Micron Technology (MU) — Company Deep Dive

Cover Metrics + Verdict

§1 Investment Thesis at a Glance

Thesis 1 · This is a "fundamental inflection confirmation" event; the 6/23 selloff should be classified as deleveraging, not an industry inflection.

On 6/23 MU fell -13.18% in a single session (closing at $1,051.77). The drop coincided in time with three events: forced liquidation of Korean leveraged ETFs, high-beta de-risking ahead of MU's print, and a broad systematic consolidation that day. It was more likely positioning/liquidity-driven, on three grounds: (1) there was no new MU-specific fundamental bad news that day (earnings had not yet been released); (2) Korea's financial regulator in May cleared 16 single-stock 2x leveraged ETFs tracking Samsung/SK Hynix, and foreign profit-taking combined with leveraged position unwinds drove Korean equities intraday -10%, tripping a 20-minute circuit breaker — an observable liquidity event with prior timing; (3) 48 hours later the earnings beat drove a +13% recovery, which is incompatible with "fundamental deterioration drove the drop".

Magnitude check: a single-day -13.18% far exceeds the -3 to -5% pullback a beta ~1.4 cyclical should see in a routine systematic consolidation (~-2 to -3%) — that 8–10pp gap cannot be explained by fundamentals and must be filled by a leverage / crowded-positioning amplifier (Korean 2x leveraged ETF unwinds + concentrated de-risking of the high-momentum bucket ahead of the print). The Korean ETF forced liquidation was therefore one trigger, not the sole root cause; pre-earnings de-risking and the systematic consolidation are coexisting competing explanations, all pointing to a "flows over logic" pattern — a flow-driven divergence — rather than an industry inflection. The 6/24 print (revenue beat consensus $35.69B by 16%; EPS $25.11 vs consensus $20.49, a 22.5% beat) and the subsequent AH +13.1% recovery more likely invalidate the "cycle has peaked" panic.

Disconfirming condition: if the post-earnings price fails to reclaim the 6/23 drop (hold above $1,051), it signals that the market's core focus has shifted to earnings durability (the quarterly beat already delivered yet the drop unrecovered), and Thesis 1 (pure deleveraging) must cede to Thesis 5 (cycle-top concern).

Thesis 2 · Shipments are supply-constrained — this is the root of pricing power.

Management explicitly states that "shipment growth is constrained by supply, not demand"; HBM3E + HBM4 is sold out through calendar 2027, with demand spilling into 2028. At the industry level, the DRAM/NAND/HBM supply-demand gaps reach 4.9% / 4.2% / 5.1% (the widest since 2011), with DRAM inventory at just 2–3 weeks. Q2 2026 DRAM contract prices are up +58–63% QoQ, NAND +70–75%.

Implication: in a supply-constrained, no-spot-market structure, price is set by the supply side — the industry basis for an 84.9% gross margin. Disconfirming condition: TrendForce quarterly contract price QoQ growth narrows to single digits, or spot (rather than contract) prices begin to soften.

Thesis 3 · Long-term agreements (Strategic Customer Agreements) have partially "bond-ified" cyclical volatility, but coverage is limited.

16 long-term agreements totaling $22B committed (including ~$18B in cash prepayments), take-or-pay, non-cancellable, covering roughly 20% of DRAM bits and 1/3 of NAND bits; the company's long-term goal is to take long-term agreements to 50% of revenue.

Implication: take-or-pay locks the volume of ~20% of DRAM + 1/3 of NAND bits, which likely dampens (rather than eliminates) the earnings-volatility amplitude of the next downturn. The magnitude must be kept straight: long-term agreements cover volume, not price; the other ~80% of DRAM bits, and the take-or-pay "floor" itself, is a committed quantity, not a committed gross margin — once the cycle rolls over, the uncovered bulk falls with contract prices, and the long-term agreements hedge only part of it. Competing explanation: this cycle's smaller volatility could also stem from a rising structural HBM mix (high margin, sticky demand) rather than the long-term agreements themselves; the two are hard to attribute separately.

Disconfirming condition: if progress toward the 50% long-term-agreement goal stalls, or if newly signed long-term agreement prices come in below current contract prices, the agreements' "volatility-dampening" role is invalidated and the pricing-power inflection is near.

Thesis 4 · Current earnings are the absolute peak and cannot be extrapolated linearly.

