Semicap service is an installed-base annuity, and the market still prices these names as capex cyclicals
semicap-service-annuity · conviction low · status open · horizon 2027-2029 (the test requires a downturn) · as of 2026-08-03
Combined service revenue at ASML, Lam, TEL and KLA grew 33% year-on-year in the June quarter against 20% for equipment, service outgrew equipment at most of them, and long-term service targets were revised UP at the two most recent updates. Service tracks the installed base and how hard it is being run, so it decays far more slowly than orders — and it funds the dividends. The archetype the desk applies to these names, and the multiple the market applies, both treat them as pure capital-equipment cyclicals.
Robust to undisclosed shares. 1 derived input under this thesis; redrawing every supply weight the industry does not publish moves none of them by more than 25%. Computed from evidence at most 17 days old (oldest input: analog-devices).
Exhibits
Exhibit 1Relative performance, indexed to 100How the names in this thesis have traded against SOXX.
Series available as data/semicap-service-annuity.csv
Exhibit 2What the conviction is actually made ofEach premise and the number it composes to. A conjunction of plausible premises is far weaker than any of them.
Service revenue is structurally decoupled from the order cycle because it tracks the installed base
† 1 premise marked supporting — shown and arguable, but the conclusion does not depend on them, so they are not multiplied into the composed figure. Citing a filed figure should not cost conviction.
86% if the 2 gates are independent, 90% if they move together. They are claims about one industry, so the truth is between and nobody can say where. Treat this as an ordering device rather than a calibrated probability — the ranking of premises is the information, not the level.
Weakest link: Global Semi Cycle at 0.90 — The cycle node the consensus prices these names against.
Process complexity keeps the installed base working harder, which raises service intensity independently of unit growth
69% if the 3 gates are independent, 85% if they move together. They are claims about one industry, so the truth is between and nobody can say where. Treat this as an ordering device rather than a calibrated probability — the ranking of premises is the information, not the level.
Weakest link: Wafer cleaning (wet processing) at 0.85 — Highest-frequency step in the flow; scales with TOTAL step count, the cleanest proxy for process complexity.
Therefore the archetype is wrong, and the mispricing is in the trough multiple rather than the peak
69% if the 3 gates are independent, 85% if they move together. They are claims about one industry, so the truth is between and nobody can say where. Treat this as an ordering device rather than a calibrated probability — the ranking of premises is the information, not the level.
Weakest link: Tokyo Electron Limited at 0.85 — Service outgrew equipment; the pattern is group-wide rather than name-specific.
The variant
Consensus
Semicap names are high-quality capital-equipment cyclicals. Earnings track WFE spend, WFE spend tracks customer capex, and capex is the second derivative of end demand — so the group is more cyclical than its customers and is valued on where the cycle sits. The service line is a stabiliser, noted and then largely ignored in the multiple.
Variant
The mix has shifted far enough to change the security. Service revenue is a function of CUMULATIVE historical shipments and of utilisation, not of current orders — an installed base that keeps growing makes the annuity grow through a downturn in orders. June-quarter service grew 33% against 20% for equipment; long-term targets now run from Lam 10%+ and ASM 12%+ to KLA 13-15% and AMAT mid-teens, with the two most recent revisions UPWARD. If a rising share of earnings is annuity that funds the dividend, the correct comparison set drifts from capital equipment toward installed-base compounders, and the trough multiple the market underwrites is too low.
Differentiator
This is a mix-shift and archetype argument, not a cycle call — it is deliberately agnostic on where WFE spend goes next, and it gets MORE interesting if the cycle turns down. The desk's own archetype layer classifies all five names as capital-intensive equipment cyclicals, which is the same error in miniature: a classification made when the mix was different and never revisited. Service is also a real-time UTILISATION signal that inflects ahead of equipment orders, so it has diagnostic value for the rest of the ontology beyond the names themselves.
