Micron Technology, Inc. (NASDAQ: MU) — The Best Quarter in Memory History, Priced as the Last One

An independent fundamental research note. Report date: 2026-07-18 Price at writing: $848.95 (2026-07-17 close) · Market cap: ~$976B · Enterprise value: ~$952B (net cash ~$24.4B) Fiscal year: ends last Thursday of August (FY2026 is a 53-week year; latest reported quarter FQ3 FY2026 ended 2026-05-28) Coverage status: UPDATE — follows the 2026-06-09 initiation ($935.89). This report stands alone but diffs against the prior thesis throughout.

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

⚡ Kimi’s Take

This block is Kimi’s own subjective opinion — the author’s independent view, offered as general information only. It is not investment advice. Everything below it is the position-free analysis; this block is the single opinion in the report.

Verdict: HOLD / do-not-chase rallies at ~$849. Not a short. The accumulation zone moves up: I would become a buyer on a further cyclical reset into roughly the $450–650 zone — provided the SCA floor is still intact when we get there. Conviction: medium.

Tag: “The best quarter in memory history, priced as the last one.”

Five weeks ago the prior report called Micron “a trillion-dollar bet that the cycle is dead” and refused to chase $936. The market has since litigated that call in both directions and, characteristically, both sides got a trophy. The business got better than the bull case required: FQ3 printed $41.5B of revenue (+74% QoQ), an 84.6% gross margin, $24.67 of EPS, and guided FQ4 to ~$50B and ~$31 EPS — while signing 16 take-or-pay Strategic Customer Agreements covering ~25% of revenue with price floors management says sit above any prior cycle’s peak margins, backed by $22B of customer deposits. And the stock got worse than the bear case demanded: after spiking +16% to a $1,255 all-time high on the print, it round-tripped −32% in three weeks — the largest positive revision wave in company history (FY27 consensus EPS up ~47% since May to $150.77) absorbed entirely by de-rating. At 5.6× FY27 consensus, the tape now prices ~55–60% permanent erosion of earnings the sell side capitalizes at ~10–11×. The embedded expectation has migrated from “the cycle is abolished” (what $936 required) to “the cycle returns, but the floor is several times higher” — which is, not coincidentally, close to my own base case.

So why not buy the de-rating? Three reasons. First, the slow anchors are still extreme: 13.4× book, the 98.5th percentile of Micron’s own history — cheap only through the forward-P/E lens that has never been cheap on a memory stock. Second, the oversupply endgame is now a dated construction schedule: Samsung pulled its ~$82.7B P5 twin-fab forward to a July 2026 groundbreaking — the amber flag the prior report pre-specified — SK hynix just raised ~$26.5B in New York, CXMT is scaling toward ~350k wafers/month, and Micron itself will spend >$40B in FY27. The floor protects ~25% of revenue; the other 75% meets that wall of 2028–29 bits. Third, the tape itself is broken: a −1.61 low-volatility factor loading, the momentum regime rolled over a month ago, and memory stocks do not bottom on the first 30% off the high. But shorting is equally wrong: DRAM ASPs are still rising sequentially (+low-60s% in FQ3, +13–18% guided for CQ3 contracts), 2027 is contracted tighter than 2026, the balance sheet carries $24.4B net cash, and short interest is a trivial 2.8%. Shorting a company printing ~$31 of quarterly EPS against contracted demand is how careers end. The honest position is patience: the de-rating has done about half the work a reset would do.

What flips me bullish: a reset into ~$450–650 (≈8–10× through-cycle earnings, ~4–6× forward book that compounds ~$25/share per quarter) with evidence the SCA floor is transacting — deposits converting to floor-priced revenue, no renegotiations, coverage rising. What flips me bearish (toward an actual short): the first quarter of sequential DRAM ASP decline, any disclosed SCA floor renegotiation or deposit forfeiture (the 2023 LTA playbook), or a 2028 bit-supply surprise from Samsung/CXMT — any one breaks the higher-floor thesis while the stock still carries 13× book.

Changes since 2026-06-09 (update summary)

What moved the thesis. (1) FQ3 FY2026 (2026-06-24) — revenue $41.46B vs ~$35.7B consensus, GAAP gross margin 84.6%, diluted EPS $24.67, FQ4 guided to $50B ± $1B / ~86% GM / ~$31 EPS: the demand question the prior report left open is settled for now. (2) The SCA contract layer — 16 take-or-pay agreements, ~$100B minimum-price cumulative RPO, ~$22B deposits, floors “well above our peak quarterly margins in any past cycle” (management): the first contractual answer to memory’s 40-year boom-bust, covering ~25% of revenue. (3) The de-rating — −32% from the 2026-06-25 all-time high of $1,255 while FY27 consensus rose ~32%: the market no longer prices persistence of peak earnings at all. (4) The supply response got scheduled — Samsung P5 Fab 2 groundbreaking pulled forward six months to July 2026 (~$82.7B program), SK hynix’s ~$26.5B US listing, CXMT’s ~$8.5B STAR IPO filing: 2028–29 oversupply is now a calendar, not a hypothesis.

What was confirmed. The ASP-driven nature of the surge (DRAM ASPs +low-60s% QoQ on low-single-digit bit growth — the cycle’s signature, again); the fortress balance sheet (net cash $6.5B → $24.4B; $9.4B of debt repaid); pro-cyclical insider behavior (zero open-market buys; ~$91M sold in five weeks at $983–$1,192, including a discretionary director sale); the competitive frame (Micron #3 in DRAM, and now diluted to a ~10% HBM4 share assumption at Rubin per UBS as Samsung re-entered).

What was falsified or degraded. The prior report’s bull falsification tests did not trigger on their master criterion — sequential DRAM ASPs are still rising, and customer inventories (~7–9 weeks) remain below the alarm threshold. One pre-specified amber flag did fire: a major competitor (Samsung) moved aggressively on new capacity (criterion b). The prior report’s bear falsification test moved closer: sustained ASP strength plus supply discipline now holds through FY2027 on third-party models (Goldman/UBS/TrendForce), and the SCA floor is a structural innovation the bear case must now digest. Net: the prior HOLD call stands, the bear’s trough anchor rises, and the debate has narrowed from “is demand real?” (settled) to “what is the uncontracted 75% of revenue worth through the 2028 supply wave?”

📈 Stock Price Action — Five-Year Event Map

Arc. Micron’s five years are the memory cycle drawn in one line: a COVID-era DRAM peak (~$96, Jan 2022), a −50% bust into the FY2023 loss (~$48, Sep 2022), an AI/HBM recovery to ~$156 (Jun 2024), a −59% round-trip on HBM-qualification lag and tariff shocks (~$64, Apr 2025), and then the steepest up-leg in the company’s history — roughly 10× in twelve months off the $103 52-week low (2025-08-01), a $1T market-cap cross in May 2026, and an intraday peak of $1,255 on 2026-06-25, the day after the FQ3 FY2026 print. As of the 2026-07-17 close of $848.95, the stock sits −32.3% below that intraday peak (−30.0% below the $1,213.56 closing high), still ~+723% above the 52-week low and −9.3% below the 2026-06-09 prior-report price of $935.89. The 2026-07-17 session itself spanned $804.00–$903.93 on 62.8M shares — an ~12% intraday range that is now a normal day for this name. [FACT — adjusted price history, accessed 2026-07-18]

Note: the 2026-06-09 memo did not carry a literal event-map table; this block rebuilds the arc from the price record and refreshes it. Prices are split/dividend-adjusted closes; intraday extremes noted where relevant. Moves are FACT; attributed causes are INTERPRETATION and are labeled as such.

Event narratives (post-2026-06-09 rows, the new material):

• Row 6a — 2026-06-11 to 2026-06-22 run to $1,211 (+11.7% on 6/11, +10.8% on 6/15, +8.7% on 6/18, +6.8% on 6/22). [FACT — prices.] Drivers [INTERPRETATION]: a broad tech rebound after Trump called off Iran strikes and the Dow jumped 930 points (6/11); pre-earnings positioning into the 6/24 FQ3 print amid relentless memory-price newsflow (HBM4E sampling at SK Hynix, Nvidia/Korea AI headlines). Sources: Investopedia markets wrap, 2026-06-11; Witology semiconductor daily, 2026-06-18; Korea Market Morning Brief, 2026-06-16.
• Row 6b — 2026-06-23, −13.2%. [FACT — Seattle Times/AP: “Chip companies, which have soared in recent weeks, sank. Micron Technology lost 13.2%.”] Driver [INTERPRETATION]: a memory-sector rout imported from Korea (KOSPI −9–10%; Samsung −12.3%, SK Hynix down double digits), compounded by rate-hike anxiety and a “demand-crack” narrative (Microsoft reportedly weighing DeepSeek R4 in Copilot) one day before earnings. Sources: Seattle Times/AP, 2026-06-23; Quantel market summary, 2026-06-23; Matterfact AI Capex Tracker, 2026-06-23.
• Row 6c — 2026-06-25, +15.7% (FQ3 reaction). [FACT — price.] FQ3 FY2026 (reported 6/24 AMC): revenue $41.46B vs ~$35.69B consensus; adj. EPS $25.11 vs ~$20.49; adj. gross margin 84.9%; Q4 guided to $49–51B revenue vs ~$43.24B consensus and ~86% gross margin. [FACT — company results as reported by press.] Source: FXLeaders, 2026-07-01 (figures block); Vittarthi market wrap, 2026-06-25.
• Row 7a — 2026-07-01, −10.6% (and −5.5% 7/2). [FACT — prices.] Drivers [INTERPRETATION]: a newly filed U.S. class-action suit naming Micron, Samsung and SK Hynix alleging conventional-DRAM supply restriction/price manipulation; announced Korean capacity additions reviving oversupply fears; profit-taking days after record highs. A same-day GM automotive memory supply agreement did not offset. Source: FXLeaders, 2026-07-01/02.
• Row 7b — 2026-07-07, −4.7%. [FACT — price.] Driver [INTERPRETATION]: Samsung’s quarterly print exceeded Nvidia/Apple profits but fell short of elevated AI expectations — Samsung fell ~8% and the memory complex (MU, SanDisk, WDC, Seagate) sold off in sympathy. Source: SIAM market note, 2026-07-07; Pluang news feed, 2026-07-02.
• Row 7c — 2026-07-15, −8.0%; 2026-07-16, −5.7%. [FACT — prices.] Drivers [INTERPRETATION]: ASML guided EUV lithography sales below consensus, reversing an initial semis rally and weighing on the whole complex (7/15); the tape was simultaneously risk-off on US–Iran escalation (strikes, Strait of Hormuz threats, Brent >$84–86). Continuation selling on 7/16. Sources: Newsquawk US Market Wrap, 2026-07-15; Univest, 2026-07-14.
• Row 7d — 2026-07-17, −0.5% close but a $804.00→$903.93 intraday round trip on 62.8M shares. [FACT — price/volume.] Driver [INTERPRETATION]: capitulation-then-bid session after three straight down days; Chinese financial press described a broad memory-chip rebound that evening. Source: 中国基金报 (China Fund News) headline via NetEase news feed, 2026-07-18 (“memory chips stage a collective counterattack overnight”) — headline observed in search results; direct URL not stable, treat as low-weight corroboration only.