FY24 gross margin ~23% → FY25 41% → FQ3 FY26 84.9%. FY26 actual EPS ~$73 (the old consensus was $58); the Q4 annualized run-rate reaches ~$124, implying a forward P/E of just ~16x (on $73) and as low as ~9.6x (on the annualized figure) — this is more likely the valuation trap of a deep cyclical: a low forward P/E does not equal cheap. A low forward P/E and "earnings about to roll over" are correlated, not inherently causal: the same low P/E is consistent both with "the market expects peak E to be unsustainable" (cyclical read) and with "the market underprices AI permanently re-rating memory" (growth read). This report judges the former to be more likely dominant, on two independent pieces of evidence (rather than the P/E itself): (1) an 84.9% gross margin has no sustainable precedent in memory history (the denominator E sits at a physical peak); (2) the FY23 empirical floor — the last cycle bottom printed a gross margin of -7.7% and a loss of $4.45/share, an EPS swing of roughly $77 over one full cycle.

Implication: valuation must separate peak from normalized earnings (see §5). Disconfirming condition: if sustained supply discipline (capex/wafers) extends this "super-cycle" beyond 2028, the peak assumption needs upward revision.

Thesis 5 · The supply response is the sword hanging over 2028.

Micron's FY26 capex is raised to ~$27B (net of government subsidies), with FY27 jumping to mid-$40B+ — the most aggressive investor among DRAM makers; the ID1 (Idaho) new fab ramps in 2027, with greenfield capacity making a "more meaningful contribution" from 2028. The historical pattern is that memory price peaks tend to follow a capex step-up by roughly 18–24 months (the physical lag from investment to capacity ramp) — but this is an inference, not a confirmation; the mechanism holds only if two undetermined conditions are met: (1) capex actually converts into effective wafer output; (2) demand does not simultaneously expand to absorb the new supply (if AI demand growth stays >35%, outpacing supply at ~16%, the new capacity is not necessarily excess). Reverse / common-cause check: a capex step-up and a price peak may also both be outcomes of the same factor (a cyclical high) — strong conditions → ample cash → confidence to expand, while strong conditions themselves are already near the top. We therefore position this as "the variable most worth tracking", not as a mechanical rule that "a capex jump must mean a price peak".

Implication: this is the most important source of downside risk and the origin of "the key metric most worth tracking" — but its trigger must be confirmed by actual supply-demand gap data (rather than the capex number itself).

Pricing & valuation nuance

The market is buying MU as "a cheap AI name at a forward P/E of 12x". At the absolute peak of the memory cycle, the lower the forward P/E, the more dangerous it is — it is the early pricing of an earnings downgrade. The right thing to look at is mid-cycle normalized earnings, not a $100 peak EPS used as the anchor. The market is pricing a deep cyclical with a growth-stock framework (P/E / PEG), and that is the central pricing error.

Key indicators to track

Metric: the transaction mode of HBM / high-end DRAM (long-term take-or-pay vs spot) + DRAM quarterly contract price QoQ.
Current level: HBM3E/HBM4 sold out through calendar 2027, no spot market; Q2 2026 DRAM contract price +58–63% QoQ.
Threshold: (1) long-term agreements converting to spot transactions upon expiry (rather than renewing as long-term agreements); (2) DRAM quarterly contract price turning negative QoQ for the first time.
Sources: TrendForce quarterly contract prices, Micron quarterly call long-term-agreement disclosures, SK Hynix/Samsung contract policy.
Dates: next observation window = MU FQ4 print (~2026-09-22 to 29) + TrendForce 2026Q3/Q4 contract-price updates.
Implication once crossed: either threshold tripping = cycle top confirmed; the thesis flips from "hold" to "reduce", and the valuation anchor shifts from peak to normalized.

§2 Business Model

Product lines. Micron is one of the world's three largest DRAM makers and one of a handful of vertically integrated NAND manufacturers (forming a DRAM oligopoly with SK Hynix and Samsung; in NAND there is also SanDisk/Kioxia). Two technology pillars:

• DRAM (the profit engine this cycle): standard DDR5, LPDDR5 (mobile), GDDR (graphics), and most critically HBM (high-bandwidth memory) — multiple DRAM dies stacked in 3D, the "standard-issue memory" for AI training/inference GPUs and ASICs, with per-GB value and gross margin far above ordinary DRAM.
• NAND (flash): enterprise SSDs, client SSDs, mobile UFS — benefiting from AI data-center storage expansion.

