Micron Technology, Inc. (NASDAQ: MU) — A Trillion-Dollar Bet That the Memory Cycle Is Dead

Report date: 2026-06-09 Price at writing: $935.89 · Market cap: ~$1.06T · Enterprise value: ~$970B (net cash) Fiscal year: ends last Thursday of August (FY2025 ended 2025-08-28; latest reported quarter Q2 FY2026 ended 2026-02-26)

With the single, clearly-labeled exception of the “Claude’s Take” block immediately below, this article contains no buy/sell recommendation and no price target; the body discusses valuation only as embedded expectations and scenarios.

⚡ Claude’s Take

This block is the author’s own independent opinion and general information only. It is not investment advice. Everything below it is position-free analysis.

Verdict: HOLD / do-not-chase at $936. Not a short. Would become a buyer only on a deep cyclical reset into roughly the $300–450 zone. Conviction: medium.

Tag: “A trillion-dollar bet that the cycle is dead.”

Micron is, today, a genuinely better business than at any prior memory peak — net cash, technology leadership at the leading edge (1-gamma EUV DRAM, G9 NAND), the #2 position in HBM, and a sold-out, contracted 2026 order book. The AI/HBM demand shock is real, the cash it is throwing off ($19.0B net income in one half-year) is real, and the “die penalty” (HBM consumes ~3× the wafer area per bit, draining commodity-DRAM supply) is a real, quantifiable structural tightener that has made this up-cycle longer and steeper than any before it. That is the bull case, and it is not stupid.

But the price has now fully discounted the proposition that memory’s 40-year boom-bust pattern has been abolished. At $936 the stock trades at the 96th percentile of its own 10-year valuation history on a composite basis (99th on price/book and price/sales), after rising roughly ten-fold off its 52-week low of $103 and crossing $1 trillion in May 2026 on its best month since 1985. The entire 2026 surge is price/ASP-driven on mid-single-digit bit growth — the textbook signature of a cycle top, where operating leverage flatters margins on the way up and savages them on the way down (this same company lost $5.8 billion in FY2023). Management is plowing peak cash flow into a ~$200B, >$25B-a-year capacity build at the same time as SK Hynix and Samsung — the canonical Marathon capital-cycle setup that seeds the next glut — while rewarding executives on bit-growth and market-share metrics with no ROIC gate, resuming buybacks at all-time highs after repurchasing nothing at the trough, and selling stock personally the entire way from $130 to $980 with zero open-market buying. What the market is mispricing is persistence: it pays for the cycle to be permanent; history says it is merely long. The forward P/E of ~8.6× looks cheap precisely because it is a low multiple on peak earnings — the oldest trap in cyclicals.

I am not short it: shorting a structurally-improved, sold-out, momentum-saturated name at a real demand inflection is how careers end, and HBM could keep 2026–27 tighter for longer than bears expect. But I would not initiate length here. What flips me bullish: a genuine cyclical reset — a 50%+ drawdown to ~3–4.5× prospective book — combined with evidence (sequential ASPs still firm, SCA coverage expanding) that the structural HBM floor is holding; that is a generational entry. What flips me bearish (toward an actual short): the first quarter of sequential DRAM ASP decline or any major competitor announcing aggressive greenfield capacity — either one breaks the “this time is different” narrative, and at 14× book the air underneath is measured in hundreds of dollars.

1. Executive Summary

Micron Technology is one of three scaled producers of DRAM and one of a handful in NAND flash — the memory and storage layer of essentially every computer, phone, car, and AI accelerator on earth. It is a commodity-cyclical business of the most extreme kind, and it is currently living through the most violent up-cycle in its history, driven by artificial-intelligence demand for high-bandwidth memory (HBM) and the broad DRAM shortage that HBM’s silicon-hungry economics have created.

The numbers are staggering and, importantly, filed and real. Revenue went from $15.5B in FY2023 (a year in which Micron lost $5.8 billion) to $25.1B (FY2024) to $37.4B (FY2025), and then the first half of FY2026 alone produced $37.5B of revenue and $19.0B of net income — H1 FY2026 by itself exceeded all of FY2025. The February-2026 quarter printed a 74.4% gross margin, a 67.6% operating margin, and $12.07 of diluted EPS. The balance sheet flipped from net debt to roughly $6.5B net cash. The stock responded by rising ~10× in twelve months and crossing a $1 trillion market capitalization.

The investment debate is not about whether the present is good — it plainly is — but about persistence and price. Two facts frame everything:

1. The surge is price, not volume. DRAM bit shipments rose only mid-single-digits sequentially in Q2 FY2026 while DRAM average selling prices rose ~mid-60% sequentially (and ~110% year-over-year). Memory’s defining mechanism is operating leverage on ASP: nearly all of a price increment falls to gross profit on the way up — and the same leverage produced negative gross margins and a multi-billion-dollar loss as recently as FY2023.

2. The whole industry is building into the peak. Micron, SK Hynix, and Samsung are all expanding leading-edge DRAM and HBM capacity simultaneously. Micron’s own capex is guided above $25B in FY2026 (from $8.4B in FY2024), inside a ~$200B U.S. fab program. In the Marathon “Capital Returns” framework, synchronized, pro-cyclical capacity addition at peak returns is the precise signal that the next down-leg is being financed now.

On business quality: Micron is an excellent operator of a structurally mediocre business. Through a full cycle, memory ROIC has historically hovered around or below the cost of capital, with excess returns appearing only at peaks and GAAP losses at troughs. Applying Greenwald’s framework, Micron has a contestable edge — economies-of-scale-plus-customer-captivity that must be re-won each technology generation — not a durable wide moat. The market-share-stability test fails (DRAM shares swing 4+ points per quarter); the through-cycle ROIC test fails. Micron is, moreover, the #3 DRAM player (~22%) behind Samsung (~39%) and SK Hynix (~29%), and the #2 HBM player (~21%) behind SK Hynix (~62%) — it carries a scale disadvantage versus Samsung, not an advantage.

On capital allocation: a split verdict. Financially conservative and counter-cyclical (deleveraging into strength, ~$6.5B net cash, never cut the dividend, no empire-building M&A) but strategically aggressive at a dangerous point in the cycle (tripling capex into the peak; resuming buybacks at all-time highs after $0 at the trough; an incentive plan that rewards bit growth and share, not returns on capital). Insiders have sold the entire move and own less than 1%.

On valuation: The stock is cheap on peak forward earnings (~8.6× forward P/E) and extreme on normalized and book measures (≈14.6× book; 96th–99th percentile of its own history). That divergence is the cyclical-top fingerprint. The embedded expectation is that AI/HBM has structurally raised memory’s through-cycle earnings power enough to justify a trillion-dollar valuation — a bet on persistence that the bull and bear both ultimately argue about, agreeing on today’s facts and disagreeing only on their durability.

The remainder of this memo argues each of these points from the evidence.

2. Business Overview

What Micron does. Micron designs and manufactures semiconductor memory and storage. Its two product families are:

• DRAM (dynamic random-access memory) — the high-speed working memory that sits next to a CPU or GPU. DRAM is ~76–79% of Micron’s revenue and the overwhelming driver of its profits. Within DRAM, the strategically critical product is HBM (high-bandwidth memory) — vertically-stacked DRAM dies bonded together and sold at a large premium for AI accelerators (Nvidia, AMD). DRAM also includes standard DDR5 server/PC memory, LPDDR (low-power) mobile and increasingly server memory, graphics memory (GDDR), and CXL-attached memory.
• NAND flash — non-volatile storage (solid-state drives, embedded storage for phones and cars, memory cards). NAND is ~20% of revenue and a structurally weaker, more fragmented, lower-margin franchise.

