Phase Space AI

Valuation

Lam Research [LRCX]

Lam Research [LRCX] — Valuation

As of 2026-07-29 · framework v1.5.1 · spot $269.51 · shared comparator set: KLAC, LRCX, AMAT, ASML, TER

0. COMPANY STATE — declared first

STATE B — cyclical / commodity-sensitive.

Dating the cycle position — required, and not the same as normalising

valuation.md: "A median of annual observations over a partial window inherits that window's start point." So the window is stated explicitly. FY2022-FY2026 contains one complete WFE cycle — the FY2022 peak (31.2%), the FY2024 memory trough (28.6%) and the FY2026 peak (35.3%). A median across it is a mid-cycle figure with a dated start, not an artefact of where the window happens to begin.

Where the forecast begins in the cycle: at or very near the peak. Three independent markers, none of them a forecast: 1. Backlog coverage is falling while revenue rises (see LRCX_Research.md §2). 2. China revenue share has already rolled over from its FY2024 peak. 3. The name trades at the 96th percentile of its own ten-year TTM P/E history.

1. Terminal margin — set from the company's own economics

Terminal EBIT margin: 31.2%

Basis. Own FY2022-FY2026 median operating margin (31.2%), spanning the FY2022 peak (31.2%), the FY2024 memory trough (28.6%) and the FY2026 peak (35.3%) — one complete cycle, dated. 4.1pp below the FY2026 trailing actual: Lam's incremental margin in FY2026 was earned on a memory-upgrade mix that its own systems-market disclosure shows is not yet a mix shift.

Opex bridge. 47.3% gross - 11.4% R&D - 5.3% SG&A + 0.6% other/net = 31.2% EBIT; constraint 31.2% <= 47.3% satisfied.

Hard constraint check: m_EBIT,T (31.2%) <= m_gross,T (47.3%)satisfied, using the five-year median gross margin (47.3%) rather than the higher trailing figure (50.5%), so the constraint is tested on the more conservative of the two.

Relation to trailing actual. Trailing FY operating margin is 35.3%; terminal is 31.2%, a gap of -4.1pp. The brief requires that a terminal margin below the trailing actual be justified rather than assumed. The justification is a named causal bridge: this is a State-B business observed at a dated cycle peak, and the terminal year is a mid-cycle year by construction. The haircut is the cycle, not a confidence discount smuggled into an operating assumption (which rule 5 forbids). Note also that the terminal margin sits ABOVE the ten-year median of 29.4% — the 10-year window is deliberately NOT used, because it embeds a structurally lower-margin pre-2017 era and using it would reject a real structural improvement on the grounds that it exceeds precedent.

2. The exit multiple is not a free parameter

EV_T / EBIT_T = (1-t)(1-g/ROIC) / (WACC-g)

With t = 0.15, g = 0.04, ROIC = 0.35, WACC = 0.095:

(0.85)(1 - 0.04/0.35) / (0.095 - 0.04) = **13.7x**

Exit multiple used: 13.7x EV/EBIT. Today's traded multiple is 40.9x, so the underwritten compression is 67% (40.9x → 13.7x). Per rule 3, an exit multiple below today's is often correct and is not itself a bear assumption; the test is consistency with terminal growth, ROIC, reinvestment and cost of capital, and the identity above is what supplies it.

Reference-class validity (rule 6). The comparator set is KLAC, LRCX, AMAT, ASML, TER — a WFE oligopoly plus ASML (litho) and TER (test) as reference. It satisfies n >= 5 mature profitable firms with a matching operating model. But it carries almost no information in the dimension being matched, and that must be said rather than hidden: KLAC, LRCX and AMAT trade at 39.5x / 40.9x / 40.6x EV/EBIT across operating margins spanning 29.5% to 41.7% and growth spanning 11.4% to 16.4%. There is no slope. This is the same failure valuation.md records as "29.9x / 29.9x / 30.4x across 8pp of exit growth, no slope". A peer-median exit multiple for this cluster is therefore UNIDENTIFIED, which is precisely why the exit multiple here is derived from the identity and the peer figure is used only as a sanity check.

3. Implied-path test — the Valuation Criteria (PRIMARY long-horizon output)

reverse_dcf.py --spot 269.51 --shares 1250.571 --net-cash 1849 --revenue 23232.7 --years 5 --wacc 0.095 --terminal-margin 0.312 --exit-multiple 13.7

Required revenue CAGR 39.7%
Demonstrated revenue CAGR (5y) 16.4%
Margin (demonstrated − required) -23.3pp
Exit multiple solved at 13.7x
Implied compression from today 40.9x → 13.7x (67%)

VERDICT: FAIL. The price requires 39.7% revenue CAGR for five years against 16.4% demonstrated — a 23.3pp shortfall. Per criteria.md this would be PASS WITH ARGUMENT only if the excess were backed by a specific evidenced driver. The named driver (memory/HBM capex) is real, but the company's own backlog and deferred-revenue disclosure runs the other way — coverage is falling, not building — so the argument fails on the company's own evidence, not on narrative grounds.