Reporting segments (FY26 switched to four end markets, replacing the legacy CNBU/MBU/SBU/EBU)

(Segment gross margins are on the company's disclosed basis; the four-unit revenue sum of $41.448B differs slightly from total revenue of $41.456B due to disclosure rounding and "other".)

Customer structure and concentration. End customers are highly concentrated among a few hyperscalers (including indirect demand pulled by the NVIDIA platform + direct Microsoft/Google/AWS/Meta data centers), and HBM in particular is "tied to GPU roadmaps" — FQ3 noted that HBM4 was shipping in volume for the NVIDIA Vera Rubin platform (ramping from March 2026, at roughly 2x the pace of HBM3E 12-high). Customer concentration is a double-edged sword: it provides the counterparties for long-term agreements, but it also means a single customer's roadmap change (e.g., a GPU delay) amplifies volatility.

Monetization and gross-margin logic. Memory is commodity-style priced (a price-taker in commodity DRAM), but HBM/high-end DRAM enjoys episodic pricing power thanks to technical barriers + capacity binding. Gross-margin logic = (contract price − cash cost) × bit shipments; in a shortage + price-increase cycle, price elasticity far exceeds cost, and gross margin spikes non-linearly (this is the source of the 23% → 85% move).

Capital intensity and operating leverage. Extremely asset-heavy: FY26 capex ~$27B (net of subsidies), FY27 mid-$40B+. Once capacity is in place and prices are rising, operating leverage is stunning (revenue +74% QoQ with opex nearly flat → operating margin 80%+); conversely, in a downturn fixed depreciation and amortization rapidly compress gross margin back down. This is the two-sidedness of a deep cyclical's operating leverage.

Transformation / structural changes: (1) segment reporting switch (FY26); (2) long-term-agreement mix target of 50% (structural volatility reduction); (3) US domestic capacity (Idaho ID1, New York) + CHIPS subsidy ($6.4B) reshaping the cost and geopolitical structure.

§3 Industry & Competitive Landscape

Value-chain position. Micron sits at the "memory device manufacturing" node, upstream of which are WFE equipment (ASML/AMAT/Lam Research) and silicon wafers, and downstream of which are GPU/ASIC makers (NVIDIA/AMD/Broadcom/Google TPU) and OEMs. HBM partially upgrades Micron from a "cyclical commodity supplier" to a "key component supplier for AI compute", but in essence it remains an oligopoly manufacturer.

Peers and comparability

Why peer multiples mislead: SK Hynix YTD +340%, SanDisk last-quarter revenue +251% — these "low forward P/Es" (SanDisk 11.76x) are low precisely because the entire industry sits on peak earnings, with the denominator (E) at an unsustainable high. A cross-sectional P/E comparison on peak E makes all three look "cheap" — that is the cyclical synchronicity of moving up and down together, and it does not constitute an independent margin of safety for MU. On HBM share, MU is only 21% vs SK Hynix's 61%, which means MU has both a "share-catch-up" upside option and the reality that in HBM — the highest-margin node — it is a share follower, not a price leader.

Supply-demand / pricing power / substitution

• Supply-demand: DRAM/NAND/HBM gaps of 4.9%/4.2%/5.1% (the widest since 2011), DRAM inventory 2–3 weeks, NAND 3–4 weeks. Supply growth (~16%) structurally lags demand growth (~35%).
• Pricing power: the supply side currently dominates; Samsung/SK Hynix have shifted to quarterly contracts (refusing long-term agreements) in order to raise prices each quarter — a real-time manifestation of pricing power.
• Substitution / switching cost: HBM is deeply tied to GPU roadmaps (HBM4 → Vera Rubin), with long qualification cycles and high switching costs, conferring short-term stickiness; but once the next generation (HBM4E) reaches volume, the highest-margin node is the most prone to localized oversupply from "new-node capacity + yield breakthrough".

TAM (provided only where it affects valuation and is sourced): management says HBM TAM will "comfortably exceed $100 billion in 2027" (pulled forward from a prior 2028 expectation). This is the quantitative anchor of the upside narrative, but TAM expansion ≠ Micron's share and margin expanding in lockstep, and must be discounted by share (target = DRAM share ~22%).