It sells under the Micron brand (OEM/enterprise) and the Crucial brand (consumer/retail). [FACT — FY2025 10-K, Item 1]

How it makes money. Micron earns a spread between the market-clearing price of a memory bit and its cost to manufacture that bit. Cost-per-bit falls over time through process-node “shrinks” (packing more bits per wafer); price-per-bit is set by the supply/demand balance of a near-commodity market. The business is therefore price-takers competing on cost, and profitability is dominated by the cycle: when supply is tight, ASPs run far above marginal cost and margins explode; when supply is loose, ASPs fall toward (or below) cash cost and the industry loses money. There is essentially no recurring revenue in the subscription sense — Micron must re-sell its output every quarter at whatever the market will bear — although long-term agreements (LTAs) and, newly, multi-year “strategic customer agreements” (SCAs) are beginning to add order-book visibility, particularly in HBM. [FACT/INTERPRETATION — FY2025 10-K; Q2 FY2026 earnings call, 2026-03-18]

Segmentation. Effective Q4 FY2025 (September 2025), Micron reorganized reporting into four market-facing business units (prior periods recast). The reorganization deliberately reframes the company around AI/data-center:

[FACT — FY2025 10-K Note 27; Q2 FY2026 10-Q Note 17 / MD&A]

The story this table tells: CMBU — the HBM-plus-hyperscale franchise — went from $1.87B (FY2023) to $13.52B (FY2025) and then booked $7.75B in the single February-2026 quarter (+163% YoY), annualizing above $30B. Data center (CMBU + CDBU) reached ~56% of total company revenue in FY2025, up from a low-30s% historically — the AI mix shift in one figure. Note also that every BU now earns a 60–76% segment operating margin: the pricing wave is broad-based, not HBM-only, which is itself the tell that this is a cyclical ASP phenomenon, not a product-specific structural win.

Customers and end-markets. Micron serves data center (the dominant and fastest-growing end market), PCs, smartphones, automotive/industrial, networking, and consumer. Customer concentration is material and rising: roughly half of total revenue comes from the top ten customers, and a single customer was ~13% of H1 FY2026 revenue (primarily CMBU — i.e., a large HBM hyperscaler/GPU customer). [FACT — FY2025 10-K; Q2 FY2026 10-Q Note 14]

Manufacturing footprint. Micron fabricates in the U.S. (Idaho, Virginia), Taiwan, Japan, and Singapore, with assembly/test in Asia and a large U.S. expansion underway (Idaho, New York). It is the only U.S.-headquartered scaled memory maker, which has become a strategic and political asset (CHIPS Act funding, U.S. data-center supply security). It is ~53,000 employees, headquartered in Boise, Idaho, founded 1978, public since 1984. [FACT — FY2025 10-K]

Verdict (Business Overview): A globally essential product, a clean and understandable business model, and a powerful AI-driven mix shift toward data center — wrapped around a commodity-cyclical core whose revenue must be re-won every quarter at market-clearing prices. Easy to understand; hard to own through a cycle.

3. Industry Dynamics

Structure: a consolidated DRAM oligopoly atop a more fragmented NAND market. DRAM is one of the most consolidated industries in technology — three players (Samsung, SK Hynix, Micron) hold ~90%+ of the market. Q1-2026 DRAM share: Samsung ~38.6%, SK Hynix ~28.8%, Micron ~22.4%, with the remainder split among Taiwan’s mature-node makers (Nanya, Winbond) and China’s CXMT. NAND is meaningfully more fragmented — Samsung, SK Hynix/Solidigm, Kioxia, SanDisk (formerly Western Digital), Micron, plus a fast-rising YMTC — which is precisely why NAND margins and returns lag DRAM through the cycle. [FACT — Omdia/TechTimes 2026-06-09; TrendForce 2026-02-26; FY2025 10-K Item 1 competitive conditions]

Why memory is the canonical commodity cycle. The product is a fungible bit sold at a market-clearing ASP; the cost base is dominated by fixed depreciation on multi-billion-dollar fabs. That combination produces extreme operating leverage in both directions: when ASPs rise, almost the entire increment is profit; when they fall, a fixed depreciation load drives the industry into losses. Demand is price-inelastic in the short run, capex is lumpy and pro-cyclical, and node transitions add bits in step-functions. The result is a 40-year history of boom-bust. The proof point sits two years back: in FY2023 SK Hynix posted a negative gross margin and a ₩7.7T operating loss, and Micron lost $5.8 billion — the same companies now printing 70%+ gross margins. [FACT — SK Hynix FY2023 reported results; Micron FY2023 10-K; INTERPRETATION on mechanism]

The pattern is not anecdotal — it is the industry’s entire history. Micron’s own diluted EPS over the last five years traces the cycle precisely: $5.14 (FY2021) → $7.75 (FY2022 peak) → −$5.34 (FY2023 trough) → $0.70 (FY2024) → $7.59 (FY2025) → $16.68 in just H1 FY2026. A business that swings from +$7.75 to −$5.34 to +$16.68 of EPS in five years is, by construction, not a stable earner; it is a leveraged bet on a commodity price. The 2018–2019 and 2022–2023 down-cycles each compressed industry margins from peak to loss inside ~18 months. The relevant question for an investor at the 2026 peak is not whether a down-cycle comes, but when and how deep — and whether HBM contracting genuinely raises the trough floor this time. [FACT — Micron 10-K EPS history; INTERPRETATION]

The structural improvement that is genuine: consolidation from four to three. One reason to take the “longer cycle” case seriously is that DRAM consolidated from four meaningful players to three over the last decade (the exit/absorption of smaller players; Micron’s own Elpida/Inotera consolidation). Fewer rational actors makes supply discipline easier to sustain, and the FY2023 down-cycle was indeed shorter and the recovery sharper than prior gluts — partial evidence that the oligopoly is behaving more rationally. But “more rational” is not “non-cyclical,” and the entry of state-backed Chinese capacity threatens to reverse the consolidation benefit at the commodity tail. [INTERPRETATION — grounded in share-history and SK Hynix peer report]

Market size and the current boom. Total memory revenue is in a once-in-a-decade up-cycle: industry trackers peg total memory at ~$365B (2025) rising toward ~$552B (2026) and a forecast peak around ~$843B (2027); DRAM alone ~$166B (2025) → ~$404B (2026). Q1-2026 conventional DRAM contract prices rose a record ~90–95% quarter-over-quarter; NAND +55–60%; Q2-2026 forecast DRAM +58–63%, NAND +70–75% QoQ. Cloud service providers are locking supply via LTAs, and meaningful new wafer capacity is “unlikely before late 2027 or 2028.” [FACT — TrendForce 2026-02-02 / 2026-02-26; Tom’s Hardware 2026]

The capital cycle (Marathon lens). This is the analytically decisive frame. Memory is almost a caricature of the supply-side capital cycle: record profits attract capacity → capacity creates oversupply → oversupply collapses pricing → losses force rationalization → repeat. The 2026 question is whether AI/HBM has changed this or merely stretched it.

Evidence the cycle is genuinely longer this time (the strongest “this time is different” case ever assembled):

• The HBM “die penalty.” One HBM bit consumes ~3× the wafer area of a standard DDR5 bit (larger die for through-silicon vias and logic, plus stack-yield loss). Every wafer diverted to HBM therefore removes ~3 wafers of commodity DRAM from the pool — tightening all of DRAM. This is a real, quantifiable supply sink and the mechanism behind the +90–95% conventional-DRAM price move. [FACT — corroborated by price action; SK Hynix peer report]
• Capex is rising but restrained relative to revenue, and skewed to upgrades. Industry DRAM capex ~$53.7B (2025) → ~$61.3B (2026), only ~14% YoY, shifting toward process upgrades, stacking, and HBM rather than raw greenfield bit capacity; trackers judge the 2026 capex adds will have “minimal impact on bit-supply growth.” 2026 bit-supply growth (DRAM ~+16–20%, NAND ~+17%) is running below the 20–30% historical norm. [FACT — TrendForce 2025-11-13]
• HBM is contracted and sold out. Micron has “completed agreements on price and volume for our entire calendar 2026 HBM supply, including HBM4,” and signed its first 5-year customer agreement. [FACT — Q1/Q2 FY2026 calls]

Why the skeptic should not be disarmed (the red flags are flashing):

• High returns are unambiguously attracting capital. All three players are expanding leading-edge capacity simultaneously — Micron’s ~$200B U.S. program (>$25B FY2026 capex), SK Hynix doubling DRAM toward ~1M wafers/month by ~2030, Samsung pushing HBM capacity hard. This synchronized, pro-cyclical build is the textbook prisoner’s-dilemma that has broken cooperation before every prior cycle. [FACT — TrendForce 2026-03-19; TheStreet 2026]
• The die penalty is a one-time level shift, not a perpetual tightening force. Once HBM mix stabilizes, normal supply growth resumes — and today’s peak-margin capex lands as supply in 2027–28, the period of maximum cycle risk. Trackers themselves warn that “if HBM demand weakens, the shift back to conventional DRAM production could create a supply glut” by mid-2026. [FACT — TrendForce/Neumonda 2026]
• The equilibrium is behavioral, and China is an indifferent agent. Supply discipline is a choice, not a law. CXMT (DRAM) and YMTC (NAND) are state-backed Chinese producers commoditizing the legacy tail with capital that does not require an ROIC — a structural wildcard the incumbents do not control. [FACT/INTERPRETATION]

