MU 17 min read

Micron Technology (MU) — Investment Tree v1

0. Company Fundamentals — what Micron is and how it earns

Figures FY2025 (ended Aug 28, 2025) and latest quarter FQ2 FY2026 (ended Feb 26, 2026). Memory is deeply cyclical — the table shows the swing on purpose.

What it is & how it earns. Micron is one of three scaled memory makers (with Samsung, SK Hynix), producing DRAM (~79% of revenue) and NAND (~21%). Most output is a price-taking commodity with violent boom-bust cycles; the differentiated, supply-constrained slice is HBM — stacked DRAM contracted ("sold out" through calendar 2026, incl. HBM4) for AI accelerators, which is why a historical cyclical is now a scarce AI input. FY2025 revenue was $37.4B (+49% YoY); the FQ2 FY2026 run-rate hit $23.86B in one quarter (+196% YoY).

Cash-flow anatomy. The swing is the story — thin FCF in an investment year, a flood at the peak:

Net income went from $8.5B (FY2025) to $13.8B in a single quarter — and the same lines were negative in FY2023. This swing is the defining feature of the business.

Balance sheet & capital allocation. ~$16.7B cash/investments vs ~$10.8B debt = ~$5.9B net cash (Feb 2026), strong for a memory maker entering heavy capex; inventory days are falling (sold-out, not a glut). FY2026 capex guided >$25B (partly de-risked by ~$6.2B CHIPS Act funding); dividend $0.60/yr (<0.2% yield), buybacks modest (~$300M/qtr); R&D ~$1.25B/qtr.

What drives it. DRAM/NAND pricing and fab utilization (at a historic peak) plus HBM mix drive everything; the >$25B capex is the key risk. The question the tree resolves: a pure commodity cyclical to sell at the peak, or has HBM structurally raised the floor?

Section I — Company Identity

Micron Technology is a US-domiciled semiconductor memory manufacturer producing DRAM, NAND, and High Bandwidth Memory (HBM) for data center AI accelerators, cloud computing infrastructure, enterprise storage, mobile, and automotive customers. Headquartered in Boise, Idaho; listed NASDAQ (MU); ~$1.06T market cap as of June 2026. CEO: Sanjay Mehrotra (since 2017, ex-SanDisk co-founder). Single-class common stock; no dual-class structure.

Revenue segments (Q2 FY2026):

• DRAM: $18.8B (+207% YoY), 79% of revenue — HBM + data center DDR5 dominant
• NAND: $5.0B (+169% YoY), 21% of revenue — enterprise NVMe SSDs; Crucial consumer exited Q2 FY2026

The analytical taxonomy for this tree separates Micron's revenue by pricing mechanism — not by the DRAM/NAND product label that market models use:

†Tier C estimates — not separately disclosed by Micron as of June 2026. All HBM revenue figures are management-derived inference from blended margins.

Section II — H-0 Thesis

Micron is mispriced on lifecycle stage because cognitive anchoring on 30 years of memory-cycle boom-bust history has prevented the market from repricing Micron's HBM business as a structurally contracted AI-infrastructure component supplier, while structural blindness in sell-side model templates (no separate "HBM contracted revenue" line) locks the cognitive bias in place.

The simultaneous facts:

1. Q2 FY2026 revenue was $23.9B — nearly Micron's entire FY2025 full-year revenue ($28.6B) in a single quarter
2. Q3 FY2026 guidance: $33.5B at ~81% GM — exceeding any historical memory-cycle peak margin (which topped ~38-40% in 2018)
3. Micron's entire FY2026 HBM supply is sold out under finalized multi-quarter contracts
4. Sell-side consensus PT: $776-804 — below the current $919 stock price
5. FY2026 capex guidance: >$25B — not the behavior of a company managing a cyclical peak
6. Crucial consumer business exit complete — deliberate concentration into AI/enterprise

The contradiction: A company doing $23.9B in one quarter on its way to $33.5B, with 75-81% gross margins, sold-out capacity through 2026, exiting consumer to focus on AI, and spending $25B on new capacity — cannot simultaneously be a late-cycle commodity memory company whose earnings will revert to FY2023 trough levels ($15.5B full-year, deeply negative OI). Either the HBM economics are structurally different from commodity DRAM, or they are not. H-0 says they are — and sell-side models haven't caught up.