Korean-equity (Samsung/SK Hynix) → US-equity (MU) spillover channels and their validity

The core analysis is single-stock fundamentals, but the 6/23 selloff was more likely triggered by a Korean-equity liquidity event (rather than Korean equities being the sole transmission source), so the spillover channels are decomposed and classified as Direct/Indirect/Sentiment — the key is to distinguish which channel is a true signal and which is mere synchronized noise:

Spillover assessment: (1) Leading — on 6/23 Korean equities fell first and MU followed in sync, while the 6/24 MU earnings beat more likely shows the follow-down was sentiment noise rather than a true change in expectations; (2) Overridden — this is more likely the essence of 6/23: fundamentals (the print would beat) were temporarily overridden by flows (forced leverage liquidation + pre-earnings de-risking); (3) Overshooting — US memory stocks' elasticity to the Korean leverage event (6/23 single-day -13.18%) was markedly larger than their fundamentals should warrant. Magnitude calibration: with no new MU-specific fundamental bad news that day, applying MU beta ~1.4 to the day's systematic consolidation of ~-2 to -3% implies a fundamentals-explicable pullback of about -3% to -5%; actual -13.18% minus a reasonable -3 to -5% leaves a gap of ~8–10pp attributable to a leverage / crowded-positioning amplifier (i.e., roughly two-thirds of the drop is sentiment/flows and only about one-third can barely be called fundamental). The next-day +13.1% nearly full recovery, on its own, is insufficient to confirm the overshoot reading; what supports it is the addition of two independent pieces of evidence — "no fundamental bad news before the drop" + "a drop magnitude far exceeding the beta-warranted reaction". On balance this is more likely an overshoot. Conclusion: of the three channels, Direct (contract-price synchronization) is the true signal usable for MU pricing, and the Sentiment channel is weighted at zero.

§4 Financial Analysis

4.1 Multi-year revenue and earnings history (full cycle, including the empirical cycle bottom)

FY26 bottom-up reconciliation (replacing stale consensus)

Key correction

The "FY26 consensus of $108.7B revenue / $58 EPS" still circulating is a pre-Q3 stale figure and is directly contradicted by the company's own quarterly numbers — the first three quarters alone have already realized $78.96B of revenue and $42.09 of EPS, and the fourth-quarter guide alone is $50B / $31, putting the full year at ~$129B / ~$73. Any forward P/E derived from $58 EPS (yielding ~20x) overstates the multiple; on the $73 true peak EPS, the current price of $1,186 implies a forward P/E of just ~16.2x; on the Q4 annualized run-rate ($31×4 = $124), only ~9.6x. This "the forward P/E looks even cheaper" effect is a danger signal at the cycle top (see §5.3).

Read

• FQ3 revenue of $41.46B beat consensus $35.69B by +16.2%; non-GAAP EPS of $25.11 beat consensus $20.49 by +22.5%. This is a 22% beat on top of an already very high bar, driven by real demand rather than a flattering low base.
• FY26 revenue jumps from FY25's $37.4B to ~$129B (3.4x), driven mainly by price (contract price +58–75% QoQ) + HBM mix upgrade, with bit shipments a secondary contributor (shipments are supply-constrained). The magnitude of this attribution checks out: contract price alone is +58–75% QoQ, and HBM's per-GB value is several times that of standard DRAM — together these are enough to explain the bulk of the 3.4x revenue, while bit volume can only grow single digits to low double digits under supply constraints. Hence the characterization as a "price-increase cycle" rather than a "volume cycle" — more sensitive to the cycle top (price can fall faster than volume, because price has no capacity stickiness).
• The empirical yardstick for earnings amplitude: from FY23's -7.7% gross margin / a $4.45 loss to FY26's ~85% gross margin / ~$73 of earnings, per-share earnings swing roughly $77 within one full cycle (from -4.45 to +73). Any normalized valuation must keep this swing in mind (see §5.3).

4.2 Cash flow and capital allocation (FQ3 FY26)

Earnings quality

• Growth organic or via M&A? Organic (price increases + HBM), no material M&A.
• Does net income convert to cash? OCF/net income ~88%, good conversion, not receivables inflation.
• Is FCF inflated by working-capital timing? Receivables/inventory turns are healthy in a shortage cycle; FCF of $18.3B is real.
• Does capex understate the next phase? No, it is front-loaded: FY27 capex of mid-$40B+ is nearly 2x FY26 — a classic cycle-top signal (high profits spawning aggressive expansion).
• Is the share base expanding? Diluted share count is 1.145B; the company uses buybacks as the primary capital return, planning to increase returns after Dec 9 (the second anniversary of the CHIPS agreement); the share base is expected to stabilize/decline, with no dilution.