The China wildcard deserves its own paragraph. CXMT (DRAM) and YMTC (NAND) are the most important structural variable the bulls tend to underweight. Backed by state capital that does not demand an ROIC, these producers have moved from negligible to meaningful share in legacy DRAM (DDR4) and mainstream NAND within a few years. Their effect is asymmetric: they cannot yet compete at the HBM/leading-edge DRAM frontier (export controls on advanced equipment constrain them), so they do not threaten Micron’s current profit pool — but they relentlessly commoditize the tail, pulling down legacy-node pricing and forcing the incumbents up the technology curve. In a downturn, Chinese capacity indifferent to losses is precisely the kind of actor that turns a normal glut into a prolonged one. The 2023 CAC ban on Micron (barring its products from Chinese “critical information infrastructure”) is the demand-side mirror of the same geopolitical fault line. For a through-cycle view, China is the single biggest reason to doubt that “consolidation to three rational players” durably holds. [FACT/INTERPRETATION — FY2025 10-K competitive conditions; CSIS 2023]

Regulatory/geopolitical layer. Memory is now a strategic industry: U.S. CHIPS Act subsidies and export controls, China’s 2023 cybersecurity ban on Micron’s products in “critical information infrastructure,” tariffs, and the OECD Pillar Two 15% minimum tax (raising Micron’s effective tax rate). These are simultaneously a tailwind (subsidies, U.S.-supply preference) and a risk (China demand, tariffs, clawbacks). [FACT — FY2025 10-K; Q2 FY2026 10-Q Note 15]

Verdict (Industry Dynamics): structurally good at the leading edge, mediocre-to-poor at the commodity tail — and not safe at any point in the cycle. The 3-player DRAM oligopoly, the HBM die penalty, contracted LTAs, and disciplined-relative-to-revenue capex make the disciplined-supply thesis the most credible it has ever been, and it is materially extending this up-cycle. But it remains a behavioral equilibrium sitting at a violent peak, with all three players building into 2027–28 and China commoditizing the tail. The honest read: a structurally improved but still-cyclical industry. The improvement is real; mean reversion is not repealed.

4. Competitive Position

Greenwald moat type: economies-of-scale + customer-captivity — but the contestable variety, not a durable wide moat. At the commodity-die layer there is no firm-level moat: Micron is a price-taker in an oligopoly where collective supply discipline, not individual differentiation, drives returns. Micron’s genuine firm-specific edges are two, and both have a half-life measured in a single technology generation.

1. Technology leadership — real now, a lead not a moat. Micron is ramping industry-leading 1-gamma (1γ) DRAM — among the first to mass-produce DRAM with EUV lithography at scale, ramping faster to mature yields than any prior node, and on track to be the majority of DRAM bits by mid-2026 — and G9 NAND (PCIe Gen6 data-center SSDs, first to market on Nvidia’s reference platform). This delivers a power-efficiency edge that matters in HBM3E, where data-center power budgets are binding. [FACT — Q2 FY2026 prepared remarks 2026-03-18; Motley Fool transcript 2026-03-18] But all three players sit within roughly one node of each other; SK Hynix and Samsung have their own EUV roadmaps; and a memory technology lead historically lasts a generation, then converges. This is a sprint that resets every node — not a structural barrier.

2. Customer captivity via qualification lock-in — real, but generation-bound. HBM is co-designed and qualified into a specific GPU architecture over multiple quarters; automotive and industrial memory carry multi-year qualification cycles. This is genuine captivity for a product generation — and Micron has parlayed it into record data-center SSD share gains for four consecutive years and HBM qualification at both Nvidia and AMD. But the captivity must be re-won at each HBM generation (HBM3E → HBM4 → HBM4E); it does not compound into a durable lock the way a software ecosystem would.

Scale is a disadvantage for Micron, not an advantage. Greenwald’s scale advantages accrue to the largest player. Micron is #3 in DRAM at ~22%, behind Samsung (~39%) and SK Hynix (~29%), and #2 in HBM (~21%) behind SK Hynix (~62%). Micron’s own 10-K concedes competitors “may have a larger market share and greater resources to invest in technology … and withstand downturns.” Samsung’s vastly larger memory scale and conglomerate balance sheet make Micron the cost-disadvantaged incumbent. [FACT — FY2025 10-K competitive conditions; Omdia/Counterpoint share data 2026]

The two decisive moat tests both fail:

• Market-share-stability test — FAILS. Greenwald’s single most reliable moat diagnostic is share stability. DRAM share swings 4+ points per quarter (SK Hynix fell 32.9%→28.8% while Samsung rose 36.5%→38.6% in one quarter; HBM leadership swung from ~62% toward ~50% across a year). Shares that move >5 points/year are the signature of no durable barrier. [FACT — Omdia/TrendForce 2026]
• Through-cycle ROIC test — FAILS. Micron earns spectacular returns at the peak but posts GAAP losses at the trough (FY2023). Excess returns appear only at peaks; through-cycle memory ROIC has historically hovered around or below the cost of capital. A business that earns its cost of capital only at the top of the cycle does not possess a moat. [INTERPRETATION grounded in FY2023 results and SK Hynix peer history]

Customer concentration is the model’s defining fragility. As HBM concentrates demand into a handful of hyperscaler/GPU budgets (Nvidia above all), the source of super-profits is also the single largest point of failure — illustrated by the June-2026 tremor when Broadcom’s soft AI guide knocked the whole memory complex (Micron −7% in a day). [FACT — CNBC 2026-06-04]

The cost-curve reality versus Samsung. A genuine moat in a commodity would show up as a durable cost advantage — the lowest cost producer can stay profitable when prices fall to a level that bankrupts higher-cost rivals. Micron has closed much of the historical cost gap through node leadership (1γ DRAM ramping faster than prior nodes, with low-single-digit DRAM cost-down in FY2025), and at the leading edge is arguably at or near cost parity with SK Hynix. But Samsung’s sheer scale — the largest memory capacity base, a captive equipment/materials ecosystem, and a conglomerate balance sheet that can absorb losses indefinitely — means Micron is not the structurally lowest-cost producer across the cycle. In the last two down-cycles it was Samsung, not Micron, that could most comfortably “print through” the trough. Micron’s cost position is competitive at the top of the cycle; it is the most exposed of the three at the bottom. That asymmetry — fine when prices are high, dangerous when they are low — is the opposite of a cost moat. [INTERPRETATION — FY2025 10-K cost commentary; SK Hynix peer report; competitive-conditions disclosure]

Verdict (Competitive Position): a genuine but contestable, finite-duration lead — “best operator in a contestable oligopoly,” not a durable moat. Micron’s 1γ DRAM, G9 NAND, and HBM3E power efficiency are real and currently valuable, and qualification lock-in provides per-generation captivity. But it is the #3 DRAM / #2 HBM player with a scale disadvantage to Samsung; both moat tests fail; and nothing yet proves HBM has permanently changed memory’s poor through-cycle economics. Be direct: this is a high-quality cyclical at a peak, not a wide-moat compounder.

5. Growth History and Forward Opportunities

The historical trajectory is real, steep, and almost entirely organic. Memory growth comes from node transitions (more bits per wafer) and pricing, not acquisitions. Revenue: FY2023 $15.5B → FY2024 $25.1B → FY2025 $37.4B → H1 FY2026 $37.5B (already exceeding all of FY2025 in two quarters). DRAM revenue: $10.98B (FY23) → $17.60B (FY24) → $28.58B (FY25); NAND: $4.21B → $7.23B → $8.50B. The only inorganic supply addition is the small Powerchip Tongluo (Taiwan) fab acquisition, which converts logic capacity to DRAM but adds no meaningful bits before FY2028. [FACT — FY2025 10-K Item 1; Q2 FY2026 10-Q]

The composition of recent growth is the analytical crux: it is price, not volume.