Falsifiers
claim: Service proves cyclical after all · criterion: In any WFE downturn through 2029 where equipment revenue falls >25% year-on-year, combined service revenue across ASML, Lam, TEL and KLA falls more than 10% · horizon: 2029-12-31 · settles: refuted
claim: The mix shift reverses · criterion: Service grows slower than equipment for four consecutive quarters at three or more of the four names · horizon: 2028-06-30 · settles: refuted
claim: The market re-rates on the annuity · criterion: Trough EV/EBIT for the group in the next downturn exceeds the prior downturn's trough by more than the semis benchmark's own multiple expansion · horizon: 2029-12-31 · settles: confirmed
claim: SCOPE — right about the business, wrong about the security · criterion: The annuity share rises as described and the group still de-rates to prior-cycle trough multiples. The mix shift was real and not priced differently · horizon: 2029-12-31 · settles: refuted
Open questions
CLOSED 2026-08-03. service_revenue_share is now a sourced field on four of five names: lam-research 36.8% (CSBG, Jun-qtr, +43% YoY), asml 30.1% (Installed Base Management, Q2-26), kla 22.4% (Services, Jun-qtr, +17% YoY), applied-materials 22.5% (AGS, FY25). tokyo-electron is NULL — service outgrew equipment but no share figure was obtained, and it is recorded as unknown rather than estimated. CAVEAT THAT LIMITS THE THESIS: segment definitions differ materially between these companies (ASML's IBM includes upgrades; Lam's CSBG includes spares, service and Reliant refurbished systems; KLA's Services is narrower). Use the ORDERING and the direction, not the decimals. The spread 22-37% is wide enough that the thesis is much stronger on Lam than on KLA or AMAT.
The +33% group service growth that prompted this thesis is CARRIED BY LAM (+43%). KLA is +17% and management described services as near the low end of its 13-15% long-term range. A group claim resting on one name is weaker than it looks — check whether Lam's CSBG mix (which includes Reliant systems, i.e. TOOLS) is inflating the comparison.
One quarter of 33% vs 20% is not a trend, and service comparisons are sensitive to installed-base mix and upgrade timing. The thesis needs several quarters before conviction can move off low.
Service growth partly reflects tools running hot, so it is coincident-to-leading on utilisation rather than an independent demand signal. Its diagnostic value for the rest of the ontology may exceed its value as a stock call.
The desk's archetype layer classifies all five as capital-intensive equipment cyclicals. If this thesis is right, that classification is wrong and every archetype-driven comparison involving them inherits the error.
Reasoning chain
Service revenue is structurally decoupled from the order cycle because it tracks the installed baseVALID
premises
ASML Holding0.95 strong
Largest installed base in the group; service and field options a long-standing material line.
Lam Research Corporation — service revenue share $370.90 strong
2.47 bn6.72 bn43%
RESEARCHED 2026-08-03 and now in the model.
why
CSBG was $2.47bn of $6.72bn in the June quarter, +43% YoY. This is the strongest single datapoint in the thesis: over a third of revenue at the largest etch vendor is installed-base annuity, not tool sales.
Global Semi Cycle0.90 strong
The cycle node the consensus prices these names against.
Orders are a flow; the installed base is a stock. Service bills against the stock and against how hard it is run, so its revenue does not fall with orders — which is precisely what makes the blended earnings stream less cyclical than the order book that the multiple is set on.
Process complexity keeps the installed base working harder, which raises service intensity independently of unit growthVALID
premises
Etch (plasma etch)0.90 strong
Step count scales with vertical structure, not wafer starts — 3D NAND layers, GAA logic.
Thin-film deposition (CVD / PVD / ALD)0.90 strong
Largest WFE category and the one most levered to structure height.
Wafer cleaning (wet processing)0.85 strong
Highest-frequency step in the flow; scales with TOTAL step count, the cleanest proxy for process complexity.
Every added process step is an added tool running, an added yield problem and an added service hour. Complexity growth therefore raises the service annuity even when unit shipments are flat.
Therefore the archetype is wrong, and the mispricing is in the trough multiple rather than the peakVALID
premises
Lam Research Corporation0.90 strong
10%
REACTIVE from derived:lam-research.demand_pull (today 147.1, band 58.85-235.4 -> p 0.98-0.75).
why
Lam's demand pull is the EQUIPMENT cycle. The thesis is about the annuity surviving a downturn, so it gets MORE interesting as pull falls — the inverse wiring is the point, and it is why this thesis is the hedge against the rest of the bank. Anchors are centred on today's reading so this makes the premise LIVE without re-rating the thesis. Long-term service target 10%+, explicitly guided.
Applied Materials, Inc.0.90 strong
Long-term service target mid-teens, revised upward at the most recent update.
Tokyo Electron Limited0.85 strong
Service outgrew equipment; the pattern is group-wide rather than name-specific.
A group whose annuity share is rising and whose dividends are funded from it should not be underwritten at a pure-cyclical trough multiple. The claim is about the DOWNSIDE case, which is why the test requires a downturn and why conviction stays low until one arrives.
Sources
June-quarter combined service revenue +33% YoY vs +20% for equipment; service outgrew equipment at most semicaps; long-term targets Lam 10%+, ASM 12%+, KLA 13-15%, AMAT mid-teens, two most recent revised upward; service funds dividends — linkacc 2026-08-03