4.3 Key-ratio snapshot

• Operating margin (non-GAAP) 81.2% — rare in semiconductor history, reflecting extreme operating leverage.
• Net cash ~$24.5B, net cash / market cap ~1.8% — ample cash but, relative to a trillion-dollar market cap, not a valuation support.
• ROIC: extremely high in the peak period (single-quarter profit of $28.86B vs the capital base), but not extrapolable; it will fall sharply in the normalized period.

§5 Multi-Method Valuation

5.1 Rationale for method selection

• EV/EBITDA peer comparison — the first choice and market anchor for deep cyclicals / asset-heavy names. MU forward peak EV/EBITDA ~9–13x and per-peer actuals are in §5.2. Distinguish peak from LTM EBITDA.
• Mid-cycle normalized P/E — the core method here: a deep cyclical cannot have a P/E applied to peak EPS; normalized earnings must be estimated bottom-up (bit volume × ASP decline × cash cost, anchored to FY23 empirics, see §5.3).
• Peak P/E — computed on FY26 true EPS $73 / Q4 annualized $124, used as the upside boundary, not relied on alone.
• DCF (sensitivity only, not included in weighting) — cash flow sits at an unsustainable peak; terminal value is >65% of EV, sharing the same assumption set as the scenario method with no independent information; narrowed to a 3×3 sensitivity matrix yielding a median ($1,015) that is excluded from valuation weighting (see §9), used only as cross-confirmation.
• Narrative scenario valuation (the primary method) — the bridge from business evidence to numbers, addressing the core uncertainty of "how much longer the cycle can run"; natural weighting output, with no manual mark-up.

5.2 Relative valuation vs peers (peer table)

MU forward peak EV/EBITDA calculation (sum of FY26 four-quarter actual operating profit)

EV = market cap $1.37T − net cash ~$24.5B ≈ $1.345T. Summing FY26 four-quarter non-GAAP operating profit by quarter:

• Blended op margin = $98B / $129B ≈ 76% (four-quarter actual weighted). Peak EBITDA = operating profit ~$98B + D&A ~$8B ≈ ~$106B.
• Forward peak EV/EBITDA = $1,345B / $106B ≈ 12.7x; on the Q4 annualized run-rate EBITDA, ≈ 9x. Taking a range of ~9–13x peak, the four-quarter sum puts the midpoint near 13x.
• Why LTM EV/EBITDA (~31x) differs so much from forward (~9–13x): the LTM denominator still includes FY25's weak quarters, understating current earnings power; the forward peak is a meaningful basis, but its denominator is unsustainable peak EBITDA. Neither is cheap nor expensive; the key remains "how many years the peak can be sustained".

5.3 Mid-cycle normalized valuation (core — bottom-up derivation)

The real question for a deep cyclical: what is normalized (mid-cycle) EPS? We use a three-step bottom-up derivation, validated against a floor from Micron's own history. Derivation path = bit shipments × (mid-cycle ASP − cash cost) × margin pass-through − opex − tax.

Step 1: bit-shipment base. This cycle is a "price-increase cycle", not a "volume cycle", but long-run bit-demand CAGR is DRAM ~20–25%/yr and NAND ~15%/yr. By the normalized year (assume 2028–29), bit shipments grow roughly 30–50% above FY25 from natural growth (including ID1/greenfield capacity ramp). This is the structural reason normalized revenue is higher than history.

Step 2: mid-cycle ASP decline (the main damage). The historical pattern is a 40–60% contract-price decline from peak to cycle bottom (the FY22→FY23 blended ASP fell ~50%+, taking gross margin straight from positive to negative). We assume a milder decline this cycle: 35–45% from the FY26 peak ASP (because take-or-pay floors ~20% of DRAM + 1/3 of NAND bits).

Step 3: cash cost + margin pass-through (anchored to Micron empirics).