Weekly shape since the prior report [FACT — CSV]: wk of 6/15 +15.5% → wk of 6/22 −0.2% → wk of 6/29 −13.8% → wk of 7/6 +0.4% → wk of 7/13 −13.3%. Two-sided, ±10–15% weekly swings in both directions — the signature of a crowded, high-beta leader being repriced, not of orderly distribution or accumulation. [INTERPRETATION]

1. Executive Summary

This is an update to our 2026-06-09 initiation on Micron Technology, written five weeks later into one of the strangest stretches in the company’s history: the stock is down 9% since that report and 32% from its all-time high, set on the best quarter the company has ever printed.

The print. FQ3 FY2026 (ended 2026-05-28, reported 2026-06-24): revenue $41.46B (+73.8% sequentially, +345.7% year-over-year), gross margin 84.6% (from 74.4% the prior quarter and 37.7% a year ago), operating margin 80.4%, diluted EPS $24.67, free cash flow ~$17.6B. Guidance for FQ4: revenue ~$50B, gross margin ~86%, EPS ~$31 — which would put FY2026 at roughly $129B of revenue and ~$73 of EPS, against $15.5B of revenue and a $5.8B loss as recently as FY2023. Nine-month net income of $47.3B is more than the company earned in its previous decade combined. The surge remains what it was in June: almost entirely price — DRAM ASPs rose low-60s% sequentially on low-single-digit bit growth.

The contract layer — the genuinely new thing. Since the prior report, Micron disclosed 16 take-or-pay Strategic Customer Agreements covering ~25% of revenue, with ~$100B of minimum-price cumulative remaining performance obligations, ~$22B of customer deposits (~$18B cash), ceilings near current market prices and floors that management states carry gross margins “well above our peak quarterly margins in any past cycle.” If the floors transact as written, memory’s through-cycle earnings floor has moved up several-fold — the first contractual crack in 40 years of pure boom-bust. If they renegotiate the way 2023’s LTAs did, they are marketing. That is now the central empirical question of the stock, and it is only ~25% of revenue either way; the other 75% is as cyclical as ever.

The de-rating — the market’s verdict. The stock made its all-time high of $1,255 the day after the print and then fell 32% in three weeks, while sell-side consensus raised FY2027 EPS estimates ~32% to $150.77. The forward multiple on that estimate compressed from ~10.9× to 5.6×. The tape is no longer pricing persistence of peak earnings at all; the sell side (89% Buy, zero Sells, targets $1,550–1,750) is pricing full persistence. The embedded expectation at $849: the cycle is not abolished, but its floor is ~4–8× any prior cycle’s — normalized EPS must hold around $57–70 against a prior-cycle peak of ~$8.

The supply clock. All three DRAM producers plus China’s CXMT are now building into a dated 2028–29 delivery window: Samsung pulled its ~$82.7B P5 twin-fab forward to a July 2026 groundbreaking (the prior report’s pre-specified amber flag), SK hynix raised ~$26.5B in a US listing, Micron raised its own program to >$250B and FY2027 capex above a $40B run-rate, and CXMT approaches ~350k wafers/month. Third-party models still have 2027 tighter than 2026 — the glut is scheduled, not present.

Business quality. Unchanged from the prior report: an excellent operator of a structurally cyclical business — #3 in DRAM (~22%), #2 and diluting in HBM as Samsung re-entered (~10% HBM4 share assumption at Rubin vs. SK hynix ~60%), through-cycle ROIC historically around the cost of capital, no durable moat under either the share-stability or ROIC test. The SCA layer is a real but modest upgrade to demand captivity, not a moat.

Capital allocation. The split verdict persists with a positive tilt: $9.4B of debt repaid at the peak (net cash $24.4B), zero peak-chasing buybacks in FQ3 (CHIPS-gated until December 2026, when management intends to return “100% of excess cash”), and capex now substantially pre-funded by customers and governments. Against that: another capex step-up into the dated 2028 supply wave, no ROIC gate in executive pay, and insiders selling ~$91M in five weeks at $983–$1,192 with zero open-market buys.

Where that leaves the debate. The prior report’s falsification architecture still governs. The bull’s master tripwire — the first sequential DRAM ASP decline — has not fired; pricing is decelerating but still rising. The bear’s tripwire — sustained ASP strength with supply discipline — now holds through FY2027 on every third-party model, with the SCA floor as new structural evidence. The risks have rotated from demand (settled) to supply (scheduled), law (a DRAM price-fixing class action filed June 25), and positioning (a momentum regime that rolled over a month ago against the most anti-low-vol stock in the factor universe). The sections below argue each point from the evidence.

2. Business Overview

What Micron does. Micron designs and manufactures semiconductor memory and storage. Two product families carry the P&L:

• DRAM (dynamic random-access memory) — the high-speed working memory that sits next to a CPU or GPU. DRAM is now $31.3B of $41.5B in FQ3 FY2026 revenue (75.6%), +343% YoY, and the overwhelming driver of profit [FQ3’26 10-Q, Note 14 / MD&A p.28]. Within DRAM the strategically critical product is HBM (high-bandwidth memory) — vertically-stacked DRAM dies sold at a large premium into AI accelerators (Nvidia, AMD). HBM carries a ~3× wafer-area “die penalty” versus DDR5, so every wafer diverted to HBM tightens conventional DRAM supply — the mechanism behind the current price supercycle [prior report, 2026-06-09; corroborated by industry pricing]. DRAM also includes DDR5 server/PC memory, LPDDR (low-power mobile and, increasingly, server), GDDR graphics memory, and CXL-attached memory.
• NAND flash — non-volatile storage (data-center SSDs, embedded storage for phones and autos). NAND was $9.9B in FQ3’26 (24.0% of revenue), +361% YoY [FQ3’26 10-Q, Note 14]. Structurally the weaker, more fragmented franchise — but in this market even NAND printed ASPs up “mid-80s percent” sequentially [FQ3’26 10-Q, MD&A p.28]. A small NOR business (~$185M) rounds out the book.

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

How it makes money — and what just changed. Micron earns the spread between the market-clearing price of a memory bit and its cost to manufacture it. Cost-per-bit falls through process-node shrinks; price-per-bit is set by supply and demand in a near-commodity market. Micron is therefore a price-taker competing on cost, with extreme operating leverage on a fixed cost base — FQ3’26 is the cleanest demonstration on record: revenue rose +74% QoQ while COGS rose only +4.8% ($6.1B → $6.4B), taking gross margin from 74.4% to 84.6% in a single quarter [FQ3’26 10-Q, pp.5, 28–30]. That leverage is symmetrical: the same flat cost base produced a negative gross margin and a $5.8B net loss in FY2023 [FY2023 10-K]. Until recently there was essentially no recurring revenue — output had to be re-sold every quarter at whatever the market would bear.

That last sentence is what the FQ3’26 disclosures materially amend. Since the June report, Micron has signed 16 Strategic Customer Agreements (SCAs) across data-center and automotive customers: take-or-pay, multi-year contracts with binding volume commitments, typically five-year terms (CY2026–CY2030; three years for automotive) [FQ3’26 call, 2026-06-24]. The signed agreements cover ~20% of DRAM volume and ~30% of NAND volume over the period — roughly 25% of company revenue — with four very large and three medium customers; management targets ~50% or more of revenue under SCAs when the program completes, of which ~40% would carry fixed prices or ceilings at or near CQ2-2026 market prices [FQ3’26 call, 2026-06-24]. The largest agreements cap pricing for existing products at roughly the CQ2-2026 market level while holding a floor through the term — and management claims floor-price gross margins are “well above our peak quarterly margins in any past cycle” [FQ3’26 call, 2026-06-24; FQ3’26 10-Q, MD&A pp.31–32 — management claim, not independently verifiable]. Fourteen of the 16 agreements carry ~$100B of cumulative minimum-price revenue over their remaining terms; ASC 606 RPO at quarter-end was only ~$5B (~$1.7–1.8B recognizable over the next 12 months), rising to ~$100B including agreements signed after quarter-end — management is explicit that RPO is a minimum, not expected revenue [FQ3’26 10-Q, Note 14 / MD&A p.33; FQ3’26 call]. Customers are posting ~$22B of deposits and financial commitments (~$18B in cash deposits) in support, largely refundable toward the back half of the terms [FQ3’26 call]. This is a genuine business-model change: a quarter of revenue — heading toward half — now has contractual volume and price-floor visibility rather than pure spot re-sale. The trade is deliberate: Micron caps its upside on contracted bits (ceilings ≈ CQ2-2026 prices, already below the guided FQ4 run-rate) in exchange for a downside floor it claims sits above every prior cycle’s peak. [INTERPRETATION]

Segmentation — the four business units, FQ3’26. Since the Q4 FY2025 reorganization Micron reports four market-facing units. The FQ3’26 print versus FQ2’26 [FQ3’26 10-Q, Note 17, pp.23–25]:

Segment operating income excludes ~$363M of unallocated costs (mostly SBC) [FQ3’26 10-Q, Note 17]. Three observations. First, every unit’s operating margin expanded 11–16 points sequentially — and the highest-margin unit is now MCBU (mobile/client, 85.7%), historically the most commoditized book in the company. When the consumer-tail segment out-earns the HBM franchise, the profit driver is indiscriminate pricing power, not product differentiation — the fingerprint of a cyclical ASP wave, not a structural mix win [INTERPRETATION]. Second, CDBU doubled sequentially (+103%) on ASPs and favorable mix, and data-center SSD revenue “more than doubled” past $5B in the quarter [FQ3’26 10-Q, MD&A p.29; FQ3’26 call]. Third, CMBU’s 78.4% OM — home of all HBM — is the lowest of the four: management has previously conceded non-HBM DRAM margins currently exceed HBM margins, since HBM CY2026 prices were fixed in advance [prior report, Q1/Q2 FY2026 calls]. MCBU’s revenue rose despite lower bit shipments — pure price [FQ3’26 10-Q, MD&A p.29].