• Q2 FY2026 vs Q1: total revenue +75%; DRAM ASP +~mid-60% with bits +mid-single-digit; NAND ASP +~high-70% with bits +low-single-digit. [FACT — Q2 FY2026 10-Q MD&A]
• Q2 FY2026 vs Q2 FY2025: total +196%; DRAM ASP +~110% (bits +~mid-40%); NAND ASP +~slightly-over-100%. [FACT — Q2 FY2026 10-Q MD&A]
• Gross margin trajectory: ~41% (FY2025) → 56% (Q1 FY26) → 74.4% (Q2 FY26). A ~33-point gross-margin expansion in two quarters on mid-single-digit bit growth is pure price/mix. [FACT — filings]

Forward opportunities management is guiding to (each a hypothesis to be validated):

• HBM TAM. Management forecasts the HBM market growing from ~$35B (2025) to ~$100B (2028) — “larger than the entire DRAM market in calendar 2024” — a ~40% CAGR, with the $100B milestone “now projected to arrive two years earlier” than prior outlook. [FACT — Mehrotra, Q1 FY2026 call] Caveat: a self-serving supplier forecast that has been revised up repeatedly — momentum, but also forecast instability. [INTERPRETATION]
• HBM share. Micron reached HBM share “in line with our DRAM share” (~20–25%) in CQ3 2025 and declines to commit to a specific forward number, preferring to “manage the mix.” [FACT — calls]
• HBM sold out and priced. “We have completed agreements on price and volume for our entire calendar 2026 HBM supply, including HBM4.” This caps 2026 HBM upside but de-risks it; notably, the Q2 beat came from non-HBM DRAM repricing, since HBM 2026 prices were already fixed. [FACT — Q1 FY2026 call]
• First 5-year SCA. A multi-year strategic customer agreement with “far stronger contract structures and specific commitments” than prior 1-year LTAs — terms undisclosed; coverage % unknown. [FACT/OPEN QUESTION — Q2 FY2026]
• Data-center SSD / NAND. #3 global enterprise SSD share, exiting CY2025 at ~15% (from ~5% in 2022); data-center NAND revenue exceeded $1B in Q1 FY2026. [FACT — calls]
• LPDDR / LP server DRAM. Sole supplier of low-power DRAM into Nvidia’s GB-family data-center platforms; sampling 192GB LP SOCAMM2 modules. [FACT — calls]

Forward guidance, handled carefully. Management guided to continued sequential growth with “price again the largest factor” and “tight conditions for the foreseeable future and certainly beyond 2026,” declining to quantify beyond the current quarter. Sell-side models embed continued ASP-led upside through CY2026. We deliberately decline to publish a precise out-quarter figure that we cannot reconcile to a primary filing; the load-bearing verified data point is the Q2 FY2026 actual ($23.86B revenue, 74.4% GM, $12.07 GAAP EPS), which supersedes the prior $18.7B Q2 guide and demonstrates how fast ASPs moved. [FACT — Q2 FY2026 call/10-Q]

NAND: the structurally weaker half, easy to forget in the HBM excitement. While DRAM/HBM dominate the narrative, NAND (~20% of revenue) is a different and worse business: more fragmented (six-plus players including a rising YMTC), more commoditized, and lower-margin through the cycle. Micron’s NAND strategy is to ride data-center SSD share gains (now ~15%, #3 globally, up from ~5% in 2022) and lead on technology (G9, PCIe Gen6), but NAND will remain a margin drag and a capital sink relative to DRAM. The bull case for Micron is overwhelmingly a DRAM/HBM case; NAND is along for the ride and would be the first to feel a demand softening. An investor underwriting Micron is underwriting DRAM. [FACT/INTERPRETATION — FY2025 10-K; calls]

Verdict (Growth): high-quality cash generation, low-quality growth durability. The growth is real and the cash is real, but its persistence is a bet on supply discipline and the HBM trade-ratio, not on a defensible volume franchise. Essentially none of the incremental revenue is defensible unit growth — it is ASP that can reverse as fast as it arrived (FY2023 is the template). In the classic memory-cycle sense, this is low-quality growth occurring at an unusually durable point of the cycle.

6. Financial Quality

The cycle in one table (GAAP, $M except EPS). Memory’s economics are unreadable at a single point in time; they must be read across the cycle.

[FACT — FY2025/FY2023 10-Ks; Q2 FY2026 10-Q]

The recent quarters (GAAP, $M except EPS):

[FACT — 10-Qs]

Reconciling the impossible feed figure. A data feed showed Q2 FY2026 “operating margin 67.6% > gross margin ~58%,” which is arithmetically impossible. The filed figures resolve it cleanly: gross margin 74.4%, operating margin 67.6%, net margin 57.8% — the feed had mislabeled the net margin (57.8%) as gross margin. The real chain (GM 74.4% → OM 67.6% → NM 57.8%) is internally consistent. [FACT — Q2 FY2026 10-Q]

Balance sheet — transformed by the boom. At Q2 FY2026: cash + investments $16.6B, total debt $10.1B → ~$6.5B net cash, flipping from ~$2.6B net debt at FY2025. Stockholders’ equity surged $18.3B (to $72.5B) in two quarters, almost entirely retained earnings. Gross debt was cut from $14.6B to $10.1B (repaying $4.6B in H1, retiring higher-coupon notes via tender). Current ratio ~2.9×. All three rating agencies upgraded Micron in 2026. [FACT — Q2 FY2026 10-Q balance sheet; capital-allocation workstream]

Two quality-of-earnings flags worth holding in mind:

1. Receivables ballooned from $9.27B to $17.31B (+$8.05B) as revenue spiked — i.e., a meaningful slice of reported H1 profit is sitting in receivables, not collected cash. Not a manipulation flag, but a cyclical-quality watch item: if a top customer slows payment or ASPs reverse, this unwinds. In H1 FY2026, operating cash flow ($20.3B) modestly exceeded net income ($19.0B), so the cash conversion is still sound — but the receivables build is the single biggest working-capital swing factor. [FACT — Q2 FY2026 10-Q cash flow / balance sheet]
2. Inventory days collapsed (inventories flat ~$8.3B while revenue doubled) — bullish for tightness, but it also means little buffer if demand softens.

Cash flow and the capex problem. Memory is brutally capital-hungry. Capex has run 33–49% of revenue every year, and is largely non-discretionary (you cannot pause node transitions without falling behind):

[FACT — FY2025/FY2023 10-Ks; Q2 FY2026 10-Q]

The structural risk is stark: in the FY2023 trough, capex stayed at $7.7B even as operating cash flow collapsed to $1.6B, producing −$6.1B of free cash flow in a single year. FY2026 capex is guided above $25B (net of government incentives), so even with H1 operating cash flow of $20.3B, full-year free cash flow — while positive — will run far below the eye-popping earnings, because the cash is being reinvested. Government incentives are now cash-flow material ($2.0B received in FY2025, $2.26B in H1 FY2026, offsetting gross capex), which is why capex should be read “net of incentives.” [FACT — Q2 FY2026 10-Q MD&A liquidity]

Dilution and SBC — well-controlled. Diluted shares rose only ~1,093M (FY2023) → 1,142M (Q2 FY2026), ~4.5% over the period; stock-based comp of ~$1B/year is <3% of revenue, low for the sector. Buybacks and tax-withholding repurchases now partly offset issuance. No aggressive insider-dilution problem. [FACT — filings]

Returns — do not capitalize the peak. ROE/ROIC are wildly cycle-distorted: negative/NM in FY2023; ~17% ROE / ~15% ROIC in FY2025; ~60% ROE annualized in H1 FY2026. The annualized peak figures are artifacts and must not be capitalized — the canonical memory-cycle valuation error. Honest through-cycle ROIC for memory has historically sat in the low-to-mid teens at best, going negative at troughs, barely covering the cost of capital over a full cycle. That is the central quantitative case on the business’s intrinsic quality. [FACT/INTERPRETATION]

One-time items that distort the run-rate. FY2023 absorbed a $1.83B inventory net-realizable-value write-down plus ~$382M of facility underutilization costs (driving gross margin negative); the unwind of that write-down then flattered FY2024 by ~$987M (selling written-down inventory at near-zero cost), meaning FY2024’s thin underlying margin was even thinner than the 22% reported. By FY2025 that benefit was fully bled out, so FY2025’s 40% GM is “clean.” The forward effective tax rate is structurally rising (Q2 FY2026 14.7%, up from 10.1%) as the OECD Pillar Two 15% minimum tax offsets Micron’s Singapore incentives — modeling forward earnings at the historical sub-12% rate would overstate net income. [FACT — FY2023/FY2025 10-Ks; Q2 FY2026 10-Q Note 15]

Normalizing free cash flow — the number that matters for intrinsic value. Strip the cycle out and the picture is sobering. Across FY2021–FY2025, cumulative operating cash flow was ~$55B and cumulative capex ~$54B — i.e., roughly five years of breakeven aggregate free cash flow despite spanning a boom, a bust, and a recovery. H1 FY2026’s $8.5B of FCF is real, but the >$25B FY2026 capex guide means even this record year converts a minority of reported earnings into deployable cash, and the next trough will again take FCF deeply negative. A business whose multi-year free cash flow hovers near breakeven outside of a peak is not, on the evidence of its own history, a cash-compounding machine — it is a capital-intensive cyclical that occasionally showers cash and periodically consumes it. The current half-year is the shower; the discipline is to remember the consumption. [FACT — FY2021–FY2025 cash-flow statements; INTERPRETATION]

Verdict (Financial Quality): the economics improve dramatically with scale at the top of the cycle and collapse with scale at the bottom. Micron is currently a cash machine with a fortress balance sheet — net cash, low dilution, disciplined debt management. But the same operating leverage that produces a 74% gross margin today produced a negative gross margin in FY2023, and the capex intensity guarantees violently negative free cash flow at the next trough. The quality is real and cyclical; it is not a through-cycle compounding machine.

