputs (optics, lasers — done, per Reuters) → Tier-1/2 order books inflect (done, Q1'26 data) → supplier P&L recognition (starts H2'26–2027) → supplier capacity investment completes (H2'27, Jenoptik Jena; 2027 Dronten ramp). ASML's own financial commitment stays contractually soft (purchase obligations +7.6% over three years; delay/cancel rights) — suppliers carry the balance-sheet risk of ASML's optionality, as they did in 2022. If the 2028 "investigating" becomes "committed" at or before the June 2027 CMD, the purchase-obligation line is where it will first be legible.

7.2 Adjacent WFE (who inherits the print)

WFE context first: SEMI's mid-year forecast — published July 14, one day pre-print — already lifted 2026 WFE to $143.9bn (+23.1%) (DRAM +39%, NAND +31%, foundry/logic +19%, test +31%), with 2027 ≈$175bn and 2028 ≈$200bn; the sell-side cluster (Cantor $145bn/2026 → $185bn/2027; Citi $145/200/250bn; Lam "~$140bn with upside bias"; KLA ">$140bn"; SCREEN's planning case $134–140bn) had converged the same way. ASML's implied 2026 system revenue (~€30.8–32.9bn ≈ $35–38bn at 1.14) is ~24% of WFE — inside the historical 22–28% litho band. The correct headline is therefore validation, not front-running: ASML confirms the already-raised WFE consensus and skews 2027–2028 revisions upward via its order coverage. The exception — the structural finding — is within memory, where ASML's +75% against memory-WFE +37% marks litho taking share of memory capex for the first time (EUV-in-DRAM).

7.3 Downstream customers (who signed the 2027 order book)

Ranked by evidentiary weight in explaining the print [EV-cust]: (1) the memory complex — SK hynix (disclosed ~$8bn ASML order book through 2027; M15X ramping; 1c at 5–6 EUV layers; "HBM demand exceeds capacity for 3 years"), Samsung (KRW ~110tn program; HBM capacity +50%; 1c conversion; P4/P5; foundry a 2027+ contributor, not the Korea driver), Micron ($27bn capex, 1γ EUV, HBM4 shipping >$1bn, US-flat-geography reconciled by its Japan/Taiwan/Singapore EUV footprint); (2) TSMC — the logic anchor and largest single low-NA buyer, whose High-NA deferral is low-NA-bullish; (3) Intel — the High-NA proof point (joint release today), not a volume driver (capex ~flat $14.6bn, tools +25% mix); (4) Kioxia/NAND — the DUV leg, disciplined and repurposing-led; (5) Rapidus — symbolic 2028 breadth (one NXE:3800E, 2nm pilot, MP 2H FY2027); (6) China domestic — CXMT above all (C8).

The named-customer discipline holds: ASML discloses no customer revenue; 2025 concentration was extreme (top customer 23.9%, top four 61.2%) [OFFICIAL AR25]; the 2027 unit-allocation estimates circulating (SK hynix ~20 / Samsung ~7 / Intel ~5 / TSMC+Micron balance) are analyst order-book reconstructions [REPORTED], used here only to characterize the shape (memory-led) of the 2027 book.

7.4 Second-order nodes (briefly, for completeness)

More EUV layers × more wafer starts propagate to: EUV photoresist and underlayers (JSR/TOK/Shin-Etsu — Japan chemistry complex), EUV masks/blanks (Hoya, AGC; written by NuFlare/JEOL e-beam, inspected by Lasertec), pellicles (ASML/Mitsui), reticle pods (Gudeng — already printing), and silicon (Shin-Etsu/SUMCO wafer adds lag fab tool-ins by ~2–4 quarters). None was independently verified today beyond Gudeng; flagged as the natural next research ring.

8. Cycle context: the 2022 analog and what is different

The last time ASML stacked capacity ambitions this steep (2022: toward 90 low-NA/year), the 2023–24 downturn arrived first and the plans were quietly rescoped — capacity plans are options ASML buys mostly with suppliers' balance sheets (C4). What is genuinely different now: (a) order coverage precedes the capacity step (2027 "close to all orders" before the 2027 capacity is even built — in 2022 the coverage was assumed, not signed); (b) the demand mix includes memory-with-EUV and sovereign/AI capex, both stickier than 2021's shortage hoarding; (c) the installed-base annuity is ~€10.7bn/year and compounding, cushioning any system air-pocket. What rhymes: extreme customer concentration, a China slice partly built on regulatory front-running, and memory-price dependence at the margin — with C7's warning that ASML's own P&L is a lagging indicator at every cycle top. Prudent stance: treat 2027 as underwritten, treat 2028 as an option whose strike is the June 10, 2027 CMD, and hedge the 2027 memory-digestion scenario by watching DRAM pricing and dep/etch orders, not Veldhoven.

9. Timing map (dated catalysts that test each conclusion)

10. Risk ledger (ranked by expected supply-chain impact)

1. Quarterly recognition slip inside an intact year (C2) — highest probability, lowest severity; hits sentiment and Q3/Q4 optics, not demand.
2. Memory price break in 2027 (C3/C7) — moderate probability, high severity; ASML reports it last; watch DRAM contract pricing, hyperscaler memory purchase agreements, dep/etch orders.
3. China policy event (C8) — MATCH Act (sales and service ban would also hit the IBM annuity in China), Dutch tightening, entity-list additions (CXMT is the name to watch); severity concentrated in the DUV/mature leg and 2027 China baseline.
4. Supplier capacity friction (C4/C5) — optics/laser scaling, Dutch grid constraint, skilled-labor; would convert the 2027 revenue potential (~€52–58bn capacity-limited scenario, NOT a forecast) into a lower deliverable ceiling.
5. Concentration — four customers = 61.2% of 2025 revenue; one architecture (AI accelerators + HBM) increasingly underwrites all of them; a single hyperscaler capex pause propagates through logic AND memory simultaneously.
6. Disclosure opacity — no numeric bookings/backlog going forward; the market must now infer order flow from supplier prints and two balance-sheet lines (contract liabilities, receivables). Section 5's contract-liability decline is the current open question.
7. FX — EUR/USD 1.142 today, a non-factor in the print; a sustained strong euro compresses the USD-WFE-share optics but not EUR guidance.