Customers and concentration. Micron serves data center (dominant and fastest-growing), PC, smartphone, auto/industrial, networking and consumer. Roughly half of revenue comes from the top ten customers [FY2025 10-K]. The single largest customer accounted for 10% of 9M FY2026 revenue (primarily CMBU — a large HBM/hyperscale buyer), down from 16% in 9M FY2025 [FQ3’26 10-Q, Note 17]. Concentration is declining as the pricing wave lifts all customer cohorts — a modest improvement versus the June report’s ~13%-of-H1 reading — but HBM demand remains structurally concentrated in a handful of hyperscaler/GPU budgets, and the 16 SCAs deepen the binding to those same buyers [INTERPRETATION].

Manufacturing footprint and the U.S. build-out. Micron fabricates in the U.S. (Idaho, Virginia), Taiwan, Japan and Singapore, with assembly/test in Asia, and remains the only U.S.-headquartered scaled memory maker — a strategic and political asset (CHIPS Act funding, U.S. supply security) [FY2025 10-K]. In March 2026 it acquired Powerchip’s Tongluo, Taiwan fab for $1.8B cash, with shipments pulled in to mid-CY2027 [FQ3’26 10-Q, Note 7; FQ3’26 call]. The U.S. program — Idaho ID1 (first wafers mid-CY2027), ID2 (late CY2028), a New York cluster that broke ground in January 2026, Manassas legacy/auto expansion, plus Singapore HBM advanced packaging (meaningful capacity H1-CY2027) — is now described by management as a >$250B U.S. manufacturing and R&D commitment, up from the ~$200B cited in our June report [FQ3’26 call, 2026-06-24; ASSUMPTION — the $250B figure is management’s; the FQ3’26 10-Q does not restate the program total]. FY2026 capex is guided to ~$27B net of government incentives, with FY2027 quarterly capex above the ~$10B FQ4 rate [FQ3’26 10-Q, MD&A p.32; FQ3’26 call]. Approximately 53,000 employees; headquartered in Boise, Idaho; founded 1978, public since 1984 [FY2025 10-K].

Verdict (Business Overview). The business model is being rebuilt mid-cycle. The commodity-cyclical core described in June — a price-taker re-selling its output at market-clearing prices every quarter — still exists, and FQ3’26’s flat-COGS/84.6%-GM arithmetic shows it at maximum torque. But 16 take-or-pay SCAs now put ~25% of revenue (targeting ~50%+) under multi-year volume and price-floor commitments, with customers posting ~$22B of deposits to hold their place in line — something no prior memory cycle has produced. Micron remains easy to understand and cyclical to its bones; what has changed is that a growing slice of the book is contracted, floor-priced and deposit-backed rather than re-won quarterly. That raises the quality of the franchise without repealing its cyclicality — the ceiling on those same contracts means the next downturn’s floor is being purchased with this boom’s upside. [INTERPRETATION]

3. Industry Dynamics

Structure: the DRAM oligopoly is intact; the question has moved from demand to the dated supply wave. Conventional DRAM share in Q1-2026 remains Samsung ~38–39%, SK hynix ~29%, Micron ~22% — unchanged from the June 9 baseline — with China’s CXMT at ~8% and rising but still commodity-bound. [FACT — Counterpoint via AInvest, 2026-07-10] The three-player structure that underpinned the June report’s “structurally improved but still cyclical” verdict is not what changed over the past five weeks. What changed is that (a) pricing kept rising but began decelerating for the first time, (b) the 2028–29 capacity wave stopped being a forecast and became a construction schedule with groundbreaking dates, and © the market started pricing (b) over (a).

HBM economics remain the structural tightener — and HBM4’s price anchor is now hard. One HBM bit consumes ~3× the wafer area of a standard DRAM bit (larger die, TSVs, stack-yield loss), so every wafer diverted to HBM removes roughly three wafers of commodity DRAM supply. That die penalty is still the mechanical reason conventional DRAM is tightening even as HBM ramps: Samsung is allocating much of its P4 cleanroom to HBM, which removes conventional supply into 2027. [FACT — DigiTimes, 2026-05-26; Studio Global, 2026-06-24] The first hard HBM4 price point is now on the record: SK hynix finalized 2026 HBM4 price and volume with Nvidia at >50% above HBM3E (finalized 2026-06-05). [FACT — SK hynix via industry press, 2026-06-05] A >50% generation-over-generation price increase in a year when Samsung re-entered and qualified at both Nvidia and AMD is the strongest single datum that the premium-tier pricing structure survives intensified competition. [INTERPRETATION]

The pricing state: still up everywhere, but the second derivative has turned. Micron’s FQ3 (May quarter) realized DRAM ASPs rose low-60s% QoQ and NAND ASPs mid-80s% QoQ on only low/mid-single-digit bit growth — the record quarter was almost entirely price. [FACT — Micron FQ3 FY2026 call, 2026-06-24] Looking forward, every category is still guided up but at decelerating rates: Q3-2026 conventional and server DRAM contract +13–18% QoQ after ~+50% in Q2 (TrendForce, 2026-07-04/09); DDR4 >+50% QoQ on legacy phase-out (DigiTimes, 2026-07-08); NAND +10–15% QoQ (TrendForce, 2026-07-07). Management’s own framing concedes the moderation: FQ4 gross-margin guidance of ~86% (vs 84.9%) “reflects a meaningful moderation in the rate of price increases” (CFO Murphy, 2026-06-24). The analytically important point: TrendForce attributes the deceleration to buyer resistance and demand destruction at the consumer edge (Apple raised Mac/iPad prices citing memory cost, 2026-06-25) and to LTA lock-ins — explicitly not to improved supply. Server RDIMM bit supply grows only ~15–20% in 2027, far below server demand growth. [FACT — TrendForce via Sina/IT之家, 2026-07-09] Decelerating-on-resistance is a different animal from decelerating-on-supply; the former is a price-level phenomenon, the latter is the cycle killer, and the cycle killer is not yet in the data. [INTERPRETATION]

TAM: forecasts revised up since June. Total memory revenue is tracking ~$552B for 2026E rising toward ~$843B in 2027E, with DRAM alone ~$404B in 2026 and third-party estimates near ~$400B+ by 2027. [FACT — TrendForce baseline; GlobeNewswire DRAM deep-dive, 2026-04-16] More important than the level: Goldman (2026-06-05) now models conventional DRAM, NAND and HBM supply-demand tighter in 2027 than in 2026, with tightness extending into 2028; UBS’s work points the same direction. [FACT — AASTOCKS/Goldman, 2026-06-05; UBS, 2026-07-07] The June report treated “2027 tighter than 2026” as a bull conjecture; it is now the consensus model. That is supportive of fundamentals — and a red flag for expectations, because consensus is now positioned for something the industry has never delivered. [INTERPRETATION]

The supply side is now a dated construction schedule — the single biggest delta since June 9. The baseline flagged “all three expanding into the peak” as a forecast. It is now on the calendar:

• Samsung: pulled P5 Fab 2 groundbreaking forward ~6 months into July 2026 (this month); the P5 Fab 1+2 pair carries a ₩120T (~$82.7B) commitment, with 70+ litho tools (~20 EUV) already ordered for 1c DRAM/HBM and tool install from Q2-2027. Output ~2028. [FACT — NineScrolls, 2026-05-13; IT之家/Sedaily, 2026-04-07]
• SK hynix: racing Yongin’s first cleanroom for 2027; scaling 1c wafers ~20k → 160–190k/month by end-2026 (8–9×); funded in part by a ~$26.5B Nasdaq ADR raise (2026-07-10) — the largest-ever US listing by a foreign company — earmarked for EUV, packaging and Yongin. [FACT — AInvest, 2026-07-10; IBD, 2026-07-09]
• Micron: raised its US program to >$250B through 2035 (from ~$200B) and poured first concrete at Clay, NY a quarter-plus ahead of schedule (2026-07-09); FY2027 capex guided above the ~$40B annualized FQ4 run-rate, weighted to construction. [FACT — GlobeNewswire, 2026-07-09; FQ3 call, 2026-06-24]
• CXMT: ~350k WSPM by end-2026 (~91% of Micron’s ~385k WSPM), per SemiAnalysis/Citrini. [FACT — Tom’s Hardware, 2026-07-16]
• Industry-wide: SEMI projects 300mm memory WFE spend >$50B for the first time in 2026 ($52B, +29% YoY), +11% to $57B in 2027. [FACT — SEMI via Futunn, 2026-07-09]

Essentially every dollar of this lands as sellable bits in 2028–2029. The mitigants are real — near-term “expansion” is mostly HBM conversion that subtracts conventional wafer supply through 2027, and 2026–27 HBM is sold out — but they only push the problem out; they do not cancel it. [INTERPRETATION]

China: scale faster, scope unchanged, new political overhang. CXMT’s Q1-2026 revenue was ~$7.3B (+700% YoY) and it is preparing a ~$8.5B STAR IPO at a ~$85.5B implied valuation. [FACT — Crypto Briefing, 2026-07-13; 247wallst, 2026-07-15] But its ~8% share is concentrated in commodity/consumer DRAM — the segment the Big 3 are abandoning anyway — and it is not yet in qualified HBM (HBM3 mass production targeted end-2026, unproven). [INTERPRETATION — mix point per Startup Fortune, 2026-05-23] Two new overhangs: Apple is lobbying Washington for clearance to source from CXMT/YMTC, with House China Committee leadership publicly pressing to block it [FACT — Tech Wire Asia, 2026-06-30; Crypto Briefing, 2026-07-16]; and reports (2026-07-15) that Washington is weighing tighter unilateral export restrictions on HBM — which would hit Micron’s China-facing HBM revenue directly. [FACT (report) — Proactive, 2026-07-15] The Apple channel-validation signal cuts both ways: it legitimizes CXMT as a supplier, and it confirms that the largest memory buyer on earth is actively looking for an exit from oligopoly pricing. [INTERPRETATION]