7. Capital Allocation

The dominant decision is reinvestment, and it is being made at the peak. Capex roughly tripled from $8.4B (FY2024) to a guided >$25B (FY2026), inside a ~$200B U.S. program (~$150B manufacturing + ~$50B R&D): two leading-edge Idaho fabs (ID1 online ~2H2027), up to four fabs in Clay, New York, and Manassas, Virginia expansion, targeting ~40% of DRAM produced in the U.S. This is supported by up to $6.1–6.4B of CHIPS Act direct funding, a 25–35% investment tax credit, and New York State incentives. [FACT — GlobeNewswire 2025-06-12; FY2025 10-K Note 20; Q2 FY2026 10-Q]

In the Marathon capital-cycle frame, this is the single largest red flag in the file. Capex nearly tripled precisely as memory prices, the stock, and free cash flow inflected up, and Micron is expanding at the same time as SK Hynix and Samsung — the canonical synchronized over-build into peak returns. The defenses are genuine: the build is partly subsidized (lowering net outlay and after-tax cost of capacity), it is secularly justified if AI/HBM demand is structural, there are only three rational leading-edge players, and it is a multi-year optionality-laden build (fabs online 2027+) rather than a spot capacity dump. But the structural improvement does not repeal the arithmetic: peak-margin capex lands as supply in 2027–28, the period of maximum cycle risk, and the incentive plan rewards exactly the bit-growth/share that encourages it. [INTERPRETATION]

Funding is conservative — the saving grace. Crucially, the build is funded by internal cash flow plus government subsidy, not leverage. Micron is repaying debt aggressively (repayments $4.6B FY2025, $4.6B in H1 FY2026) and holds ~$6.5B net cash. So being wrong on the capex will damage returns on capital but will not break the balance sheet — an important distinction that materially limits the downside of the bet. [FACT — filings]

Shareholder returns are a distant third priority.

• Dividend: initiated August 2021 ($0.10/qtr), never cut (including through FY2023), raised to $0.115 (Sept 2025) and ~30% to $0.15 (March 2026) — but at ~$0.60/year it yields ~0.06%, a token. [FACT — filings]
• Buybacks: $425M (FY2023) → $300M (FY2024) → $0 (FY2025) → $650M (H1 FY2026). The pattern is pro-cyclical and value-destructive on timing: zero repurchased near the trough, resumed at all-time highs (~$975). The $10B authorization (2018, no expiry) is constrained by CHIPS funding covenants. [FACT — FY2025 10-K issuer-purchases; Q2 FY2026 10-Q]

M&A — genuinely organic. Micron does not chase deals; it builds capacity. Its formative deals are old (Elpida 2013, Inotera/Rexchip consolidation), it wound down the Lexar brand (2017) and exited 3D XPoint (selling the Lehi fab to TI, 2021), and the trailing-24-month 8-K corpus shows no acquisitions. This removes integration/overpayment risk and concentrates all capital-allocation risk in the capex decision. [FACT — capital-allocation workstream; 8-K corpus]

Insider behavior — no conviction at the top. Across an 18-month sample of Form 4s, there were zero open-market purchases by any officer or director, and a continuous stream of (predominantly 10b5-1-planned) sales. CEO Sanjay Mehrotra sold steadily from ~$130 all the way to ~$980 — every leg of the move. The entire 18-person director-and-officer group owns <1% of shares outstanding (Mehrotra ~1.08M shares). The planned nature mutes the signal, but the total absence of buying through a 10× move, plus low and shrinking skin-in-the-game, is a mildly bearish-to-neutral tell that offers no conviction confirmation at these prices. [FACT — Form 4 corpus via data.sec.gov; DEF 14A ownership table]

Compensation — rewards the wrong thing for a cyclical. CEO FY2025 total pay ~$30.9M, ~82% equity. The short-term plan is 50% profitability (non-GAAP net income / operating margin) + 50% confidential strategic goals; FY2026 tightened to require both profitability gates for above-target payout. The long-term plan uses relative TSR vs. the semiconductor index (target 55th percentile, with a negative-absolute-TSR cap) plus HBM3E market-share (25%) and data-center-SSD share (25%), with NAND-FCF and U.S.-DRAM-expansion stretch conditions. Say-on-pay passed at a soft 84%. The structure has useful guardrails (relative TSR, the negative-TSR cap) but contains no explicit ROIC or return-on-capital metric — it rewards bit growth and market share, which in a capital-cycle business is precisely the incentive that encourages over-building into the peak. [FACT — DEF 14A 2025-11-25]

Verdict (Capital Allocation): a split decision — financially conservative, strategically aggressive at the wrong point in the cycle. Management deleverages into strength, never cut the dividend, and avoids value-destructive M&A — all correct. But it is making the defining bet of the company’s history — a ~$200B capacity build — at the most dangerous moment in the cycle, resuming buybacks at all-time highs, and paying executives on share/bit growth rather than returns on capital. The balance-sheet discipline caps the downside of being wrong; the ROIC consequence is unhedged.

8. Changes and Headwinds — Last Two Years

Strengthening the franchise:

• AI/HBM demand inflection and the recovery from FY2023. The single largest change: from a $5.8B loss in FY2023 to ~$24B TTM net income, driven by HBM and the DRAM shortage. [FACT]
• Business-unit reorganization (Q4 FY2025). New CMBU/CDBU/MCBU/AEBU segmentation reframes the company around data center (now ~56% of revenue) and improves margin transparency — a presentation and strategic upgrade. [FACT — Q4 FY2025 call]
• Balance-sheet transformation. Return to net cash, three credit-rating upgrades in 2026, record free cash flow, aggressive deleveraging. [FACT]
• Technology leadership cemented. 1γ DRAM (EUV) ramping faster than any prior node; G9 NAND first to market on Nvidia’s reference platform; HBM4 ramping ~2× faster than HBM3E. [FACT — calls]
• U.S./diversified fab footprint + CHIPS funding — strategic and political asset; SCAs lengthening order-book visibility. [FACT]
• Leadership continuity — Mehrotra (Chairman/CEO), Murphy (CFO), Sadana (CBO), Bhatia (EVP Ops) stable through the up-cycle; board refreshed (Beyer/McCarthy retirements; Björlin added June 2026). [FACT — proxy / 8-Ks]

Headwinds and risks:

• The June-2026 “memory peak” debate — the live bear catalyst. Raymond James warned ASPs may peak by mid-2026; Barron’s ran “Brace for the Memory-Price Peak.” On June 4, 2026, Broadcom’s soft AI guide triggered a sector “sell-the-news,” knocking Micron ~7% in a day and ~13% off its peak. [FACT — TheStreet 2026; CNBC 2026-06-04]
• Parabolic stock / sentiment saturation. Micron crossed $1 trillion on May 26, 2026, +84% in May (best month since 1985), ~+1,000% over the trailing year. A 2× short ETF (“MUZ”) launched in June 2026; Renaissance Technologies reportedly trimmed. A momentum positive and a mean-reversion risk simultaneously. [FACT — CNBC/Benzinga 2026-05-26]
• China CAC ban (May 2023) — largely absorbed. China barred Micron from “critical information infrastructure”; the directly-affected revenue subset was limited (most China revenue is consumer/mobile), and the AI surge has overwhelmed it. A residual overhang, not a current driver. [FACT — TechCrunch/CSIS 2023]
• Tariffs/export controls — unquantified. Management explicitly excludes potential new tariffs from guidance. [OPEN QUESTION — calls]
• Rising effective tax rate (Pillar Two) structurally lowers forward net income vs. the historical rate. [FACT]

Verdict (Changes/Headwinds): the structural changes strengthen the long-term franchise; the near-term changes introduce acute cyclical/sentiment risk at the point of maximum optimism. The business is structurally better than at any prior peak; the stock is priced as if the cycle is abolished. Those two truths are the crux of the entire thesis.

