Micron

Phase A — Understand the Business

Lens 1 · Company Overview

Micron is the only US-based — and one of three globally relevant — manufacturers of both DRAM and NAND flash, a true IDM that designs, fabricates, and (for HBM) packages its own product. FQ2'26 revenue split DRAM $18.8B (79%) / NAND $5.0B (21%). The franchise has re-centered on HBM — vertically stacked DRAM with a logic base die, sold as the memory tier bolted onto AI accelerators. Customers: NVIDIA is the anchor (HBM3E into Blackwell, HBM4 into Vera Rubin), plus AMD, hyperscalers, legacy mobile/auto/industrial. Competitors: SK Hynix, Samsung (DRAM/HBM); Samsung, SK Hynix, Kioxia/Sandisk (NAND). The defining 2026 shift: HBM moved from quarterly spot pricing to multi-year (3–5yr) fixed price-and-volume contracts — Micron's entire CY2026 HBM supply is contracted. That is what converts a memory-cyclical into something closer to a contracted-capacity utility.

Lens 2 · Supply Chain

Micron sits in the memory rail: Equipment (ASML EUV, Applied Materials, Lam, KLA) → Micron fabs → stack + base-die → TSMC CoWoS interposer → NVIDIA/AMD package → hyperscaler. Two chokepoints outside Micron's control gate its output: TSMC CoWoS (sev 9, NVIDIA holds ~60%) and the HBM4 base-die (increasingly foundry-made). Upstream, hostage to ASML's EUV monopoly. Downstream concentration intensifying: 2026 HBM industry-wide pre-booked by NVIDIA + OpenAI. Micron's own new capacity (Idaho ID1, Taiwan) adds no wafer output until H2 2027 — so through 2026 it sells a fixed pie at rising prices.

Lens 3 · Competitive Advantages

The moat is structural oligopoly, not proprietary tech. Three suppliers control HBM — SK Hynix ~62%, Micron now #2 (~21%) ahead of Samsung (~17%). Overtaking Samsung into #2 is the year's competitive headline and a genuine share win — but the cross-company frame matters: Micron is the fast-closing #2 in a 3-player cartel, not the leader. Real edges: HBM3E power efficiency (~30% lower, won NVIDIA design-ins), IDM capture of stacking/packaging margin, capital intensity ($25B+/yr) + 12–18mo capacity-to-output lag making supply structurally slow → ~68–81% gross margins sustainable through 2026. Bargaining power at cycle-high: customers get only 50–67% of demanded HBM, signing take-or-pay-style contracts. Vulnerability: still a commodity at the core; all three are expanding — the 2027–28 HBM4 ramp is the structural-tightness expiry.

Lens 4 · Segments

Phase B — Measure Performance

Lens 5 · Earnings Result (FQ2'26, ~Mar 18 2026)

A blowout. Revenue $23.86B, +196% YoY / +75% sequential; non-GAAP EPS $12.20 vs ~$8.79 est (+39%). GAAP GM 74.4%, op margin 67.6%; OCF $11.90B, net capex $5.0B → adj FCF $6.9B; cash & investments $16.7B. The guide is the story: FQ3 record $33.5B ±0.75, GM ~81%, EPS ~$19.15. Dividend +30% to $0.15/qtr. Stock dipped after-hours despite the beat — the market balked at >$25B FY26 capex (the memory reflex: capex = future oversupply).

Lens 6 · Earnings Calls (sentiment trend)

Monotonic escalation from "recovery" to "structural shortage." FQ1'26: HBM TAM $35B→$100B pulled forward 2 years. FQ2'26: "entire CY2026 HBM supply" contracted; DRAM ASPs +mid-60s% sequential, NAND +high-70s%. The move to multi-year contracts is presented as a permanent change to revenue volatility. Tone risk: cycle-peak confidence is historically when memory teams are most wrong about durability. FQ3 reports June 24, 2026 — "the print that decides the AI memory trade."

Lens 7 · Comps

Lens 8 · Stock-Price Catalysts

What moves MU: (1) memory-cycle DRAM/NAND ASP direction (dominant); (2) HBM allocation/qualification news (now co-equal); (3) next-quarter guidance > the print. 2022–23 downturn → $5.83B FY23 net loss, $1.83B inventory write-downs; 2024–26 up-cycle the mirror (+84% in the month into June 2026). Signature recent catalyst: Jensen Huang certifying Samsung+SK Hynix+Micron for Vera Rubin HBM4 (June 5, 2026) — qualification is the catalyst in a sold-out market. Counter-intuitively, beats sell off on capex.