Marathon capital-cycle framing: the cycle is behaving exactly as the framework predicts — with a longer fuse. Record industry margins (Samsung ~52% operating margin in Q2; Micron 81.2% in FQ3) are attracting record capital ($52B WFE, +29%), which is precisely the supply-side response the capital-cycle framework says ends every boom. What is genuinely different this time is the fuse length and the contract layer: 5-year-plus LTAs with 60–70% of volume/price locked (UBS, 2026-07-07), Micron’s SCAs, and HBM’s die penalty have converted more of the up-cycle into contracted, forward revenue than in any prior cycle. The industry has, in effect, publicly scheduled the end of the shortage for 2028 — and the equity market has begun trading to that date, not to the next two quarters of prints. [INTERPRETATION]

Verdict (Industry Dynamics): structurally attractive through FY2027; scheduled oversupply 2028–29 — more attractive near-term and less attractive long-term than at the June 9 baseline. The near-term case is better evidenced than five weeks ago: pricing up in every category, the deceleration demand-side not supply-side, 2027 modeled tighter than 2026 by Goldman/UBS/TrendForce, HBM4 priced >50% above HBM3E, and falsification tests (a) sequential-ASP-decline and © inventory-build both untriggered. The long-term case is worse: test (b) — a major competitor announcing aggressive greenfield capacity — is flashing amber on Samsung P5, and the 2028–29 convergence of Samsung P5 ×2, SK hynix Yongin/M15X, Micron NY/Idaho/Hiroshima/Tongluo and CXMT’s 350k WSPM is now a matter of construction schedules, not conjecture. The honest base case: the up-cycle is intact through 2027 with unusually high revenue visibility; the industry’s own capex commits it to oversupply risk concentrated in 2028–2029, with CXMT the swing factor at the commodity tail. The cycle has not been repealed; it has been dated. [INTERPRETATION]

4. Competitive Position

Greenwald verdict from the baseline — restated and still standing: a contestable edge, not a durable moat. The June 9 report classified Micron as economies-of-scale + customer-captivity of the contestable variety: at the commodity-die layer there is no firm-level moat, returns are set by collective oligopoly discipline rather than individual differentiation, both decisive moat tests (share stability, through-cycle ROIC) fail, and whatever edge exists must be re-won each technology generation. Five weeks of new evidence sharpens that verdict in both directions — a genuinely new contract layer on one side, a real HBM share dilution on the other — but does not overturn it. [INTERPRETATION]

Does the SCA contract layer change the captivity analysis? Modestly — it is the strongest challenge to the June verdict, and it still does not clear the moat bar. Since the baseline, Micron disclosed 16 signed Strategic Customer Agreements: typically 5-year terms (calendar 2026–2030; 3-year for auto), take-or-pay with binding commitments, ceiling prices at current CQ2 market levels and floor prices through the term, covering ~20% of DRAM volume and ~30% of NAND volume — ~25% of company revenue, with ~$100B of cumulative minimum-price revenue and ~$22B of customer deposits/letters of credit behind them (14 of 16 agreements). Management’s target is ≥50% of revenue under SCAs. [FACT — Micron FQ3 FY2026 call, 2026-06-24] Post-quarter, the model extended into automotive: GM (2026-07-01), Ford (2026-07-06), and a Qualcomm/Harman/Tier-1 cluster (2026-07-16). [FACT — company PRs; Reuters, 2026-07-16] Management claims the floor prices support gross margins “well above our peak quarterly margins in any past cycle” — if accurate, that is contractual demand captivity with a margin floor, something no memory maker has ever had, and it directly attacks the “trough returns to losses” half of the failed ROIC test. [INTERPRETATION]

Three reasons it upgrades the analysis only modestly. First, coverage is ~25%, not the business — the remaining ~75% of revenue is still a spot/contract commodity exposed to full cyclicality. Second, the captivity is generation-bound and re-negotiable: the 2022–23 down-cycle demonstrated that long-term agreements in memory are honored in good times and re-traded in bad ones; take-or-pay terms in semiconductors have historically been softened, stretched, or settled rather than enforced through a deep glut. [INTERPRETATION — 2022–23 industry LTA experience; no SCA has yet been stress-tested through a downturn] Third, floors cut both ways: ceilings at CQ2 prices cap Micron’s upside on a quarter of revenue if the shortage extends, and floors set near current price levels could become liabilities if CXMT-driven commodity deflation resets the market far below them — counterparties facing punitive floors in 2028 have every incentive to litigate, renegotiate, or walk, exactly as LTA counterparties did in 2023. [INTERPRETATION] Net: the SCA layer is a real upgrade to through-cycle downside protection and revenue visibility — captivity purchased with price ceilings — but it is a five-year contract portfolio, not a compounding moat. The captivity must still be re-won at renewal, and the first renewal arrives precisely when the 2028–29 supply wave does.

The HBM pecking order changed since June 9 — against Micron. The baseline framed HBM as “SK hynix ~62%, Micron #2 at ~21%.” That frame is stale. Samsung has now fully qualified HBM4 at both Nvidia and AMD, shipped ~$1B of HBM4 in its first four months, and is taking the #2 slot on Nvidia’s Rubin platform; UBS’s HBM4/Rubin share assumptions are SK hynix ~60% / Samsung ~30% / Micron ~10% — against Micron’s ~21% HBM3E share. SK hynix remains the leader at ~60–70% (Q1-2026 HBM revenue share ~70% per TrendForce estimate; ~56% on other tracker cuts — the estimate range itself is wide). [FACT — AInvest, 2026-07-10; Wallstreetcn/UBS, 2026-04-08; IBD, 2026-07-09] A halving of modeled HBM share in one generation is exactly the “captivity re-won each generation — and sometimes lost” dynamic the baseline described. [INTERPRETATION]

Micron’s HBM posture is deliberate, not accidental — but deliberately smaller. Management stated it plainly: “HBM market share, we strategically are choosing it to be close to our DRAM share… because of the trade ratio of HBM… puts pressure on non-HBM supply in the industry.” (Mehrotra, FQ3 call, 2026-06-24). [FACT] The logic is sound: at a ~3× die penalty, chasing HBM share beyond DRAM share cannibalizes the conventional DRAM supply that is currently printing low-60s% QoQ ASP increases; Micron’s HBM4 12-high ramp is tracking 2× faster than HBM3E with >$1B HBM4 revenue already shipped, so execution is not the constraint — allocation choice is. [FACT — FQ3 call, 2026-06-24] But the consequence stands: on UBS’s numbers Micron is now the #3 HBM supplier in the generation that defines the AI trade, and “we chose to be smaller” is a strategy, not a moat. It is also reversible by competitors — nothing stops Samsung, with the largest wafer base in the industry, from making the opposite choice in 2027. [INTERPRETATION]

Micron’s real edges — updated. (1) Sole US memory maker. This edge appreciated materially since June: the >$250B US program is now paired with explicit administration embrace (the President touting MU publicly, 2026-07-01; Commerce Secretary at the Clay, NY pour, 2026-07-09), CHIPS-linked capital-return relief (restriction lapses 2026-12-09), and a policy tailwind if Washington tightens HBM export controls against Chinese competition. In a strategic industry, political capital is real capital — Micron is the only memory maker that can monetize it. [FACT — GlobeNewswire, 2026-06-30/07-09; NYT, 2026-07-01] (2) 1γ EUV DRAM execution — ramping to mature yields faster than any prior node, still at/near the leading edge with an ASML multiyear EUV agreement concluded. [FACT — FQ3 call, 2026-06-24] (3) Balance sheet: net cash $24.4B after a $4.4B debt paydown in FQ3, with upgrades from all three major agencies — the deepest trough-survival buffer in the company’s history. [FACT — FQ3 call, 2026-06-24]

Where Micron remains disadvantaged — unchanged or worse. Scale: still #3 in DRAM (~22%) behind Samsung (~38–39%) and SK hynix (~29%), and now likely #3 in HBM4 — Greenwald’s scale advantages accrue to the largest player, and Samsung’s conglomerate balance sheet can still “print through” any trough more comfortably than Micron. [INTERPRETATION — as baseline, reconfirmed by share data 2026-07-10] HBM share vs SK hynix: the leader has ~60–70%, Rubin incumbency, HBM4E sampling in 2H26, and is now directly accessible to US investors via its Nasdaq listing (~$26.5B raised, 2026-07-10) — which also funds its capacity. [FACT — IBD, 2026-07-09] Taiwan concentration: Micron’s leading-edge DRAM output is heavily concentrated in Taiwan (with the Tongluo fab adding shipments from mid-2027), a geopolitical single point of failure none of its edges mitigate. [FACT — FQ3 call, 2026-06-24; baseline 10-K risk factors] And customer concentration — the demand side of the SCA book is a handful of hyperscalers; four “very large customers” anchor the 16 agreements. [FACT — FQ3 call, 2026-06-24]

Verdict (Competitive Position): durable advantage — no; stronger franchise within a crowded market — yes, marginally. Weighing the new evidence both ways: the SCA layer is the most credible improvement to Micron’s competitive economics in its history — contracted demand on ~25% of revenue (targeting ≥50%), a claimed margin floor above prior-cycle peaks, and $22B of customer cash commitments — and it meaningfully raises the trough the June report assumed. Against that, the HBM4 share dilution (~21% → ~10% on UBS’s Rubin model) confirms that Micron’s captivity is re-negotiated every generation and this generation it ceded ground to Samsung; and the >$250B build makes Micron the most capex-aggressive of the three — the highest beta to the very 2028–29 supply wave its own construction is helping create. The moat tests still fail: share is not stable (HBM4 shares moved double-digit points in one generation), and through-cycle ROIC is unproven until an SCA floor survives an actual downturn. Micron is a better-protected #3 than it was on June 9 — but still #3, in an industry whose schedule now shows the crowd arriving in 2028. [INTERPRETATION]

5. Growth History and Forward Opportunities

The trajectory, updated. Growth is real, steep, and almost entirely organic (node transitions and pricing, not M&A). The cycle in one table [FACT — FY2021–FY2025 10-Ks; FQ3’26 10-Q p.5; FQ4 guide per FQ3’26 call, 2026-06-24]:

DRAM revenue has gone $10.98B (FY23) → $28.58B (FY25) → $70.1B in 9M FY2026 alone [FQ3’26 10-Q, Note 14; prior 10-Qs]. FY2026 is tracking to ~3.5× FY2025 — and FY2025 was itself a record. Nothing in Micron’s history remotely resembles this slope; the June report’s caution that “none of the incremental revenue is defensible unit growth” must be re-tested against the FQ3 data. It survives the test.