9. Risk Analysis

The two risks that dominate the thesis are the first two — cyclical ASP reversion and peak-earnings de-rating — and they are correlated. The probability of a total loss is negligible (net cash, essential product); the probability of a large drawdown from current levels on a cycle turn is, in our view, high. The asymmetry at $936 is unattractive: limited incremental upside if the structural case is already largely priced, against hundreds of dollars of downside to a normalized book/earnings multiple if the cycle behaves like every prior cycle.

10. Valuation Discussion (Embedded Expectations)

The setup. At $935.89, market cap ~$1.06T and EV ~$970B (net cash). Multiples: trailing P/E ~44.8×, forward P/E ~8.6×, EV/EBITDA ~26×, P/S (TTM) ~18×, P/B ~13.4–14.6×. On Micron’s own 10-year history, the composite valuation percentile is 96th (P/B 99.8th, P/S 99.8th, P/E 89.5th). [FACT — public market multiples; own-history valuation percentiles, accessed 2026-06-08]

Reading the multiples correctly. The juxtaposition of a low forward P/E (~8.6×) with an extreme price/book (~14.6×) is the fingerprint of a cyclical at a top. Cyclicals look cheapest on P/E at the peak (a low multiple on peak EPS) and most expensive on P/E at the trough (a high multiple on depressed EPS) — the Peter-Lynch warning that a cyclical at a single-digit P/E is often a sell, not a buy. The forward P/E is therefore the least reliable lens here; price/book and price/normalized-earnings are the more honest anchors, and both say expensive.

Peer comps (storage/memory and AI-semis). The memory-storage peers (Seagate, Sandisk/WDC) trade at elevated multiples too — the whole storage complex has re-rated on AI — while the broader AI-semis (Nvidia, Broadcom) carry higher absolute multiples on faster, less-cyclical growth:

[FACT — public market comps, accessed 2026-06-09] The comp table cuts both ways: MU’s forward P/E is the lowest in the group (again, the peak-EPS illusion), but its cyclicality is by far the highest — the right peer for normalized-earnings risk is not Nvidia (a secular grower) but Micron’s own FY2023 self, which lost money. The storage names (STX, WDC) trading at 30–80× trailing confirm the entire complex is priced for AI persistence.

A reverse look at the embedded expectation. Turn the valuation around: at ~$1.06T market cap, even crediting a generous durable through-cycle net margin of ~25% (well above any prior-cycle average and roughly double FY2025’s), the market is paying ~$1.06T for a business that would need to sustain ~$25–30B of annual normalized revenue at that margin to earn ~$6–7.5B of normalized net income — i.e., ~140–175× normalized earnings, or, put differently, the price only makes sense if normalized earnings are not $6–7.5B but the $30B+ annualized run-rate of H1 FY2026 and that run-rate holds. The entire valuation rests on the run-rate being the new normal. That is the embedded bet, stated arithmetically.

Embedded-expectations analysis — what must be true to justify $936? The market is paying ~14.6× book and ~$1.06T for a business whose through-cycle normalized earnings power, on history, is a fraction of the current run-rate. For the current price to be “correct,” memory’s through-cycle economics must have structurally and permanently stepped up — i.e., HBM + data-center mix + supply discipline must hold normalized EPS durably in the tens of dollars (not the low-single-digits-to-teens of prior cycles), and the market must continue to capitalize those earnings at a non-trough multiple. In short, the price embeds the abolition (or at least the multi-year suspension) of the memory cycle. That is a strong-form bet.

Scenario analysis (illustrative, GAAP, deliberately wide — no point estimate). These are scenarios for thinking about asymmetry, not forecasts:

[INTERPRETATION — scenario construction from filed financials, industry data, and the published sell-side range; not a recommendation or price target]

The genuinely wide distribution is the point. The current price (~$936) sits toward the bull/structural end on book and sales (99th percentile of its own history) while looking cheap only through the peak-EPS forward-P/E lens. Sell-side targets ($1,050–$1,625) sit squarely in the bull scenario and implicitly underwrite persistence. The bear anchor is not exotic — the 52-week low was $103, and a 2–4× multiple on a growing book is the historical norm for memory outside of peaks.

What the market is underwriting correctly vs. incorrectly (our read). Correctly: that AI/HBM demand is real, that 2026 is contracted and tight, that Micron’s technology execution is strong, and that the balance sheet is fortress-grade. Potentially incorrectly: that this through-cycle earnings step-up is permanent and that synchronized industry capex will not, as it always has, eventually restore the glut. The embedded expectation is persistence; the historical base rate is mean reversion. No price target. No recommendation (see Claude’s Take for the labeled exception).

11. Variant Perception

Consensus. The market has re-rated Micron to $1T+ and sell-side targets have cascaded higher (UBS $1,625; Cantor $1,500; Wells Fargo $1,220; Morgan Stanley $1,050 — all raised late May/June 2026). Consensus is underwriting the structural AI-supercycle case: that this is not a normal memory cycle, that HBM’s rising trade-ratio plus cleanroom lead-times plus multi-year SCAs hold supply short of demand for “2–3 years,” and that data-center mix permanently lifts through-cycle margins. [FACT — TheStreet 2026]

Strongest bull case. The HBM die penalty (~3× wafers/bit, rising each generation) is a real, mechanical wafer sink; per-node bit-growth is declining (supply-side scarcity); cleanroom buildout runs to ~FY2028; multi-year SCAs with partially-fixed pricing dampen the historical price collapse; Micron has genuine technology leadership and a net-cash balance sheet. If demand is “supply-constrained for the foreseeable future,” then ~8.6× forward earnings is absurdly cheap and the stock is early, not late.

Strongest bear case. Margins went from loss-making (FY2023) to 74% gross margin (Q2 FY2026) in roughly two years — that amplitude is the cycle, not its abolition. The entire surge is ASP-driven on mid-single-digit bit growth; ASPs reverse as fast as they rise. Three players are all adding greenfield DRAM capacity into 2027–28 — the Marathon capital cycle says high returns attract capital and mean-revert. The forward P/E is a peak-EPS illusion; the honest anchors (price/book ~14.6×, 99th percentile of own history) scream top. After +1,000% in a year, with insiders selling, a 2× short ETF launching, and the first peak warnings landing, the stock prices perfection.

The crux: bull and bear agree on the facts and disagree only on persistence. Both accept that AI demand is strong, that supply is tight today, and that margins are ASP-driven. The entire disagreement is whether that lasts. The market has fully priced persistence.

The 3–5 assumptions that matter most, and what falsifies each:

The single most important variant-perception insight: the falsification test is mechanical and observable. The first quarter of sequential DRAM ASP decline, or any major competitor capacity announcement, breaks the structural narrative. Until then, the tape is momentum; after then, it is a falling knife.

12. Fact vs. Interpretation

13. Open Questions

1. Standalone HBM revenue and units — Micron discloses the CMBU envelope but not a clean HBM dollar/unit figure in filings; sourced only from calls. What is the true HBM revenue and gross margin vs. non-HBM DRAM? (Management has conceded non-HBM DRAM margins currently exceed HBM margins.)
2. SCA coverage — what fraction of FY2026–28 DRAM/NAND bits are under multi-year strategic agreements, and how price-protective are they when spot rolls over?
3. The ~13%-of-revenue customer — identity, durability, and whether HBM concentration is rising toward a single-customer dependence.
4. FY2026 H2 free cash flow after the >$25B capex ramp — how much of the reported profit actually converts to deployable cash?
5. Through-cycle normalized margin/ROIC — what is the right normalized assumption if HBM has partially (not fully) raised the floor? This is the entire valuation question.
6. The June-9-2026 8-K (Item 5.02 director/officer change) — confirm whether any leadership change is material (board addition of Björlin appears routine; verify no negative officer departure). [Flagged by transcripts workstream]
7. Tariff exposure — explicitly excluded from guidance; magnitude unquantified.
8. Exact remaining buyback authorization and the precise CHIPS-covenant constraints on repurchases.

14. What Must Be True

Bull case — what must be true, and its falsification test. For the bull (structural supercycle) to be right, memory’s through-cycle earnings power must have permanently stepped up: HBM’s rising die penalty + cleanroom lead-times + multi-year SCAs must hold DRAM supply short of demand into ~FY2028, all three producers must maintain capex discipline despite peak returns, and data-center mix must keep the through-cycle gross-margin floor structurally above prior cycles.