H-0 confidence: 65% (post-Stage 3; elevated from partial pre-tree estimate)

Section III — Market Consensus Map

Bull narrative (Wells Fargo $1,220, Susquehanna $1,750, Raymond James $1,100): HBM is not commodity memory — it is designed into AI accelerator chips at the chip-package level, with 2-4 year design cycles and 10-20× per-GB ASP premiums. Supply constraints are structural (TSV stacking technology, not just capital). AI hyperscaler capex is multi-year committed ($180-190B Google, $80B+ Microsoft, $131B Amazon for FY2026). At 11-14× forward non-GAAP P/E, MU is dramatically cheap versus AI-infrastructure peers (AVGO ~28-35×, TSM ~20-25×).

Bear narrative (HSBC $330, Standard Chartered CIO): 30-year memory cycle always reasserts. All "this cycle is different" arguments (2018 server DRAM, 2021 crypto NAND) ended the same way — supply flooding, margin collapse, 50-60% stock decline. Samsung, under mobile-division financial pressure, will break supply discipline as it did in 2012, 2014, 2018. Micron's $25B+ capex creates FY2027-2028 oversupply. Current margins (75-81%) are definitionally a cyclical peak.

The shared blind spot: Both bulls and bears treat Micron as a memory company with an HBM overlay. Neither camp has a model that treats HBM revenue as a structurally separate, non-cyclical segment and applies commodity-cycle adjustment only to the remainder. This structural blindness — locked in by Micron's own non-disclosure of HBM as a separate segment — is the primary mispricing mechanism.

Orphan premium: ~$115-143 above average sell-side PT — the market has begun pricing something the sell-side hasn't modeled. The question is whether that "something" is momentum/narrative (bear view) or the HBM-structural identity switch (bull view).

Current sell-side distribution: ~38 of 44 analysts Buy/Overweight; ~5 Hold/Neutral; ~1 Sell. Price-target range $330-$1,750 — a 5× spread consistent with a lifecycle-classification dispute, not a routine valuation disagreement.

Section IV — Category Taxonomy

Five revenue categories defined by pricing mechanism, not product type:

Micron Technology (MU)
│
├── Cat 1: HBM (High Bandwidth Memory — AI Accelerator)     [~25-35% of rev†, ~85-90% GM†]
│   Identity: Contracted AI-infrastructure component. Qualified per-chip. Non-cyclical.
│
├── Cat 2: Data Center DRAM (non-HBM Server/Cloud DDR5)     [~20-30% of rev†, ~60-70% GM†]
│   Identity: Semi-contracted, premium commodity DRAM. AI-server proliferation driver.
│
├── Cat 3: Client DRAM (Mobile / PC / AI-PC LP5X)           [~15-20% of rev†, ~40-55% GM†]
│   Identity: Spot/seasonal cyclical. AI-PC LP5X adds premium but remains consumer-facing.
│
├── Cat 4: Enterprise NAND (AI Data Center NVMe SSDs)        [~21% of rev, ~50-65% GM]
│   Identity: Semi-contracted enterprise storage. Crucial consumer exit complete.
│
└── Cat 5: Embedded / Automotive / Specialty                 [~4-6% of rev†, ~55-65% GM†]
    Identity: Long-cycle qualified. Growing, stable. Non-cyclical pricing.

Economic center of gravity: Category 1 (HBM) is the margin center — the blended 75-81% GM is impossible to explain without HBM at 85%+† gross margin. HBM is where nearly all incremental economic profit is generated in FY2026.

Category reassignment hypothesis: When HBM crosses 40-50% of total revenue (plausible by Q4 FY2026-FY2027†), the analytically correct peer group for MU's dominant revenue shifts from Samsung Semiconductor (memory cycle, EV/S 4-8×) to AVGO (contracted AI silicon, EV/S 15-25×). This multiple migration is the structural basis of the bull thesis.