The composition is still price, not volume — emphatically. The FQ3’26 10-Q decomposes the quarter [FQ3’26 10-Q, MD&A p.28]:

• QoQ (FQ3’26 vs FQ2’26): DRAM ASPs up low-60s% with bit shipments up low-single-digit %; NAND ASPs up mid-80s% with bits up mid-single-digit %. Revenue +74% QoQ on ~5% bit growth — the increment is ~95% price.
• YoY: DRAM ASPs up ~260% on bits +low-20s%; NAND ASPs up ~310% on bits +low-double-digit.
• 9M FY2026 vs 9M FY2025: DRAM ASPs ~+140% on bits ~+30%; NAND ASPs ~+130% on bits +low-20s%.

Gross margin has gone 37.7% (FQ3’25) → 56% (FQ1’26) → 74.4% (FQ2’26) → 84.6% (FQ3’26) on essentially flat COGS [FQ3’26 10-Q, pp.5, 28]. Price-led growth is the least durable kind: the same ASP lever that added ~$65B of annualized revenue in nine months subtracted ~$15B of revenue in FY2023. That is not a forecast; it is the mechanism. [INTERPRETATION]

Data-center mix is the structural part of the story. Data-center revenue exceeded $25B in FQ3’26 — >60% of total revenue, annualizing above $100B — up from ~56% of FY2025 revenue and a low-30s% share historically [FQ3’26 call, 2026-06-24; FY2025 10-K]. Data-center SSD revenue passed $5B in the quarter, more than doubling sequentially [FQ3’26 call]. The mix shift is genuine and persistent: AI infrastructure demand has durably re-pointed the customer base toward hyperscalers, and the SCAs formalize that relationship. This is the strongest evidence that through-cycle demand quality has improved since June — even though the pricing of that demand remains cyclical. [INTERPRETATION]

HBM — execution on schedule, disclosure still thin. Management reports over $1B of HBM4 revenue already shipped, with the HBM4 12-high volume ramp tracking twice as fast as HBM3E 12-high and mature yields expected significantly faster [FQ3’26 call, 2026-06-24]. HBM share strategy is deliberately held “close to our DRAM share” (~20–25%) because the ~3× die penalty pressures non-HBM supply industry-wide [FQ3’26 call]. Notably absent from this call: any “sold out for CY2026/2027” language, a discrete HBM revenue figure, or HBM TAM/share percentages — prior calls carried the sold-out framing; its omission here is noted, not interpreted. [OPEN QUESTION — whether HBM CY2027 pricing/share negotiations are less favorable than the silence implies.] HBM remains housed inside CMBU with no standalone disclosure; CMBU’s $13.8B (+78% QoQ) is the closest observable proxy [FQ3’26 10-Q, Note 17]. Singapore advanced-packaging capacity comes online H1-CY2027 [FQ3’26 call].

Forward opportunities (each a management hypothesis, not evidence):

• SCA coverage expansion. From ~25% of revenue contracted today toward ~50%+, with ~40% carrying fixed prices or ceilings near CQ2-2026 levels; 14 of 16 signed SCAs embed ~$100B of minimum-price cumulative revenue, and ~$22B of customer deposits/commitments (~$18B cash) back them [FQ3’26 call, 2026-06-24]. If completed, this is the single largest forward change in Micron’s revenue quality — a contracted, floor-priced core under the spot book.
• New capacity. ID1 (Idaho) first wafers mid-CY2027; ID2 late CY2028; Tongluo, Taiwan shipments mid-CY2027 (pulled in ~a quarter, second EUV-capable cleanroom started); Singapore HBM packaging H1-CY2027; next-gen DRAM/NAND nodes in volume production 2H-CY2027 [FQ3’26 call]. FY2027 capex guided above the ~$10B FQ4 quarterly rate — implying a >$40B FY2027 run-rate [FQ3’26 call; INTERPRETATION].
• Industry demand. Management’s CY2026 outlook: industry DRAM bit growth low-to-mid-20s% (raised), NAND ~20%; server unit growth raised to high-teens. Demand “continues to significantly exceed supply,” with tightness expected beyond CY2027 and no line of sight to supply catching demand even into 2028 [FQ3’26 call, 2026-06-24 — management framing].
• FQ4 FY2026 guidance: revenue $50B±1B (a record), GM ~86%, EPS $31±1 — with an explicit caveat that the GM guide “reflects a meaningful moderation in the rate of price increases”: prices still rising sequentially, but decelerating [FQ3’26 call]. Note FQ4’26 is a 14-week quarter (53-week year), flattering sequential comparisons by ~8% [FQ3’26 10-Q, Note 1].

Verdict (Growth): price-driven cyclical growth with new contractual ballast — quality upgraded versus June, but not secular. Our June verdict was “high-quality cash generation, low-quality growth durability.” FQ3’26 forces a revision at the margin, and only at the margin. The growth remains overwhelmingly a price phenomenon — DRAM ASP +low-60s% QoQ on low-single-digit bits — and price-led growth is reversible by construction (FY2023 remains the template). What has genuinely improved is the durability framework around the growth: ~25% of revenue is now under multi-year take-or-pay contracts with price floors management claims exceed all prior peak margins, customers have posted ~$22B of commitments to secure supply, and data-center demand above a $100B annualized run-rate is contractual in a way no prior upcycle achieved. Call it what it is: still cyclical growth, but the trough is being engineered upward and roughly half the book is on a path to contracted visibility. This is higher-quality growth than the June report credited — cyclical with ballast, not yet secular. The falsification test is unchanged and mechanical: the first quarter of sequential DRAM ASP decline (management itself now guides to decelerating price increases), or a major competitor’s aggressive greenfield announcement, breaks the upgraded case. [INTERPRETATION]

6. Financial Quality

UPDATE — supersedes Section 6 of the prior report of 2026-06-09. New primary evidence: FQ3 FY2026 10-Q (filed 2026-06-25; quarter ended 2026-05-28) and the FQ3’26 earnings call (2026-06-24). The baseline verdict (“economics improve dramatically with scale at the top, collapse with scale at the bottom; not a through-cycle compounding machine”) is tested below against the new quarter. $ millions unless noted.

What changed since 2026-06-09, in one sentence: the boom went vertical — revenue +73.8% sequentially on essentially flat COGS, gross margin from 74.4% to 84.6%, net cash from ~$6.5B to ~$24.4B — and the mechanism is confirmed as price, not volume. [FACT/INTERPRETATION]

The FQ3’26 print (GAAP)

[FACT — FQ3’26 10-Q, Consolidated Statements of Operations p. 5; MD&A pp. 28–30. Note: management’s “84.9% GM / $25.11 EPS / 81.2% OM” on the call are non-GAAP; the GAAP figures above are the ones that matter.] [INTEREST INCOME NOTE: interest income turned positive net (+215) and interest expense went to ~zero — the balance-sheet transformation now shows up in the P&L. One flag: −321 of unexplained other non-operating expense, not disaggregated in the 10-Q; see quality flags below.]

9M FY2026 [FACT — 10-Q p. 5]: revenue 78,959 (+203% YoY), gross margin 76.6%, operating income 55,589 (70.4%), net income 47,268, diluted EPS $41.40 — nine months of FY2026 have already earned ~5.5× FY2025’s full-year EPS ($7.59).

Segments — the price rise is universal, not an HBM artifact

[FACT — 10-Q Note 17, pp. 23–25; MD&A p. 29. Segment OI excludes −363 of unallocated items, mostly SBC.] Every segment expanded operating margin 11–16 points in one quarter, and the highest-margin unit is now MCBU — mobile/client, historically the most commoditized book. When the commodity segment prints 85.7% operating margins, you are looking at an industry-wide price phenomenon, not a mix story. [INTERPRETATION]

The decomposition: this is a price event

The single most important datum of the quarter [FACT — 10-Q MD&A p. 28; FQ3’26 call, 2026-06-24]: QoQ, DRAM ASPs rose in the “low-60% range” while bit shipments grew low-single-digit; NAND ASPs rose “mid-80% range” on mid-single-digit bits. YoY, DRAM ASPs are up in the low-260% range. Consolidated revenue +73.8% with COGS +4.8% and bits ~flat means the incremental ~$17.6B of quarterly revenue fell almost entirely to gross profit. DRAM is $31.3B (75.6% of revenue), NAND $9.9B (24.0%) [FACT — 10-Q Note 14]. The baseline thesis — economics improve as the AI-driven shortage persists — is confirmed on this datum, not falsified. Management guides FQ4’26 revenue ~$50B ± $1B at ~86% GM, i.e., ASPs still rising but with a “meaningful moderation in the rate of price increases” [management framing — FQ3’26 call; treat as guidance, not evidence]. The direction is still up; the second derivative has turned negative. [INTERPRETATION]

The operating-leverage mechanism — and its symmetry

The margin trajectory across three quarters is the cleanest demonstration of memory operating leverage on record: GM 37.7% → 74.4% → 84.6%, while quarterly COGS moved $5,793 → $6,105 → $6,400 — essentially a fixed ~$6B cost base. Opex fell to 4.2% of revenue (R&D 3.2% + SG&A 1.0%). With costs fixed, every ASP dollar is ~100% gross-margin incremental; that is how revenue +73.8% becomes operating income +106.5% in one quarter. [FACT/INTERPRETATION — 10-Q p. 5]