Phase C — Judge People & Books

Lens 9 · Management

Sanjay Mehrotra, CEO since 2017, co-founder of SanDisk — a genuine memory operator-founder. Navigated the 2018–19 and 2022–23 down-cycles; positioned the HBM3E power-efficiency win that opened NVIDIA. Net worth ≥$968M, mostly Micron equity on multi-year metrics. Capital allocation textbook-disciplined-aggressive: reinvest > balance sheet > return; FY26 capex >$25B, FY27 construction +>$10B YoY; dividend deliberately small. Mild red flag — CEO selling: ~$36M sold late May 2026 near all-time highs under a Jan-2026 10b5-1 plan (procedurally clean, but worth noting after an 84% run; per-share figures internally inconsistent in sources → treat as n/a pending Form 4).

Lens 10 · Forensic Red Flags

Memory is an accounting-cyclical business; the items cut favorably now, which is the risk. (1) Write-down/recovery asymmetry — $1.83B FY23 write-downs now sell through at full price, flattering current GM; part of 74–81% margin is cycle-mechanics. (2) Capex/depreciation timing — record margins earned on an under-depreciated base; FY26's $25B+ converts to depreciation in 2027–28, a built-in margin headwind. (3) HBM demand pull-forward / double-ordering — biggest concern: customers getting 50–67% of demand over-order across all three suppliers; multi-year contracts partly defend, but if AI capex disappoints, take-or-pay becomes a dispute surface. (4) Customer concentration on NVIDIA/hyperscalers (exact %: n/a). (5) China/policy tail (2023 CAC ban precedent; current China %: n/a).

Phase D — Project & Stress-Test

Lens 11 · Forward Projection (EPS, next 3 fiscal years)

Bottom-up; Micron FY ends late Aug (FY26 = Sep'25–Aug'26). The wide dispersion IS the memory signature.

Lens 12 · Bull vs Bear

Bull. Micron is the #2 supplier in a 3-player HBM cartel selling a fixed pie at rising prices, and the multi-year fixed-price contracts have — for the first time — taken the volatility out of the model: 2026 is contracted, data-center is 31% and rising, and the power-efficiency design win keeps it in every NVIDIA generation. At ~9.8× forward it's priced for a cyclical roll-over the contracts are designed to prevent — so the asymmetry is in the multiple re-rating as the market accepts memory has structurally changed. Bear (impairment vectors). (1) It's still a commodity at the core; all three players are spending $200B+ combined, and the 2027 HBM4 ramp is the dated expiry of structural tightness. (2) Today's 80% margins are partly cycle-mechanics — written-down inventory selling through + an under-depreciated asset base — that reverse together when the cycle turns. (3) Entry is at all-time highs after +84% in a month. Pre-mortem (2027): HBM4 capacity from all three lands together, AI-capex growth slows from a hyperscaler digestion, ASPs roll 30%, and the take-or-pay contracts become a litigation/renegotiation surface rather than a floor. Margins compress from 80% to 40% in three quarters — the memory bust the single-digit forward multiple was always pricing. Contrarian view: The market refuses to capitalize peak HBM earnings (hence ~10× forward) — but it's also not pricing the genuine possibility that multi-year contracts have permanently lowered the trough. The variant perception is "this cycle's trough is higher than every prior trough," and the contracts are the evidence.

Lens 13 · Devil's Advocate

"Sold out" and "multi-year contracts" are exactly what every memory peak says before the bust. Double-ordering inflates the backlog: buyers getting 50–67% of demand order 150% across three suppliers to secure allocation — so the "sold out" book is partly phantom. The CEO is selling into all-time highs. The 80% gross margin is the most mean-reverting number in the S&P. If AI capex disappoints 20–30%, FY28 EPS goes from ~$48 to ~$10 and a "cheap" 10× forward becomes 30× collapsing earnings. The single impairment scenario: synchronized HBM4 capacity + a capex air-pocket → the classic memory glut, and Micron — the #3-becoming-#2, not the leader — gets the worst of the allocation shake-out.

Lens 14 · Management Questions

1. What is HBM as a clean % of revenue, and why stop disclosing it? 2. Are the multi-year HBM contracts take-or-pay on volume, or price-only — and what's the enforcement if a customer cancels? 3. How much of the "sold out" book is double-ordering across suppliers? 4. As FY26's $25B+ capex depreciates in 2027–28, what's the gross-margin headwind at flat ASPs? 5. What's NVIDIA + top-3 customer concentration? 6. What HBM4 16-Hi share do you target vs SK Hynix/Samsung? 7. China revenue exposure post-CAC-ban? 8. At what HBM ASP does the FY27 capacity become an oversupply risk? 9. Why is the CEO selling near highs? 10. What's the trough-EPS you'd defend if AI capex paused? 11. NAND vs DRAM capital allocation priority? 12. Idaho/Taiwan capacity timing and its effect on 2027 supply?