Sensitivity — over the exit multiple, the highest-variance parameter

Exit EV/EBIT Required revenue CAGR
13.7x (identity-derived, used) 39.7%
13.9x 39.3%
16x 35.4%
18x 32.2%
22x 27.0%
26x 22.9%
30x 19.4%
35x 15.8%
40.9x (= today's traded multiple) ~12.4%
45x 10.1%

Flip point: the exit multiple at which today's price only requires the demonstrated 16.4% CAGR is 34.1x — against a traded multiple of 40.9x. So the bet embedded in today's price is not that LRCX grows: it is that LRCX still trades at ~34x EV/EBIT in 2031, having compressed barely at all, five years into a cyclical business. Stating it that way is the point of the instrument.

4. Twelve-month target — separate instrument, own multiple history

Target: $271.73 (+0.8% to spot $269.51).

Why the 90th percentile and not the median. This is the calibration judgement, made explicitly. The median (15.3x) would put the target at $125 (-54%) and would be mean-reversion by reflex — exactly the error the brief names as the most expensive recurring one in this project. The 90th percentile keeps the multiple in the top decile of the name's own range, conceding that this cycle is larger than its predecessors and that the earnings are real, while declining to underwrite the 96th percentile as a base case for a further twelve months. The whole 12-month result in this cluster is a multiple judgement, not an earnings judgement — earnings estimates are rising and that is already in the NTM figure.

Note on target-below-spot. Item B16 warns that a process whose every target sits below spot is expressing a house market view rather than valuing companies. That check is passed here on cluster construction rather than by assertion: two of the three targets come out above spot (KLAC +2.4%, LRCX +0.8%) and one below (AMAT −9.8%), from the same method and the same percentile anchor. The dispersion comes from the names' own multiple histories, which is what it should come from.

5. Trough / mid-cycle / peak earnings power (State-B required output)

On 1,250.6m shares, applying the name's own ten-year operating-margin markers to TTM revenue of $23,233m, at a 15% tax rate:

Cycle position Operating margin EBIT ($m) Net income ($m) EPS P/E at $269.51
Trough 23.7% 5,506 4,680 3.74 72.0x
Mid-cycle (terminal) 31.2% 7,249 6,161 4.93 54.7x
Peak (trailing) 35.3% 8,200 6,970 5.57 48.4x

Note this holds revenue at the trailing level and flexes only margin, so it isolates margin cyclicality. Revenue is itself cyclical, so a true trough combines both and is worse than the top row.

6. Downside case with a named cause (Downside Criteria — MEASURED)

Named cause: the receivables build proving to be pulled-forward revenue. DSO rose 12.9 days as receivables grew 58.1% against revenue of 30.0%. If that reverses — i.e. if Q4 FY2026's 15.1% sequential revenue jump was shipment timing rather than demand — the September-2026 and December-2026 quarters absorb the payback, and Lam re-rates on decelerating revenue while the memory narrative is still priced in. Combine the FY2024 trough operating margin (28.6%) with a 25% revenue decline to $17.4bn — both printed inside the last three years — and EBIT is ~$5.0bn, net income ~$4.2bn, EPS ~$3.38. At the p50 of its own ten-year multiple (15.3x) that is $52/share, −81%. At the p75 (22.1x), $75/share, −72%.

Probability: 25% — the highest of the three, because Lam is the only name carrying the receivables flag and the largest gap between its multiple percentile and its backlog trend. Not a going-concern case: net cash $1.85bn. No going-concern flag.

Secondary named cause: NAND concentration in reverse. Memory has been a flat 42% of systems revenue for three years; the market is paying for that share to rise. If HBM capex is satisfied by DRAM capacity conversion rather than greenfield NAND, the share does not rise and the mix argument for the multiple disappears without any revenue decline at all.

7. Criteria summary

Criteria Type Verdict
Quality BINDING PASS — COMPOUNDER archetype (profitable, 50.5% gross / 35.3% operating margin, ROIC >> WACC, evidenced reinvestment in R&D at 11.4% of revenue). Note the screen tagged LRCX INFLECTION; that is wrong and is a downstream consequence of a defective operating-margin input (see LRCX_Financial_Model_Notes.md).
Valuation BINDING FAIL — required 39.7% vs demonstrated 16.4% CAGR, margin -23.3pp
Downside MEASURED scored, §6
Liquidity BINDING PASS — mega-cap, $337bn market cap; vehicle = equity
Momentum MEASURED negative near term — spot $269.51 vs prior close $269.59 (-0.0%) on a cluster-wide drawdown day
Catalyst MEASURED see LRCX_Catalyst_Calendar.md
Consensus MEASURED revisions one-sided positive; see §4
Peer Spread MEASURED 96th percentile of own 10y P/E; cluster shows no multiple dispersion (§2)
Sub-sector MEASURED Semi-cap equipment (WFE)

No position verdict is issued. The memo scores Criteria; the book decides.