The mechanism is symmetrical. The same arithmetic that added 10+ GM points per quarter on the way up will subtract them on the way down: when ASP momentum breaks, ~$6B/quarter of COGS does not fall with revenue, and the FY2023 trough (−9% GM, −$6.1B FCF) is the empirical proof [FACT — FY2023 10-K, carried from baseline]. Management itself conceded the marginal mathematics on the call: “incremental price yields less in gross margin expansion” [FQ3’26 call, Q&A]. What is new versus the baseline is a partial circuit-breaker (below). [INTERPRETATION]

Balance sheet — transformation complete

[FACT — 10-Q p. 7, Notes 3–9, MD&A pp. 31–33]

• Cash + investments 30,128 (cash 24,995; ST 1,027; LT marketable 4,106) vs 11,936 at FYE25.
• Total debt 5,722 (current 582 + long-term 5,140) vs 14,577 at FYE25. 9M repayments of 9,380 fully retired the 2028 Notes, 2029 Term Loan A, 2029 A/B and 2030 Notes, with partial prepayments across 2031–2035 maturities. Interest expense is now ~zero.
• Net cash ≈ 24,406 — vs ~6.5B one quarter ago and ~2.6B net debt at FY2025. (Cross-check note: ROIC’s feed folds ~$654M of finance-lease obligations into debt; the filing figure is used here. A further $1.34B of Lease-SPE finance-lease liabilities sits off-balance-sheet per Note 3.)
• Equity 100,724. Receivables 31,025 but trade DSO ≈ 59 days — flat vs FQ2’26 (~59) and FYE25 (~58) [DERIVED: 26,894/41,456×91]. Inventories 8,567, essentially flat since FYE25 with finished goods drawn down to 621 (~7% of inventory) — a sold-out, allocation-driven market, not a build.

Cash flow — real, and enormous

FQ3’26 (derived as 9M minus H1 from the two 10-Qs; ties to ROIC quarterly data): operating cash flow 25,388, capex 7,826 gross (7,093 net of 733 government incentive proceeds), free cash flow ~17.6B on the OCF-minus-gross-capex definition. 9M FY2026: OCF 45,702, capex 19,602 (net 16,613), FCF 26,100 vs 1,596 in 9M FY25. [FACT/DERIVED — 10-Q p. 10; FQ2’26 10-Q] Cash conversion: 9M OCF/net income = 0.97; FQ3 = 0.90 — the gap is the mechanical receivables build against revenue that nearly doubled, not a collection problem (DSO flat). [INTERPRETATION]

Quality-of-earnings flags — carried forward and updated

1. Receivables +$19,953 over 9M [FACT — 10-Q p. 10]. Benign while DSO holds ~59 days; the falsification trigger is unchanged: DSO creeping above ~60 days while ASPs stall.
2. Consideration payable to customers (price-adjustment and return accruals): $3.32B, up from $2.55B at FQ2’26 and $1.19B at FYE25 [FACT — 10-Q Note 14]. Reported revenue is already net of growing give-backs — benign now, but this accrual is a margin cushion that runs in reverse when pricing softens.
3. −$321M of other non-operating expense in FQ3 (vs −98 in FQ2) is not disaggregated in the 10-Q; partially consistent with debt-prepayment costs but composition undisclosed. [FACT — 10-Q p. 5; OPEN ITEM]
4. FQ4’26 has 14 weeks (FY2026 is a 53-week year) [FACT — 10-Q Note 1]: sequential FQ4 optics will overstate underlying momentum by ~8%. Adjust before extrapolating the ~$50B guide.
5. Customer deposits will flatter future reported cash flows. ~$18B of cash deposits is expected (~$10B landing in FQ4), only ~$0.4B collected by 2026-05-28 [FACT — 10-Q MD&A; FQ3’26 call]. Management says these land in financing cash flow and will not touch FCF [call]; either way, headline cash balances and at least one cash-flow statement will look better than underlying economics warrant. The deposits are refundable to customers, weighted to the back half of contract terms — they are a liability in substance. [INTERPRETATION]

The new circuit-breaker: SCA contract economics

The structural change since the baseline is the strategic customer agreements: 16 signed, typically 5-year terms (CY2026–2030), take-or-pay with binding volumes, covering ~20% of DRAM and ~30% of NAND volume (~25% of revenue; target ≥50%). The largest agreements carry price ceilings approximating the Q2-CY2026 market price and price floors through the term, and management states floor-price gross margins are “well above our peak quarterly margins in any past cycle”; 14 SCAs carry ~$100B of minimum-price RPO. [management framing — FQ3’26 call, 2026-06-24; corroborated by 10-Q MD&A pp. 27, 31–32 disclosing $22B of expected deposits/commitments and ~$5B current RPO.] If the floor claim is accurate, ~a quarter of revenue now has a contracted downside margin above anything Micron earned in any prior cycle — a genuine dampener on trough economics. But note what is being traded: ceilings ≈ current spot mean Micron has sold away the upside on contracted volumes from here, and pricing specifics are withheld. [INTERPRETATION]

Verdict — do the economics improve with scale?

Emphatically yes on the way up, mechanically symmetrical on the way down — with the amplitude now deliberately dampened. An 84.6% GAAP gross margin on a flat ~$6B cost base is the best single-quarter economics this company has ever produced, and the sequential DRAM ASP datum (+low-60s%) confirms the improvement is still in force, not yet rolling over. The balance sheet is no longer a risk factor: net cash ~$24.4B, debt cut ~61% in nine months, interest expense ~zero. But the engine is price on a fixed cost base, and that engine runs both directions; the quality flags above (flat DSO, growing price-adjustment accruals, a 14-week FQ4, incoming deposit flattery) define where the first cracks would show. What the baseline could not know is now visible: management is using the SCA structure — ceilings near spot, floors above prior-peak margins — to trade peak torque for trough durability. If the floors are real, the next downturn’s floor is structurally higher than FY2023’s; the operating leverage itself, however, remains exactly as symmetrical as it was on 2026-06-09. [INTERPRETATION]

No BUY/SELL recommendation or price target is made in this section.

7. Capital Allocation

UPDATE — supersedes Section 7 of the prior report of 2026-06-09. New primary evidence: FQ3 FY2026 10-Q (filed 2026-06-25), FQ3’26 earnings call (2026-06-24), Form 4/144 filings through 2026-07-17, and a 2026-07-09 company announcement. The baseline verdict was a split decision — financially conservative, strategically aggressive at the wrong point in the cycle. One baseline error is corrected below (the H1 buybacks were executed at ~$260, not “at all-time highs ~$975”).

(a) The capex supercycle — bigger, and further out

The defining capital decision has been stepped up again. FY2026 capex guidance was raised from “above $25B” (FQ2’26 10-Q) to “approximately $27 billion, net of government incentives”, with FQ4’26 alone guided to ~$10B and FY2027 quarterly capex guided above FQ4 levels — implying a >$40B FY2027 run-rate, with more than half the YoY increase from construction capex pulling in cleanroom capacity. [FACT — FQ3’26 10-Q MD&A p. 32; FQ3’26 call, 2026-06-24 (Murphy)] FQ3 capex paid was $7.83B gross / $7.08B net of $733M incentive proceeds; PP&E purchase obligations rose to $2.93B from $2.10B in February. [FACT — 10-Q p. 10, Note 7]

The build is now concrete — literally: the four-fab Clay, NY project broke ground in January 2026 with first concrete poured more than a quarter ahead of the baseline schedule; Boise ID1 is on track for first wafers mid-CY2027 and ID2 late CY2028; the Tongluo, Taiwan fab (acquired from Powerchip for $1.8B cash, closed March 2026) is retrofitting for meaningful shipments from mid-CY2027 with a second EUV-capable cleanroom started; a multiyear ASML EUV supply agreement (1-delta and beyond) was concluded; next-gen DRAM/NAND nodes enter volume production 2H CY2027. [FACT — FQ3’26 call; 10-Q Note 7] The envelope remains a >$250B US program through 2035 (~$200B manufacturing/R&D plus the 2026-07-09 pledge of up to $3B into the US supply chain, including $500M of strategic financing to GlobalWafers plus a 10-year wafer supply agreement). [FACT — Micron press release, 2026-07-09; not yet in an SEC filing]

The capital-cycle critique from the baseline is not diminished by this — it is amplified. Capex was guided up by ~$2B at what may prove to be the cycle peak, FY27 adds another step-function, and this supply lands into the 2028 window management itself now dates (“even as we expect industry supply to improve gradually in 2028”). Peak-margin capex landing as supply in the year of maximum cycle risk is the canonical memory-cycle error; the only defense is that three rational players, customer pre-commitments, and floors change the payoff. That is a hypothesis, not yet evidence. [INTERPRETATION]

(b) Deleveraging — the windfall went to the fortress

The dominant use of the FQ3 cash gusher was debt extinguishment: $9.38B repaid/prepaid over 9M FY2026 (incl. a $4.3B cash tender in FQ3), cutting total debt from $14.58B at FYE25 to $5.72B; cash + investments ended at $30.1B → net cash ≈ $24.4B, nearly 4× the ~$6.5B at the baseline. Interest expense is now ~zero and all three rating agencies have upgraded (incl. BBB+). [FACT — 10-Q Note 9, MD&A p. 33; FQ3’26 call] This is the conservative-financier pattern asserting itself at the top: deleverage into strength, exactly the opposite of prior cycles where peak capex was debt-funded. It converts the downside case from “balance-sheet impairment” to “returns impairment” — a materially better failure mode. [INTERPRETATION]