• Empirical floor (hard evidence): full-year FY23 non-GAAP gross margin -7.7%, a loss of $4.45/share; the worst single quarter (FQ3'23) gross margin -16.1%, operating margin -39.2%, EPS -$1.43. The floor is negative, not $0.
• Empirical mid-cycle anchor (hard evidence): FY25 was the last full "mid-cycle" year, gross margin 41%, EPS $8.29. This is the empirical reference for "moderate pullback but no collapse".
• Reasons for a structural upward revision: a rising long-run HBM share of DRAM + the 50% long-term-agreement goal reducing ASP volatility lift the mid-cycle gross-margin midpoint from the historical ~30% to ~38–45% — this is "better than the last cycle", not "80% forever".

Normalized EPS — three tiers (bottom-up)

• Take neutral normalized EPS ≈ $20 as the primary anchor (gross margin ~40%, revenue ~$57B, diluted shares ~1.13B). The floor is guarded by the FY23 empirics (negative).
• Apply a reasonable cyclical P/E (12–18x) to normalized EPS: neutral $20 × 15x = $300; upper $24 × 18x = $432; lower $10 × 13x = $130.
• Mid-cycle normalized midpoint ≈ $300, range $130–$432 (assuming "peaks on schedule and reverts to normalized", excluding any super-cycle-extension premium).

Method-divergence warning

The peak method (FY26 true EPS $73 × 14x = $1,022, or Q4 annualized $124 × 10x = $1,240) and the normalized method ($20 × 15x = $300) differ by >3x (>100%), which directly drives confidence to "low". The two represent, respectively, the "cycle-top market price" and the "mid-cycle clearing price", and true value depends on how many years the peak can be sustained. The scenario table below bridges this time dimension; the fact that divergence is >3x is itself the reason we cannot assign "medium" confidence (see §5.6).

5.4 Three-scenario table (narrative → numbers)

Each EPS anchor comes directly from the foregoing reconciliation: peak true EPS $73 (FY26) / Q4 annualized $124; transition EPS between peak and normalized; normalized EPS $20 (§5.3 bottom-up). The weighted value is the table's natural output, with no after-the-fact mark-up.

(Probabilities sum to 25% + 50% + 25% = 100%. Arithmetic reconciliation: Bull $1,440×0.25 = $360; Base $850×0.50 = $425; Bear $260×0.25 = $65; sum of the three = $850, the valuation midpoint.)

Base case two-year EPS path

The earnings anchor for the Base per-share value is derived from a bottom-up build (volume × ASP × GM → revenue → op margin → net → EPS) for each of FY27 and FY28, then summed and divided by two. The two-year basis aligns with the long-term agreements (take-or-pay locked through CY2027, demand spilling into 2028) + the capex realization timing:

Year	Revenue anchor (provenance/logic)	non-GAAP op margin	implied net (12% tax)	EPS (÷1.13B shares)	Cycle positioning
FY27	~$150B (between FY26's $129B and the Q4 run-rate annualized ~$200B)	58%	$150B×58%×0.88 ≈ $76.6B	≈ $68	high but slowing
FY28	~$105B (moderate pullback, not a collapse: ASP falls ~30–35% from peak)	40%	$105B×40%×0.88 ≈ $37.0B	≈ $33	moderate pullback
Two-year average	—	—	—	(68+33)/2 ≈ $50	Base earnings anchor

Derivation conclusion: two-year average EPS = $50.2 ≈ $50, ×17x = $854 ≈ $850, consistent with the Base per-share value. FY27 ~$68 is already markedly below the FY26 peak of $73 (slowing), FY28 ~$33 falls further (moderate), and neither touches the Bear normalized $20. Disconfirming condition: if actual FY27 EPS falls below ~$55, or FY28 below ~$25 (sliding toward the Bear path), the Base two-year average shifts down, the Base price is cut from $850, and the midpoint shifts down in tandem.

Justification for the Base 17x multiple

The reasonable mid-cycle multiple for a pure-cyclical memory stock is 12–15x. The Base case uses 17x, above this range, justified by one specific normalized-gross-margin assumption: this cycle's take-or-pay long-term agreements + a rising structural HBM high-margin mix lift the mid-cycle gross-margin midpoint from the historical ~30% to ~40–45% (see §5.3 neutral tier). For every ~10pp upward revision to the gross-margin midpoint, the reasonable multiple for a cyclical rises about 2–3x (smaller earnings amplitude → an implied decline in the discount rate). Thus 12–15x (historical GM ~30%) → 17x (this cycle GM ~40–45%) is a direct result of this gross-margin assumption.