Section V — Investment Tree

H-0: MU is mispriced on lifecycle stage — HBM has AI-infrastructure economics
     sell-side models do not separately value; identity-switch thesis.
     (H-0 confidence: 65%)
     |
     ├── L1A: HBM product economics                              ✅ Confirmed
     │   "Does HBM have contracted, platform-locked economics
     │    distinct from commodity DRAM?"
     │   |
     │   ├── L1A 1.1: FY2026 HBM supply sold out under binding contracts?
     │   │   ✅B — "committed" language confirmed Q2 FY2026 earnings;
     │   │          force-majeure terms undisclosed†
     │   |
     │   ├── L1A 1.2: HBM ASP maintaining 10-20× DDR5 commodity premium?
     │   │   ✅B — 75-81% blended GM trajectory confirms ASP premium
     │   │          holding; direct HBM ASP not disclosed†
     │   |
     │   ├── L1A 1.3: Customer switching from Micron to SK Hynix HBM
     │   │            faces 6-12 month re-qualification process?
     │   │   ✅B — Chip-package level qualification confirmed; duration
     │   │          Tier C inference†
     │   |
     │   └── L1A 1.4: HBM GM structurally >80%?
     │       ✅C — Algebraic inference from blended margin math†;
     │              direction confirmed, magnitude Tier C
     │
     ├── L1B: Supply discipline durability                        ⚠️ Holds near-term
     │   "Can the three-player HBM oligopoly maintain pricing
     │    discipline through FY2027?"
     │   |
     │   ├── L1B 1.1: Has Samsung achieved volume HBM4 production?
     │   │   ⚠️B — Samsung behind on HBM4 qualification as of June 2026;
     │   │          resolution path 1-2 quarters (LOAD-BEARING WATCH)
     │   |
     │   ├── L1B 1.2: HBM capacity constrained by TSV packaging,
     │   │            not just capital (>18mo leadtimes)?
     │   │   ✅B — TSV stacking bottleneck confirmed; exact lead-times
     │   │          Tier C†
     │   |
     │   └── L1B 1.3: No DRAM producer announced HBM oversupply
     │                before FY2028?
     │       ✅B — No credible oversupply announcement as of June 2026;
     │              sell-side supply-demand models confirm tightness FY2027
     │
     ├── L1C: Micron competitive position                         ⚠️ Improving; #3 position
     │   "Is Micron gaining HBM share and qualifying
     │    in next-generation programs?"
     │   |
     │   ├── L1C 1.1: Micron qualified in major AI accelerator programs
     │   │            (NVDA Blackwell, AMD MI300X, next-gen)?
     │   │   ✅B — NVDA Blackwell + AMD MI300X confirmed; hyperscaler
     │   │          custom silicon unconfirmed†
     │   |
     │   ├── L1C 1.2: Micron HBM4 achieves comparable bandwidth-per-watt
     │   │            to SK Hynix HBM4?
     │   │   ⚠️B — Management claims parity; not independently verified;
     │   │          SK Hynix first-mover in Rubin credible
     │   |
     │   └── L1C 1.3: Micron HBM4 yield ramp faster than HBM3E?
     │       ⊗B — Unresolved management claim; Q3 FY2026 earnings
     │             (June 24) is first resolution window
     │
     ├── L1D: Cycle-floor analysis                                ✅ Floor structurally higher
     │   "Does HBM contracted floor prevent FY2023-style collapse
     │    even in a commodity-DRAM bust scenario?"
     │   |
     │   ├── L1D 1.1: Bear-scenario quarterly floor with HBM contracts
     │   │            delivering $6B+/quarter?
     │   │   ✅C — Floor ~$15-16B/quarter vs FY2023 trough ~$3.9B/qtr;
     │   │          HBM revenue Tier C†
     │   |
     │   ├── L1D 1.2: FY2026 capex >$25B creates non-HBM overcapacity
     │   │            risk if demand normalizes FY2027-2028?
     │   │   ⚠️B — Overcapacity risk confirmed; HBM capacity non-fungible
     │   │          with commodity DRAM (partial mitigation)
     │   |
     │   └── L1D 1.3: FY2023 trough comparison vs HBM-floor scenario?
     │       ✅C — HBM floor ($20.4B+ annual gross profit) > FY2023 total
     │              gross profit (negative); floor scenario is not existential†
     │
     └── L1E: Valuation regime                                    ✅ SoTP supports ≥$919
         "Does a SoTP model separating HBM and commodity segments
          produce fair value above, at, or below $919?"
         |
         ├── L1E 1.1: Forward non-GAAP P/E on $919 at FY2027 EPS
         │            appears cheap on AI-infrastructure lens?
         │   ✅B — ~12× fwd NTM non-GAAP P/E vs AI-infra peers 20-35×;
         │          cheap IF HBM thesis holds
         |
         ├── L1E 1.2: SoTP with HBM at 18× P/E + commodity at 10× P/E
         │            produces fair value at or above $919?
         │   ✅C — SoTP base case $790-945†; straddles current price;
         │          HBM revenue Tier C†
         |
         └── L1E 1.3: Bear target $330 requires HBM contracts to fail?
             ✅B — HSBC $330 requires $15-20B annual revenue (FY2023
                    trough level) — directly contradicted by "committed"
                    HBM sold-out language