© Shareholder returns — zero, by constraint, with a dated gate

• Buybacks: $0 in FQ3’26. The entire 9M program spend remains $650M for 2.5M shares — an average of ~$260/share, all executed in H1 FY2026. [FACT — 10-Q Note 11] Correction to the baseline: the prior report characterized the resumed buybacks as executed “at all-time highs (~$975)”; the filed average price shows they were executed at roughly a quarter of the late-June price ($1,077–$1,192) — accretive, not peak-chasing. The FQ3 pause coincides with the repeated disclosure that repurchases are “subject to … restrictions applicable under our CHIPS Act direct funding agreements” [10-Q Note 11]; ~$2.16B of the $10B authorization remains. The gate lapses 2026-12-09 (second anniversary of the CHIPS agreements), and management states that “over time, we expect to return 100% of our excess cash to shareholders.” [management framing — FQ3’26 call] Intent is not allocation; the test comes after 2026-12-09, at whatever price then prevails. [INTERPRETATION]
• Dividend: $0.15/quarter, re-declared 2026-06-24, payable 2026-07-21 ($171M in FQ3; $437M in 9M). Never cut — but at ~$1,100/share the yield is ~0.05%: a signaling device, not a return of capital. [FACT — 10-Q Note 11]

(d) The supercycle is now substantially pre-funded by customers and governments

The under-appreciated structural change since the baseline: strategic customer agreements carry $22B of expected cash deposits and related financial commitments (~$18B cash deposits; ~$10B landing in FQ4’26) for agreements concluded to date — only ~$0.4–0.5B had been collected by quarter-end. [FACT — 10-Q MD&A pp. 31–32; FQ3’26 call] Add $733M of government incentive proceeds in FQ3 alone ($2.99B in 9M, vs $1.29B a year earlier), up to $6.4B of federal CHIPS grants, the 35% investment tax credit, and up to $5.5B of New York State incentives. [FACT — 10-Q MD&A] Caveats: the deposits are refundable (returned to customers, back-half-weighted) and are a liability in substance; and customer pre-funding is itself cyclical demand signaling at the top. Still, the share of the enlarged capex program resting on Micron’s own balance sheet is materially smaller than at the baseline — the single most anti-fragile feature this memory upcycle has that prior ones lacked. [INTERPRETATION]

(e) Insider behavior — zero buys; ~$91M sold in five weeks at $983–$1,192

Since 2026-06-09: zero open-market purchases (Form 4 code P) by any officer or director; four filers sold or disposed of ~81,300 shares for ~$91.2M. [FACT — Form 4s filed 2026-06-11 through 2026-07-17; Form 144s filed 2026-06-26, 06-30, 07-01]

The planned sales still inform: Mehrotra pre-committed in January to sell into the rally (~10% of direct holdings); Arnzen sold ~29% of her direct holdings at $1,084; and Dugle’s sale carries no 10b5-1 footnote — a discretionary sale by a director at $1,150. Officers and directors collectively own <1% of the company. The asymmetry is the tell: the company last bought shares at ~$260; the insiders are selling at roughly four times that. Whatever management says about durable floors, the people with the best information are monetizing, not accumulating. [FACT/INTERPRETATION]

(f) Compensation — unchanged, still no ROIC gate

No new DEF 14A since 2025-11-25 (next proxy ~Nov 2026); the baseline critique stands untouched: short-term incentives on profitability plus confidential strategic goals, long-term on relative TSR and bit/market-share metrics, with no explicit ROIC or return-on-capital gate — precisely the incentive structure that encourages over-building into a peak. Say-on-pay passed at a soft 84%. One board change: Alexis Black Björlin appointed director 2026-06-09 (board now nine, eight independent). [FACT — DEF 14A 2025-11-25; 8-K Item 5.02, 2026-06-09] Meanwhile share count keeps creeping up — 1,122M (FYE25) → 1,129M outstanding — because equity-comp/ESPP issuance (2M ESPP shares at $92.77; RSUs at a $227 weighted-average grant value) dwarfs the token buyback. [FACT — 10-Q balance sheet, Notes 12–13]

Verdict — has management allocated capital intelligently?

Split verdict, and the split widened. On financing discipline, the record since 2026-06-09 is the best in the company’s cyclical history: $9.4B of debt retired at the peak rather than into the trough, net cash quadrupled to ~$24.4B, zero buybacks chased at $1,100+ (the H1 buys at ~$260 look excellent in hindsight), and a ~$27B capex program substantially pre-funded by ~$22B of customer commitments and ~$3B/year of government incentives. The historical way memory cycles destroy shareholders — a leveraged balance sheet meeting a price collapse — has been substantially engineered away. On per-share alignment, every structural critique from the baseline is intact or worse: insiders accelerated monetization into the highs (~$91M in five weeks, zero buys, one discretionary director sale) at ~4× the price the company itself last paid for stock; another capex step-up was committed into a 2028 supply wave management itself now dates; share count keeps rising because comp issuance dwarfs buybacks; and the incentive plan still rewards bit growth and share with no ROIC gate. The fortress is stronger; the per-share engine still leaks. Whether “return 100% of excess cash” (post-2026-12-09) converts intent into allocation is now the single most important open capital-allocation question. [INTERPRETATION]

No BUY/SELL recommendation or price target is made in this section.

8. Changes and Headwinds — Last Two Years

The five-week delta (2026-06-09 → 2026-07-18). Since the prior report Micron printed the best quarter in its history — and the stock fell. FQ3 FY2026 (reported 2026-06-24) beat consensus by ~16% on revenue ($41.46B vs ~$35.7B; non-GAAP EPS $25.11 vs ~$20.4; non-GAAP gross margin 84.9%) and guided FQ4 ~15% above the Street (revenue $50B ± $1B, GM ~86%, EPS $31 ± $1) [FACT — company PR, GlobeNewswire, 2026-06-24]. The stock rallied +16% to an all-time-high close of $1,255 the next day, briefly overtaking Meta in market cap [FACT — Reuters, 2026-06-25] — and then gave it all back plus more: −32% from the peak into the 2026-07-17 close of $848.95, which is −9.3% below the prior report’s $935.89 reference price [FACT — adjusted price history]. The update to the June memo’s central tension — “the business is structurally better; the stock is priced as if the cycle is abolished” — is that the market has begun resolving that tension on its own: FY27 consensus EPS went up ~32% over the window while the price went down ~32%. Nothing in the five weeks weakened the demand facts; everything about the tape weakened the demand for the stock. [INTERPRETATION]

The two-year arc, in one paragraph. From the FY2023 trough ($5.8B GAAP loss, negative gross margin) through the AI/HBM inflection (2023–24), the HBM-share wobble and tariff-shock round trip (mid-2024 → April 2025, −59%), the steepest up-leg in company history (≈10× in twelve months off the August 2025 low), the $1T market-cap cross (May 2026), and the FQ3’26 blowout and all-time high (June 2026) — every structural upgrade catalogued in Section 8 of the baseline report (net-cash balance sheet, 1γ/G9 technology leadership, CMBU/CDBU resegmentation, CHIPS-funded US footprint) has been confirmed or extended. What has changed is not the arc but where the market believes we are in it. [FACT for the events; INTERPRETATION for the framing]

Changes that strengthen the thesis (all post-baseline):

1. The FQ3 print and FQ4 guide (2026-06-24). Revenue +346% YoY / +74% QoQ; operating cash flow $25.4B; adjusted FCF $18.3B; data-center revenue >$25B in the quarter, a >$100B annualized run-rate; the guide implies sequential acceleration, not deceleration [FACT — company PR; FQ3’26 earnings call, 2026-06-24].
2. The SCA business-model shift — the centerpiece. 16 multiyear take-or-pay Strategic Customer Agreements (5-year terms, CY2026–2030; 4 very large + 3 medium customers), covering ~20% of DRAM volume / ~30% of NAND volume / ~25% of revenue; ~$100B minimum-price cumulative RPO; ~$22B of customer deposits (~$18B cash, unrestricted, returnable late in term); ceiling prices near CQ2 market, floor prices management claims sit “well above our peak quarterly margins in any past cycle” (prior all-time peak quarterly GM ~47%, FY2018) [FACT — Mehrotra/Murphy, FQ3 call, 2026-06-24]. Management targets ~50%+ of revenue under SCA terms eventually; Stifel called the terms a “business model reset” [2026-06-25]. This is the first contractual downside structure in the company’s history. Whether it transacts as written in a down-leg is unproven — see Section 11.
3. SCA extension into automotive. New agreements with GM (2026-07-01), Ford (2026-07-06), and a cluster with Qualcomm, Harman and Tier-1 suppliers for AI-vehicle memory (2026-07-16) [FACT — GlobeNewswire PRs 2026-07-01/07-06; Reuters 2026-07-16]. Extends take-or-pay contracting beyond data center into a segment with 3-year terms and legacy-node (Manassas) supply.
4. >$250B US program + Clay, NY first concrete (2026-07-09). Planned US investment raised to >$250B through 2035; first concrete poured at the New York megafab a quarter-plus ahead of schedule (Commerce Sec. Lutnick, Gov. Hochul attending); a separate up-to-$3B US supply-chain program alongside [FACT — GlobeNewswire / Reuters / CNBC, 2026-07-09]. Read both ways: management confidence in durable demand, and a further step-up in capex intensity.
5. Capex guide raised. FY2026 capex now ~$27B (FQ4 ~$10B), with FY2027 quarterly capex guided above FQ4 levels — implying a >$40B FY2027 run-rate, more than half of the increase from construction [FACT — Murphy, FQ3 call, 2026-06-24]. The baseline’s “capex over-build” risk is being exercised on schedule, funded by record FCF. [INTERPRETATION]
6. Debt paydown to net cash $24.4B. Debt cut $4.4B in FQ3 (including a $4.3B cash tender) to $5.7B; net cash $24.4B; upgrades from all three rating agencies; management commits to returning 100% of excess cash once the CHIPS-agreement restriction lapses on 2026-12-09 [FACT — Murphy, FQ3 call]. The baseline’s “balance-sheet transformation” is now complete; the debate has moved entirely to the P&L side.