A checkable anchor for "+10pp GM → +2–3x multiple" (source: Nasdaq / Simply Wall St 2026-06): memory stocks exhibit a repeatedly validated "inverted" historical pattern — at the absolute earnings peak (high GM), the market awards only 4–8x forward P/E (peak trailing P/E can be as low as ~3x); when the cycle turns down and GM goes negative/thin, the multiple expands to 18–24x (the denominator E approaches zero while the price has already bottomed first). This "low multiple at peak / high multiple at trough" shows that the multiple the market gives memory stocks is essentially pricing "the sustainability of current GM". Applied in reverse to a mid-cycle anchor: when the sustainable mid-cycle GM midpoint is revised from ~30% (corresponding to 12–15x) to ~40–45%, adding 2–3x to 17x is directionally consistent with the historical scatter. This is an "approximate anchor", not a precise historical point. Implied COE back-out as a backstop: 17x ≈ 1/(COE − g); with perpetual g = 2.5% → implied COE ≈ 8.4%; 12–15x → implied COE ≈ 9.2–10.8%. That is, "raising the GM midpoint from 30% to 42% compresses implied COE by ~1–2.4pp", self-consistent in order of magnitude with "+10pp GM → +2–3x". Disconfirming condition: if the actual 2028–29 mid-cycle gross margin falls below 35%, the 17x should pull back to 13–14x, cutting the Base price from $850 to ~$680, with the midpoint shifting further down.

Note on the midpoint basis

The valuation midpoint = the scenario table's natural weighted value = Base case $50 × 17x = $850, with no off-table mark-up or reverse-fitting. The three upside arguments — super-cycle extension / long-term-agreement support / HBM premium — are all decomposed into the EPS × multiple of the Bull ($120 annualized EPS) and the Base (17x supported by the GM assumption), with no additional coefficient applied. Against the current price of $1,186, the $850 midpoint implies roughly -28% downside — the current price sitting nearly 30% above the midpoint reinforces the core "limited-win, large-loss" thesis.

5.5 Multi-method valuation cross-validation table

Where the divergence comes from: the largest divergence is peak vs normalized (peak ~$1,022–1,190 vs normalized midpoint $300, a gap of >3x), not WACC or peer incomparability. The three "pure-peak bases" (peak P/E $1,022, peak EV/EBITDA in line, current price $1,186) all land in $1,000–$1,200, while the scenario-weighted value, which bridges the probability of a cyclical pullback, is only $850 — this ~$336 gap (current price $1,186 vs scenario midpoint $850) is precisely the core divergence between the market and this analysis: the market essentially prices it on pure peak earnings, applying almost no discount for "the 2028 cycle reverting on schedule", whereas the scenario table assigns the Bear 25% weight (normalized midpoint $300) and the Base 50% a moderate pullback, naturally compressing the midpoint to $850. The current price is already ~28% above scenario fair value. Upside requires the super-cycle to extend beyond expectations (Bull $1,440, only +21% of room); downside risk comes from the cycle reverting on schedule (Bear normalized midpoint $300, -75% from spot). DCF does not participate in weighting (see §9), because a terminal-value share of >65% makes it merely a re-shell of the above assumptions, with no independent information.

5.6 Confidence determination

§6 Management & Insider Activity

Insider transactions: CEO Sanjay Mehrotra's Form 4 and 10b5-1 plan disclosures are covered below (SEC Form 4 filings and contemporaneous reporting, as of 2026-06); the Forms 3/4/5 of other executives/directors (CFO, business-unit heads) are not covered line by line, so no assertion is made on management's overall direction. 10b5-1 plan sales are not treated as a bearish signal.

CEO insider transactions (Form 4 / 10b5-1)

Read (is this "planned reduction at the cycle peak"?) — the core is the plan-adoption date, not the sale date

• Both tranches were executed under the same Rule 10b5-1 plan, adopted on 2026-01-30. When the plan was adopted in late January, MU was still around ~$500 (the ~$520 execution price on the 5/1 tranche reflects that level), before this year's parabolic 1–6 month run-up (from ~$500 to $1,050+).
• This does not constitute a bearish "management opportunistically cashing out at the peak" signal: the sale prices, quantities, and dates were preset in January, and the CEO did not know at adoption that it would rise to $950 — the 5/29 tranche landing at a high is automatic plan execution. This is exactly the design purpose of a 10b5-1 plan (insulating the information advantage), and reading it as "the insider calling the top" overstates the signal.
• One neutral observation: the plan "continuing to execute on its set cadence near the peak" itself shows management did not actively pause/curtail the plan in May (10b5-1 allows modification/termination within a compliance window) — interpretable as "management did not use action to express the strong signal that 'the stock is undervalued and shares should be kept'", far weaker than "active peak cash-out", and used only as background, not entering the thesis.