Verdict tally: 12 ✅ · 3 ⚠️ · 0 ✗ · 1 ⊗

Section VI — The Argument (Voice)

Stories lie, structure doesn't. Micron's story in 2026 is one of the most seductive in the semiconductor universe: a company that barely survived FY2023 ($15.5B revenue, -$5B operating income) is now on a trajectory to do $33.5B in a single quarter at 81% gross margins. The numbers are not wrong. The story is not fake. The question is what the numbers mean.

Bears read them as a cycle peak — the pattern they've seen in 2014, 2018, and 2022. Every DRAM cycle looks like "this time it's different" at the top. The high priests of 30-year memory cycle theory have the historical record on their side. They have buried many short-cycle memory bulls.

But the bear case has a structural problem: it requires you to ignore the contractual structure of HBM. "Sold out through 2026" is not a demand forecast — it is a statement about signed contracts. "Committed" is the word management chose, not "expected" or "anticipated." In commodity memory, revenue visibility is zero beyond the current quarter. In HBM, Micron can see FY2026 and is already taking FY2027 commitments. These are different products by any analytical definition.

The structural blindness that sustains the mispricing is not malice — it's spreadsheet architecture. Sell-side analysts built their MU models before HBM existed as a commercial product. The model template has a DRAM line and a NAND line. It doesn't have a "HBM contracted AI-infrastructure" line. Until Micron forces the issue by filing a segment disclosure, the analysts cannot change the model without breaking their historical comparable series. So they apply a cycle-peak discount to the whole company and wait for the "inevitable" bust.

This is the same cognitive trap NVDA bears fell into for two years: "GPU gaming company at 40× is a bubble." The category error — GPU gaming vs. AI compute — was structural, not factual. NVDA was not hiding the data. The data was in the earnings. The bears just had the wrong framework applied.

For MU, the resolution is the same: multiple quarters of evidence + a formal structural trigger (HBM segment disclosure) that forces model-template change. The question is not IF this happens — the data is too loud — but WHEN. And whether the $919 price already reflects the answer.

It doesn't. At 12× forward non-GAAP P/E, the market is applying a "might be cyclical peak, might be structural" probability-weighted discount. The bull thesis at $1,650+ requires no earnings growth — just multiple expansion from 12× to 18×. That is a pure identification bet, not an earnings bet. The single most important number to watch is not Q3 revenue (though that matters) — it is whether HBM is formally disclosed as a separate segment.

Section VII — Scenario Regime Analysis

Three scenarios are defined by the resolution of the HBM-identity-switch hypothesis:

Why 30/50/20 vs. the 25/50/25 default prior (K.3.3): The HBM "committed" sold-out language shifts bear probability down from 25% to 20% — it is hard to construct a bear path to $550 without either (a) HBM contracts failing, or (b) AI hyperscaler capex cutting >20% in unison. Both are observable and neither has fired. The 5-point shift from bear to bull is conservative given the sold-out language, but preserves the meaningful downside magnitude (-40%) that this cyclical-capital-intensity business carries.