11. Quantitative appendix (auditable)

Machine-checkable inputs/outputs: fable-5-analysis/bridge_model.py, bridge_model_output.json. Official FY2025 baselines [AR25]: total €32,667.3m; systems €24,474.3m; IBM €8,193.0m; EUV €11,602.7m (NXE 44u/€10,445.8m + EXE 4u/€1,156.9m); ArFi 131u/€10,311.4m; logic €16,054.1m; memory €8,420.2m; geography Korea €8,159.6m / Taiwan €8,337.9m / China €9,519.7m (29.1%) / US €4,089.1m; RPO €46.5bn (65% ≤12m); system backlog €38.8bn; contract liabilities €19,372.6m; purchase obligations €12.7bn (€9.6bn ≤1y); Zeiss purchases €4,407m. H1 2026 [Q2PKG/INT26]: total €18,093.4m; systems €12,844.2m; IBM €5,249.2m; EUV €7,897.4m (29 NXE + 3 EXE); memory €6,401.5m; logic €6,442.7m; Korea €7,085.5m; China €2,883.3m; AR €7,253m; contract liabilities €17,428.6m (columns verified [Dec-25 | Jun-26] against the US-GAAP AR note).

Key derived results (all reproduced in Sections 5–6): FY guide mid €44.0bn = +34.7% YoY; segment-sum check €43.56bn; Q4 implied €14.41bn (12.9–15.9); H2 logic +82% HoH; memory 2026e €14.74bn; China H2 €5.92bn (~€2.96bn/q); capacity paths 65→~85→~104–110 (low-NA) and 130→~169→~208–220 (immersion; 2028 = "investigating", compounding vs base-year reading both shown); 2027 capacity-limited revenue potential €52–58bn under explicit ASP/unit assumptions (scenario ceiling, not a forecast); Zeiss purchase-intensity history 15.2%/18.1%/18.0% of system sales (2023/24/25) with no 2026 point forecast (audit boundary).

Consensus deltas [EV-market]: Q2 revenue +6.0% vs LSEG €8.80bn; net income +11.5% vs €2.62bn; Q3 guide ~+11% vs ~€10.36bn; FY guide +11.7% vs €39.40bn (consensus source not pinned to a single named outlet — flagged); Q3 GM guide +~490bps vs 52.1%.

12. Source ledger

Official (local copies under sources/): ASML Q2 2026 press release, US GAAP statements, IR presentation, video transcript (quotes in §1/§6 verbatim), 2026 Statutory Interim Report (capacity-plan wording; contract liabilities; commitments note); ASML 2025 Annual Report US GAAP (all FY2025 tables); ASML/Intel joint release 2026-07-15 (High-NA milestone). ASML Q2 2026 results page: https://www.asml.com/en/investors/financial-results/q2-2026 Frozen calibration corpus: ~/logs/theta/edge-network/asml-hub-round-20260714/ (MANAGEMENT-CALIBRATION.md, FORECAST-SCORECARD.md, HISTORICAL-MANAGEMENT-CALL-CORPUS.md). Prior audited workstream (imported, not recomputed): ZEISS-READTHROUGH-CONCLUSION.md + checkpoints/ + calculations/ in this directory. Same-day evidence files (per-claim URLs and FACT/REPORTED/ESTIMATE tags inside): fable-5-analysis/evidence-market-reaction.md, evidence-customers-leading-edge.md, evidence-china-node.md, evidence-peers-wfe.md, evidence-suppliers-upstream.md. Selected load-bearing external items (dates as published): Reuters "ASML expected to shine light on capacity and China challenges" (TRUMPF/Zeiss supply security; 2026-07-14); The Elec (SK hynix KRW 11.95tn ASML order book; 2026-03); Micron FQ3'26 8-K (sec.gov, 2026-06-24 — independently re-verified by the orchestrator); SEMI Mid-Year Equipment Forecast (2026-07-14); TSMC Q1'26 call (capex $52–56bn "high end"; 2026-04-16); TrendForce/Tom's Hardware (TSMC High-NA deferral; 2026-05/06); Samsung Q2 prelim (2026-07-07); Nikkei Tokyo close reports (2026-07-15); VAT Q1 trading update (2026-04-16); Jenoptik Q1 (2026-05-12); Gudeng Q2 record (2026-07-13/14); Federal Register / Bits&Chips (export-control regime); Senate Foreign Relations / press (MATCH Act, 2026-04); Bloomberg/Nikkei (CXMT IPO pricing, 2026-07-14).

Known limitations. Pre-Q&A boundary (the call may add bookings color, China specifics, and margin mechanics that modify §5/§6 emphasis); no verified ASML.AS close existed at time of writing; three memory-customer print magnitudes (Samsung KRW 171tn revenue scope; SK hynix net-vs-operating line; KOSPI +8%/index-level tape) rest on same-day secondary sources and are used directionally only; the FY consensus figure (€39.40bn) recurred across independent syntheses but was not pinned to one named outlet; 2027 low-NA customer allocation is an analyst reconstruction, not ASML disclosure.

---

End of report. No commit, push, merge, or external publication performed. All working artifacts live under fable-5-analysis/ in this directory.