• Falsification test: the first quarter of sequential DRAM ASP decline, OR any major competitor (especially Samsung) announcing aggressive greenfield DRAM capacity, OR rising DRAM inventory days. Any one of these breaks the “this time is different” thesis. The test is mechanical and will be observable in quarterly ASP/inventory data and competitor capex announcements.

Bear case — what must be true, and its falsification test. For the bear (cyclical top / peak-earnings de-rating) to be right, the current ASP-driven margins must prove transient: synchronized 2027–28 capacity must restore oversupply, ASPs must revert, and the market must re-rate Micron from ~14.6× book toward a normalized 2–4× book on mean-reverting earnings.

• Falsification test: multiple consecutive quarters of sustained sequential ASP strength accompanied by demonstrable supply discipline (no greenfield capacity announcements, SCA coverage expanding, inventory days staying low) through FY2027 — i.e., the cycle visibly failing to turn on schedule. If Micron prints durable double-digit-dollar EPS with bit-growth-led (not just ASP-led) revenue and the industry holds capex, the structural case is confirmed and the bear is wrong.

The elegance — and the discomfort — of this name is that both falsification tests key off the same observable: the trajectory of sequential DRAM ASPs and competitor capacity. The thesis will resolve in the data, quarter by quarter.

15. Source Appendix

See the Source Appendix (Appendix B, below) for the full citation list. Primary sources relied upon: Micron FY2021–FY2025 Forms 10-K; Q1 & Q2 FY2026 Forms 10-Q; DEF 14A (2025-11-25); Form 4 corpus (data.sec.gov); FY2026 earnings-call transcripts (Q3 FY2025 through Q2 FY2026, plus the May-2026 J.P. Morgan TMT conference); SK Hynix public results; industry data (TrendForce, Omdia, Counterpoint); and contemporaneous financial press (CNBC, Benzinga, TheStreet, Barron’s). Quantitative orientation from public market data, with every material figure reconciled to filings.

No buy/sell recommendation and no price target appears anywhere in the body of this article; the only position is the clearly-labeled Claude’s Take block, which is the author’s own independent opinion. Not investment advice.

APPENDIX A — Standard Diligence Questionnaire

Standard Diligence Questionnaire — Micron Technology, Inc. (NASDAQ: MU)

Supplemental to the analysis above. Answers grounded in primary filings; Fact / Interpretation / Assumption labels applied where it matters. As of 2026-06-09.

General

What thoughtful questions have other investors asked about this company? The central question every serious investor asks about Micron is the same: “Are these peak earnings, and has AI structurally changed the memory cycle?” Specific sub-questions in the 2026 debate: (1) How much of the margin surge is HBM-specific vs. a general DRAM-shortage ASP spike? [Answer: largely the latter — management concedes non-HBM DRAM margins currently exceed HBM margins; the surge is broad ASP, with HBM as the cause of the shortage rather than the sole beneficiary.] (2) When does the next down-cycle arrive, and how deep, given synchronized capacity additions landing 2027–28? (3) How price-protective are the new multi-year SCAs when spot rolls over? (4) Is Micron’s HBM share durable against SK Hynix (leader) and a recovering Samsung? (5) Does the >$25B capex program destroy through-cycle ROIC? These are the right questions; the memo’s view is that the market has answered them too optimistically.

Cyclicality & Earnings Nature

Are earnings at a cyclical high or low? [FACT/INTERPRETATION] A cyclical high — and an extreme one. Gross margin went from negative (FY2023) to 74.4% (Q2 FY2026); annualized H1 FY2026 ROE ~60%. These are peak-cycle figures; the canonical error would be to capitalize them.

Driven by the external environment or internal actions? Overwhelmingly external (the AI-driven memory shortage and ASP spike), amplified by genuine internal execution (1γ DRAM ramp, HBM qualification). The ASP move — not Micron’s volume or cost actions — is the dominant driver.

How stable are revenues? Highly unstable through a cycle. Revenue swung $30.8B (FY2022) → $15.5B (FY2023) → $37.4B (FY2025) → ~$48B+ annualized (H1 FY2026 run-rate). There is no recurring-revenue floor; output is re-sold quarterly at market-clearing prices, now partly cushioned by emerging multi-year SCAs.

Outlook for products/services? Demand for memory bits grows secularly (~mid-to-high-teens % annually for DRAM/NAND), but price — not volume — drives the P&L swings. The forward question is supply/demand balance, not end-demand existence.

How big will this market be — growing, shrinking, domestic or international? [FACT] Growing and global. Total memory ~$365B (2025) → ~$552B (2026), forecast ~$843B peak (2027); HBM TAM ~$35B (2025) → ~$100B (2028) per management. The market is real and large; the risk is cyclicality and capacity, not market size.

Business Quality & Competitive Moat

Is the industry getting more or less competitive? [INTERPRETATION] Structurally less competitive at the leading edge (consolidated to 3 DRAM players; HBM concentrated) but with a rising China threat (CXMT/YMTC) at the commodity tail. Net: more rational at the top, more contested at the bottom.

How profitable is the business (ROIC, ROE)? [FACT] Wildly cycle-dependent: NM/negative (FY2023), ~17% ROE / ~15% ROIC (FY2025), ~60% annualized ROE (H1 FY2026 — a peak artifact). Through-cycle ROIC has historically been poor — around or below the cost of capital. That is the defining quality concern.

How profitable is the industry — how many competitors, what barriers to entry? Three scaled DRAM producers (~90%+ share); high capital barriers (a leading-edge fab is tens of billions of dollars; EUV know-how is scarce). But the barriers protect an oligopoly that still competes on a commodity, so industry profitability is high at peaks and negative at troughs.

Can the business be easily understood? Yes — clean model (sell memory bits at a spread to cost). The difficulty is not understanding it but timing the cycle.

Can it be undermined by foreign low-cost labor? Not labor (it is capital/technology-intensive), but it can be undermined by foreign state capital — China’s CXMT/YMTC building subsidized capacity indifferent to ROIC, commoditizing legacy nodes.

Do brands matter? Minimally. Crucial has modest consumer pull; the business is overwhelmingly B2B where qualification, performance, power efficiency, and price — not brand — decide.

What is the nature of competition? Cost-per-bit leadership, technology-node timing, and HBM qualification wins. It is a capital and engineering race, re-run each generation.

Customers’ switching costs? Real but generation-bound: HBM/auto qualification creates multi-quarter lock-in for a product generation, but customers re-qualify (and can re-allocate) each generation. Not a compounding lock-in.

Financial Condition & Balance Sheet

Assets not fully recognized on the balance sheet? [INTERPRETATION] Technology/IP and the U.S. fab optionality (subsidized by CHIPS/ITC) are arguably worth more than book in the bull case. Conversely, the fab base is carried at cost and would be impaired in a deep downturn.

Off-balance-sheet liabilities? Purchase commitments for capital equipment (substantial given the >$25B capex program) and CHIPS-funding covenant obligations (including buyback restrictions). No unusual off-balance-sheet leverage identified. [Verify exact purchase-commitment figures in the FY2025 10-K commitments note.]

How conservative is the accounting? [INTERPRETATION] Reasonably conservative — large NRV inventory write-downs taken promptly in FY2023, transparent segment recast. Watch item: the H1 FY2026 receivables build (+$8.05B) means a slice of reported profit is uncollected; not aggressive, but cyclical-quality-sensitive.

How CapEx-hungry is the business? [FACT] Extremely — capex is 33–49% of revenue every year and largely non-discretionary. This is the single most important structural feature: it drives violently negative free cash flow at troughs (−$6.1B FY2023).

Capital Allocation & Management

How much FCF does the business generate, how does management use it, what is the philosophy? [FACT] FCF is violently cyclical (−$6.1B FY2023; +$8.5B H1 FY2026). Priority order: (1) reinvestment/capex (dominant — the ~$200B U.S. build), (2) deleveraging, (3) token dividend + opportunistic buybacks. Philosophy: build for secular AI demand, fund with internal cash + government subsidy, keep the balance sheet conservative.

Significant acquisitions recently? No — growth is organic. Only the small Powerchip Tongluo fab acquisition (no meaningful bits before FY2028). No empire-building; low integration/overpayment risk.

Buying back shares? Yes, but pro-cyclically: $0 in FY2025 (near trough) → $650M H1 FY2026 (at all-time highs). Value-destructive timing on the one lever where price matters.

Issuing large amounts of new shares to insiders? No — dilution is low (~4.5% over the cycle); SBC <3% of revenue.