Regime transition logic:

• Bull requires: Q3 ≥$33.5B at ≥81% GM + HBM segment disclosure (T2) OR Samsung HBM4 failure confirmed (T3) + FY2027 HBM committed language (T5)
• Base requires: Q3 in-line; no segment disclosure; GM sustains >75% for two quarters
• Bear requires: Q3 miss (RF1) OR RF2/RF4 (GM <65%, multi-hyperscaler capex cut)

SoTP sanity check (Tier C†):

†All HBM revenue splits Tier C — algebraic inference from blended margins and management disclosures.

Section VIII — Implied Probability Analysis

Market-implied pricing: At $919.53, the market is applying approximately 12× forward non-GAAP P/E to ~$76 NTM EPS (consensus-aggregated†). This is NOT a scenario-probability framework — it is a "cheap on forward earnings if earnings are sustainable" framework. The market is embedding a regime-classification uncertainty discount, not a scenario distribution.

The orphan premium: Current price ($919.53) sits $115-143 above the average sell-side consensus PT ($776-804). The market has begun pricing the identity-switch before sell-side models have adopted it. This orphan premium is consistent with institutional investors having channel-check evidence that HBM contracted revenue is structural; but it is also consistent with momentum overshoot. The distinction resolves at T1 (Q3 FY2026 earnings, June 24).

Probability gap analysis:

Key insight: The MU thesis is primarily a multiple-expansion bet (12× → 15-18× P/E as AI-infrastructure identity is confirmed), not an earnings-growth bet. Flat earnings from current trajectory, re-rated from 12× to 18×, produces approximately $1,368 — a +49% return from current price. The earnings need not grow; they need to be classified differently.

Section IX — Durability Summary

Full analysis in durability_test.md.

Holdability verdict: Selective hold 1-3% (K.3 rule: 17-21 durability = selective hold; no fatal-flag cap override). Not a core compounder — cyclical capital-intensity (Q3) and moat uncertainty (Q2) prevent the 22-25 durability score that unlocks the 3-7% compounder allocation. Exceptional positive note: US-only HBM + CHIPS Act creates a domestic-manufacturing premium the framework doesn't fully capture (exceptional_positive_overrides: no — not yet crystallized into earnings).

Section X — Forward-Looking: Triggers and Red Flags

Full playbooks in triggers_redflags.md.

Trigger watch

Red flag watch

Load-bearing watch item (L1B 1.1 ⚠️/L1C 1.3 ⊗): Samsung HBM4 yield and qualification timeline is the single most volatile node in the tree — it simultaneously drives T3, T1 context, and RF3. Q3 FY2026 earnings call (June 24) is the first resolution window.

Section XI — Long-Term Holdability Verdict

Durability 21/25, 0 fatal flags → K.3 rule: Selective hold 1-3%

Micron is conditionally ownable for a 5-10 year investor who:

1. Accepts that the commodity-memory-cycle risk creates potential -40% drawdowns even in a hold-through strategy
2. Has conviction that HBM represents a structural, not cyclical, inflection in Micron's economics
3. Sizes appropriately for cycle risk (1-3%, NOT the 3-7% compounder allocation)

NOT recommended as a core compounder position alongside Costco, ISRG, or VEEV. The cyclical capital-intensity (Q3: 3/5) and moat uncertainty (Q2: 3/5) prevent the 22-25 High durability rating that unlocks the compounder allocation.

Correlated-exposure paragraph (K.3.4): MU carries ai-capex-high cycle exposure — first-order correlation to AI hyperscaler capex. If a holder simultaneously holds NVDA + TSM + AVGO + AJNMY at meaningful weights, the combined ai-capex-high bucket may exceed 15% of portfolio. Per K.4 correlated-factor rule:

• If NVDA+TSM+AVGO+AJNMY already >12% combined → drop MU hard cap from 3% to 1-2%
• The bear scenario that fires RF4 (multi-hyperscaler capex cut) hits the entire cluster simultaneously — size so a concurrent -40% drawdown across the cluster does not impair the broader portfolio

Position state as of June 11, 2026:

Section XII — Investment Scorecard

K.3.5 Format B (15-question long-term hold scorecard). Weight tiers: Critical 5× · Load-bearing 3× · Important 2× · Confirming 1×. Max = 39.