Disclosable capital-allocation facts (from the print / call)

• Dividend recently raised 30%; management "intends to grow the dividend gradually".
• Buybacks are the primary capital-return tool; subject to CHIPS agreement terms, the plan is to increase shareholder returns after 2026-12-09 (the second anniversary of the CHIPS agreement signing).
• There is a historical $10B buyback authorization (no expiry, not mandatory, constrained by CHIPS).

§7 Guidance & Outlook

Latest management guidance (FQ4 FY26, quarter ending ~August 2026)

Key management remarks (call)

• HBM3E + HBM4 "sold out through calendar 2027, with demand extending into 2028"; HBM4E qualification in progress.
• HBM TAM "comfortably exceeds $100 billion in 2027" (pulled forward from 2028).
• HBM share target = "in line with DRAM share" (i.e., converging toward ~22%, currently 21%).
• Long-term agreements: 16 in total, $22B committed (including ~$18B in cash prepayments), take-or-pay, covering ~20% of DRAM bits + 1/3 of NAND bits, with a long-term goal of 50% of revenue.
• Capex: FY26 ~$27B, FY27 mid-$40B+, mostly invested in fab construction; greenfield making a "more meaningful contribution" from 2028.
• "Shipment growth is constrained by supply, not demand."

Guidance exclusions / risk language: the company gave no specific "fulfillment ratio" guidance (because allocation strategies differ across segments such as automotive/defense); buybacks are constrained by CHIPS.

What would confirm / disconfirm the thesis

• Confirm: FQ4 (~9/22–29) again beats the $50B guide + new long-term agreements renewed at prices not below the current price + TrendForce 2026Q3 contract prices continuing positive QoQ.
• Disconfirm: contract-price QoQ growth narrows to single digits or turns negative + long-term agreements converting to spot upon expiry + capex realization driving a supply inflection in H2 2027.

Next earnings date: approximately 2026-09-22 to 09-29 (FQ4 FY26).

§8 Catalysts & Risks

8.1 Upside-catalyst table

8.2 Downside-risk matrix (with probability/impact + observable signal)

8.3 Bull/Base/Bear operating scenarios (one line each)

§9 DCF Assumptions & Sensitivity

In the peak period DCF is used for sensitivity only and is not included in valuation weighting; narrowed to a sensitivity matrix yielding a median, it does not participate in weighting.

Core assumptions

Sensitivity matrix (per-share fair value, rows = normalized FCF, columns = WACC; perpetual g = 2.5%)

• DCF median (neutral FCF × neutral WACC) ≈ $1,015, full-matrix range $670–$1,560.
• Terminal value as a share of EV >65% (the normalized-period cash flows are suppressed by the front-loaded fade, with value mainly in the perpetuity segment), so DCF is essentially "a re-shell of normalized FCF + perpetual-growth assumptions" — it shares the same assumption set as the scenario/relative methods, merely re-expressed, and does not constitute independent evidence.
• Conclusion: DCF is not included in weighting. Its median $1,015 is in line with the "pure-peak bases" (peak P/E $1,022, current price $1,186) and markedly above the scenario-table weighted midpoint of $850 — exposing DCF's systematic upward bias in the peak period: it takes the elevated near-term FCF as the starting point, implicitly assigning too much weight to "peak continuation". The scenario table corrects this bias with the 25% Bear weight (normalized midpoint $300), arriving at $850. DCF is "the scenario method with too little cyclical-pullback discount applied", used only as cross-confirmation.

§10 Sources & Disclaimer

Itemized sources (filing/IR-grade)

Disclaimer

Produced by NewsLiquid; all data and conclusions are independent research outputs. Deep-cyclical valuation depends heavily on the unknowable variable of "cycle duration", and peak earnings are not extrapolable; this report provides ranges and scenarios rather than a false-precision single target price. It is not investment advice.