Compensation policy of directors/management? [FACT] CEO ~$30.9M FY2025 (~82% equity). STI: 50% profitability (non-GAAP NI/op margin) + 50% strategic. LTI: relative TSR vs. semis (target 55th pct, negative-TSR cap) + HBM3E share (25%) + DC-SSD share (25%) + NAND-FCF/US-DRAM stretch. Say-on-pay 84% (soft). Critical gap: no explicit ROIC/return-on-capital metric — rewards bit growth and share, which encourages over-building.

Motivations of management? [INTERPRETATION] Competent operators and conservative financiers, but incentivized toward growth/share over capital efficiency, and with low personal ownership (<1%) and a clear pattern of selling into strength (no open-market buys $130→$980). Aligned to grow the franchise; not strongly aligned to through-cycle ROIC.

Valuation & Market Data

Is the stock an ADR, MLP, or K-1 issuer? No — common stock of a U.S. C-corporation (NASDAQ: MU). Standard 1099 reporting.

Dividend policy? [FACT] Token. Initiated Aug 2021; never cut; ~$0.60/year currently (+30% raised March 2026); ~0.06% yield; ~2% payout. Immaterial to total return.

How profitable is the business? At present, extraordinarily (57.8% net margin Q2 FY2026); through a cycle, modestly-to-poorly (loss-making at troughs). See ROIC above.

Is net income diverging from cash from operations? [FACT] In H1 FY2026, OCF ($20.3B) modestly exceeded NI ($19.0B), so conversion is sound — but a large chunk of profit sits in the +$8.05B receivables build (uncollected). A watch item, not a red flag.

Risks & Downside

What factors would cause the stock to decline? [FACT/INTERPRETATION] (1) The first sequential DRAM ASP decline (the master falsifier); (2) a competitor (esp. Samsung) announcing aggressive greenfield capacity; (3) AI-capex/demand wobble (the June-2026 Broadcom-guide selloff is the template); (4) peak-earnings de-rating from 14.6× book; (5) HBM share loss; (6) tariff/China/policy shocks. The first two are the highest-signal triggers.

Risk of a catastrophic loss? [INTERPRETATION] Of a large drawdown, high (cyclical reversion from a 96th-percentile valuation). Of a permanent capital impairment / bankruptcy, very low — net cash, IG balance sheet, essential product. The downside is a deep cyclical drawdown, not a zero.

Chance of a total loss? Negligible. Even at the FY2023 trough Micron remained solvent with ample liquidity; today it holds ~$6.5B net cash.

Recent News & Events

Has the business environment changed recently? [FACT] Yes, dramatically and in both directions: the AI/HBM boom drove revenue and the stock to records (crossed $1T May 26, 2026; +84% in May, best month since 1985), while June-2026 brought the first “memory-price peak” warnings (Raymond James: ASPs peak by mid-2026) and a Broadcom-soft-AI-guide sector selloff (MU −7% on June 4). The environment is euphoric and simultaneously showing the first cracks.

Significant acquisitions? No (organic; small Tongluo fab only).

Change in accounting policies? Segment reorganization in Q4 FY2025 (CMBU/CDBU/MCBU/AEBU) — a reporting/presentation change, not a policy manipulation; prior periods recast.

Recent changes — new markets, facilities, management? [FACT] Major facility expansion (Idaho fabs, New York Clay fabs, Manassas, Singapore HBM packaging) under the ~$200B U.S. program + CHIPS funding; stable senior management (Mehrotra/Murphy/Sadana/Bhatia); board refresh (Beyer/McCarthy retirements, Björlin added June 2026). Rising effective tax rate (Pillar Two). China CAC ban (2023) largely absorbed.

APPENDIX B — Source Appendix

Source Appendix — Micron Technology, Inc. (NASDAQ: MU)

All facts in this article trace to a public source below. Primary (filings) before secondary. Accessed 2026-06-09 unless noted.

A. Primary SEC filings

B. Earnings-call & event transcripts (company IR / public transcript sources)

C. Peer & cross-read data (public)

• SK Hynix (000660.KS) — public financial results: direct memory peer for industry structure, HBM/AI framing, capital-cycle analysis, the FY2023 trough (−₩7.7T operating loss), and peak/through-cycle ROIC benchmarks.
• TSMC, Infineon, Nvidia — public filings and disclosures, used for semiconductor-industry and AI-demand cross-read.

D. Industry & market data (secondary)

• Omdia / TechTimes — DRAM Q1-2026 market share (Samsung 38.6%, SK Hynix 28.8%, Micron 22.4%), 2026-06-09. https://www.techtimes.com/articles/318052/20260609/samsung-leads-dram-market-share-386-sk-hynix-trails-revenue-tops-profit-margins.htm
• TrendForce — 4Q25 DRAM revenue / Samsung regains #1, 2026-02-26. https://www.trendforce.com/presscenter/news/20260226-12937.html
• TrendForce — 1Q26 memory contract price record highs, 2026-02-02. https://www.trendforce.com/presscenter/news/20260202-12911.html
• TrendForce — 2026 capex / limited bit-supply impact, 2025-11-13. https://www.trendforce.com/presscenter/news/20251113-12780.html
• TrendForce — Micron $25B FY26 capex + first 5-year deal, 2026-03-19. https://www.trendforce.com/news/2026/03/19/news-micron-ramps-fy26-capex-to-25b-signs-first-5-year-customer-deal/
• TrendForce — Samsung/SK Hynix tapped for Nvidia Rubin HBM4, 2026-03-09. https://www.trendforce.com/news/2026/03/09/news-samsung-sk hynix-reportedly-tapped-as-nvidia-rubin-hbm4-suppliers-shipments-could-start-in-march/
• Astute Group / Counterpoint — HBM share (SK Hynix 62%, Micron #2 ~21%), 2026. https://www.astutegroup.com/news/general/sk-hynix-holds-62-of-hbm-micron-overtakes-samsung-2026-battle-pivots-to-hbm4/
• Tom’s Hardware — DRAM/NAND Q2-2026 contract price forecast; Micron US fab timeline. https://www.tomshardware.com/pc-components/dram/dram-and-nand-contract-prices-to-climb-again-in-q2
• Neumonda — Memory Market 2026 scarcity/glut risk, 2026. https://www.neumonda.com/memory-market-2026-scarcity-strategy-and-security-of-supply/
• TheStreet — Micron ~$100B/$200B US fab program; Raymond James ASP-peak warning; Morgan Stanley $1,050 PT. https://www.thestreet.com/investing/stocks/micron-building-massive-memory-fabs-in-new-york-as-supercycle-ramps

E. Capital-allocation & event sources

• GlobeNewswire — Micron / administration ~$200B U.S. investment announcement, 2025-06-12. https://www.globenewswire.com/news-release/2025/06/12/3098344/14450/en/Micron-and-Trump-Administration-Announce-Expanded-U-S-Investments-in-Leading-Edge-DRAM-Manufacturing-and-R-D.html
• Micron Q2 FY2026 earnings press release (8-K Ex-99.1), 2026-03-18. https://www.sec.gov/Archives/edgar/data/0000723125/000072312526000004/a2026q2ex991-pressrelease.htm

F. News / sentiment (financial press)

• CNBC — Micron crosses $1 trillion market cap, 2026-05-26. https://www.cnbc.com/2026/05/26/micron-stock-trillion-market-cap.html
• Benzinga — Micron best month since 1985 (+84% May), 2026-05. https://www.benzinga.com/markets/market-summary/26/05/52887979/
• CNBC — Broadcom soft AI guide triggers chip selloff (MU −7%), 2026-06-04. https://www.cnbc.com/2026/06/04/chipmaker-equities-micron-marvell-broadcom-intel.html
• TechCrunch / CSIS — China CAC ban on Micron, 2023-05-21. https://techcrunch.com/2023/05/21/china-bans-micron/
• Published sell-side analyst price-target moves (Cantor $1,500, Wells Fargo $1,220, UBS $1,625, Morgan Stanley $1,050), late May / June 2026, as reported in the financial press. Third-party signals; validated against primary sources.

G. Quantitative orientation (public market data, reconciled to filings)

• Public market data providers — price, market cap, enterprise value, valuation multiples, dividend yield, ownership and short-interest, multi-year financials, and peer comps (STX, WDC, AVGO, NVDA, TXN). Own-history valuation percentiles (composite 96th, P/B 99.8th, P/S 99.8th), accessed 2026-06-08/09. Third-party aggregated data; every material figure reconciled to SEC filings.
• SEC submissions API — data.sec.gov/submissions/CIK0000723125.json (filing index, Form 4 enumeration).