Scorecard

K.3.5 weighted score: 34.0 / 39 = 87% → High-conviction (≥85%)

Note: xii_score in INDEX_META = 87%, consistent with this computation.

Scorecard summary

Micron scores 87% on the K.3.5 15-question Format B scorecard — firmly in the high-conviction band (≥85%). The thesis is structurally complete: a well-evidenced H-0, documented mispricing mechanism, confirmed business model persistence, and clean fatal-flag record. The three deductions are:

1. Supply discipline uncertainty (Q4 ⚠️): Samsung's HBM4 qualification timeline is live uncertainty — reduces the load-bearing branch without invalidating the thesis.
2. Competitive position watch (Q5 ⚠️): HBM4 yield ramp is an ⊗ node; Q3 FY2026 earnings (June 24) is the first resolution window.
3. Correlated-exposure constraint (Q15 ⚠️): ai-capex-high cluster management — position cap requires adjustment if total cluster exceeds 12%.

The 22-point gap between H-0 confidence (65%) and xii_score (87%) is the distance the thesis must travel from structural identification to market confirmation. Primary catalysts are reclassification events — HBM segment disclosure, two consecutive >75% GM quarters, and/or Samsung confirmed absent at NVDA Rubin — not earnings beats.

Final verdict: Hold-with-sizing

Initiate 1% starter position (watch mode before Q3 June 24, 2026). Scale to 2-3% post-Q3 FY2026 confirmation (revenue ≥$33.5B at ≥80% GM). Maintain 3% hard cap (selective-hold band, durability 21/25). Exit on RF1 (Q3 miss ≥5%), RF2 (GM <65%), or RF4 (multi-hyperscaler capex cut).

2-minute pitch

Micron is the cheapest AI-infrastructure asset in the semiconductor universe — not because the market is wrong about risk, but because it is applying the wrong framework. At 12× forward P/E, the market treats HBM as peak-cycle commodity DRAM. The evidence says otherwise: HBM is sold out under multi-quarter "committed" contracts at 10-20× ASP premium to commodity DDR5, qualified at the chip-package level with 6-12 month switching costs, generating $6.9B FCF in a single quarter. The bear case ($550) requires HBM contracts to fail — those contracts have been publicly confirmed as "committed." The bull case ($1,650) requires no earnings growth — just multiple expansion from 12× to 18× as the market reclassifies HBM from commodity to AI-infrastructure. Q3 FY2026 on June 24 is the first major evidence event. If it confirms (≥$33.5B at ≥80% GM), this is a 2-3% selective-hold position with 30% probability of +79% and 20% probability of -40%. The asymmetry ratio of 2.96× is the position.

Risk types applicable (K.4)

When NOT to buy (K.5 anti-pattern check)

Do NOT add to MU in these conditions:

1. RF1 fires (Q3 revenue ≥5% below $33.5B): Exit starter immediately; do not average down into a thesis failure
2. RF2 fires (GM <65% any quarter): Exit all; re-evaluate thesis from scratch
3. RF4 fires (two or more hyperscalers cut capex >20%): Exit all; the ai-capex-high cluster collapses simultaneously
4. Samsung confirmed primary at NVDA Rubin (RF3): Cap at 1%; the competitive differentiation thesis has failed
5. ai-capex-high cluster already >15% of portfolio: MU is the marginal addition pushing correlated exposure into unsafe territory; skip until cluster is trimmed
6. If sizing it as a compounder: MU is not COST or ISRG. Durability 21/25 and cyclical capital intensity classify it as selective hold (1-3%), not compounder (3-7%). Sizing above 3% requires a durability score of 22-25 that MU does not currently have.