Headwinds that emerged or sharpened:

1. Antitrust class action (filed 2026-06-25, disclosed over following days). Garciaguirre et al v Samsung Electronics et al (N.D. Cal., Judge Cousins) names Micron, Samsung and SK hynix; alleges coordinated restriction of conventional DRAM (DDR3/DDR4) output under cover of the HBM pivot; seeks class status, injunction, treble damages [FACT — Invezz, 2026-07-01]. Mitigant: a materially similar 2018 suit was dismissed, affirmed by the 9th Circuit in 2022. Scott+Scott opened a fiduciary-duty investigation 2026-07-06 [FACT — BusinessWire]. It is an overhang on the exact pricing mechanism the bull case celebrates — small probability, large awkwardness. [INTERPRETATION]
2. Burry short (2026-07-01 → 07-03). Scion disclosed an MU short on Substack: “destroyer of capital,” 42-year median ROIC ~4%, extension over the 200-DMA exceeding the dot-com peak, “sold out” HBM claims dismissed as scarcity bias [FACT — Benzinga / Invezz, 2026-07-03]. Intellectually it is the baseline bear case repackaged; practically it gave the tape’s side of the argument a name. [INTERPRETATION]
3. Insider selling, zero buying. ~81,300 shares / ~$91.2M sold in five weeks at $983–$1,192: CEO Mehrotra 40,000 sh / $46.34M @ ~$1,158 (6/26, 10b5-1 plan dated 2026-01-30); EVP Arnzen 40,000 sh / $43.36M @ ~$1,084 (7/1, 10b5-1); Director Dugle 1,300 sh / $1.50M @ ~$1,150 (6/30, no 10b5-1 footnote — discretionary); zero code-P open-market buys [FACT — Form 4 filings, reviewed 2026-07-18]. Both large sales were pre-planned before the rally. The informative datum is not what insiders did but what they did not do — buy — at any point in a 32% drawdown. [INTERPRETATION]
4. Competitor mobilization — the supply response is now funded and dated. SK hynix listed on Nasdaq via ADR (2026-07-10; priced $149, >7× oversubscribed, raised ~$26.5B — the largest-ever US listing by a foreign company), earmarked for EUV, advanced packaging and Yongin [FACT — Invezz / IBD, 2026-07-09/10]. Samsung pulled P5 Fab 2 groundbreaking forward ~6 months to July 2026 (₩120T / ~$82.7B twin-fab program) and is now fully qualified on HBM4 at Nvidia/AMD — UBS’s HBM4/Rubin share estimates run SK hynix 60% / Samsung 30% / Micron 10%, a relative dilution of Micron’s HBM position vs. the baseline’s “#2” framing [FACT — NineScrolls, 2026-05-13; AInvest/UBS, 2026-07-10]. CXMT is preparing an ~$8.5B Shanghai STAR IPO (~$85.5B implied valuation; DRAM share ~8%, roughly tripled YoY per Counterpoint) while Apple lobbies Washington for clearance to source CXMT/YMTC memory [FACT — 247wallst, 2026-07-15; Tech Wire Asia, 2026-06-30]. Separately, Washington is reportedly weighing tighter unilateral export restrictions on HBM — a direct MU revenue risk [FACT (report) — Proactive, 2026-07-15].
5. Second-derivative demand signals. Meta reportedly developing plans to lease out surplus AI training/inference capacity [FACT (report) — Proactive, 2026-07-15]; CoreWeave reportedly exploring financial hedges against a drop in memory costs [FACT (single-source report) — 247wallst, 2026-07-17 — flagged]. Sophisticated buyers behaving as if prices are nearer peak than floor, even as DRAM ASPs sit at 10-year highs with the rate of increase decelerating (Q3 contract +13–18% QoQ vs ~+50% in Q2 — TrendForce, 2026-07-09; management’s own “meaningful moderation in the rate of price increases,” FQ3 call). [INTERPRETATION]
6. The tone arc: euphoric → mixed → turning cautious. Coverage ran euphoric 6/24–25 (“margin king,” overtaking Meta), mixed 6/26–7/8 (sell-the-news despite a beat-and-raise), and turning cautious from 7/9 onward — even good news (the >$250B capex plan, Samsung’s 19× profit print, TSMC’s record) was sold, and the framing shifted from “how high” to “has it peaked” [FACT — press tone read, 2026-07-18]. The sell side never wavered: 45 July analyst opinions, 89% Buy/Strong Buy, zero Sells, PTs clustered $1,550–1,750 (BofA $1,550; HSBC $1,700; TD Cowen $1,600; KeyBanc $1,750) [FACT — Fool / Finbold, 2026-07-11/14]. The divergence between published consensus and the tape is the defining feature of the window — see Section 11.

Verdict — do the changes strengthen or weaken the thesis? Both, and the mix has shifted since June. Strengthening: the business-model argument is now better evidenced than at any point in the company’s history — contracted revenue, forfeitable deposits, floor pricing claimed above all prior peaks, a completed balance-sheet transformation, and 2027 modeled tighter than 2026 by Goldman, UBS and TrendForce [FACT — Goldman/UBS/TrendForce, as cited in Appendix B]. The “cycle is dead” case graduated from narrative to contract for ~25% of revenue. Weakening: everything bearing on the duration of peak economics degraded — the 2028–29 capacity wave is now publicly scheduled and funded on all four fronts (Samsung P5, SK hynix Yongin, CXMT, Micron’s own >$40B FY27 capex); pricing momentum decelerated; a legal challenge now hangs over the pricing mechanism itself; insiders sold the peak and bought nothing; and the tape — which led fundamentals at every prior memory top — de-rated the stock 32% on the best quarter ever printed. [INTERPRETATION]

The honest net read: the five weeks raised the floor of the thesis (the SCA structure plausibly doubles the trough-earnings anchor — Section 10©) while lowering the ceiling (the end of the shortage is on the calendar). The distribution narrowed at both ends, and the market spent five weeks repricing the ceiling. Whether that is healthy discipline or the first leg of the mean reversion the baseline warned about is the question Sections 10 and 11 take up.

This section contains no buy/sell recommendation and no price target. Facts are dated and sourced; all causal attributions are labeled interpretation.

9. Risk Analysis

What dominates now versus June: the de-rating risk has moved from the matrix onto the tape, and it leads. On June 9 the two dominant risks — cyclical ASP reversion and peak-earnings de-rating — were flagged as correlated and prospective. Five weeks later the second has fired without the first: the stock fell 32% from its all-time high while every fundamental datum (pricing, guidance, contract coverage, 2027 supply models) improved. That is the defining feature of the current risk posture — the market is no longer paying for peak earnings regardless of how long the peak lasts, and the bear catalysts it is trading on (Burry’s short, the antitrust filing, CoreWeave hedging, Meta leasing, CXMT’s IPO) are all second-derivative or legal/narrative events, not demand breaks. For the report’s purposes the de-rating row is now descriptive, not predictive: any holder at $935 has already absorbed the drawdown the June matrix assigned “High/High,” and the open question is whether the multiple stabilizes at a de-rated plateau or continues to compress toward through-cycle levels while fundamentals stay strong. [INTERPRETATION]

The risk mix has rotated from demand risk to supply, law, and positioning risk. In June the live threats were demand-side: would AI capex hold, would ASPs roll, would inventories build. Today the demand-side falsification tests are less triggered than they were (ASPs up, inventories below alarm, 2027 modeled tighter), while three non-demand risks have escalated materially: (1) supply — Samsung P5’s pulled-forward groundbreaking converted falsification test (b) to amber, the first criterion to degrade since the baseline, and put a date (2028) on the end of the shortage; (2) law — the N.D. Cal. price-fixing class action is a new, unmodelable tail risk that, whatever its merits (the 2018 precedent favors the defendants), invites discovery into exactly the “HBM pivot as coordinated restriction” narrative that underpins current pricing; and (3) positioning/structure — HBM4 share dilution to ~10% on Rubin and the SCA book’s concentration in four very large customers mean Micron’s franchise quality is more questioned now even as its contracted revenue is larger. The residual demand risk (AI-capex digestion) is real but second-order: Meta/CoreWeave items are reports of hedging and leasing, not cancellations. [INTERPRETATION]

Net risk posture: lower probability of near-term fundamental break, higher certainty of a dated fundamental wall. The SCA floor layer and $24.4B net cash position genuinely truncate the left tail relative to any prior Micron cycle — ruin is off the table and even the trough case now has a contracted margin floor (claimed, untested) under a quarter of revenue. But the distribution’s center has shifted adversely: the equity has begun de-rating on schedule risk that is now construction-dated rather than hypothetical, the premium tier is more contested, and two legal/policy vectors (antitrust, HBM export curbs) carry headline risk in both directions. The matrix’s top row for the next update is no longer “will ASPs fall” — it is the Q4-2026 contract reset (first flat print = falsification test (a) trips), Samsung P5/P6 capex disclosures, and the Apple-CXMT lobbying outcome. [INTERPRETATION]

10. Valuation Discussion (Embedded Expectations)

Update to Section 10 of the baseline (2026-06-09, $935.89). All per-share math is rebuilt at the 2026-07-17 close of $848.95 using ~1.15B diluted shares (FQ3’26) → market cap ~$976B; net cash $24.4B / debt $5.7B → EV ~$952B. yfinance’s market cap ($958.8B) uses ~1.13B basic shares — a ~2% discrepancy, flagged; we use diluted throughout. No price target; no recommendation.

(a) Multiples — now vs. baseline.

[FACT — AZI valuation_index pulled 2026-07-18; yfinance quote/stats/eps_trend accessed 2026-07-18; FQ3’26 figures per company PR 2026-06-24]

The picture the multiples paint has inverted since June. In June the stock was expensive on every slow-moving anchor (book, sales) and cheap only on peak forward EPS. The anchors have barely moved (P/B 13.4×, 98.5th percentile — still extreme), but the fast-moving multiples have collapsed because the denominators doubled while the price fell: trailing P/E 44.8× → 19.2×, EV/EBITDA ~27× → ~14×, FY27 forward P/E ~9× → 5.6×. The de-rating is not multiple compression of fear on static earnings; it is the market refusing to pay up for earnings that have already been revised into the numbers. A practical warning on data hygiene: headline aggregator multiples (AZI’s 40.0× trailing P/E, 16.6× P/S) are still computed off pre-print denominators and flatter the de-rating in the wrong direction — any multiple quoted on MU this month must be checked against a post-FQ3 denominator before use. [INTERPRETATION]

(b) The multiple-compression spine: −32% price, +32% estimates, multiple halved.

The stock made its all-time high ($1,254.81 intraday, 2026-06-25) on the FQ3 print — a ~16% revenue beat with FQ4 guided ~15% above the Street — and has since fallen 32.3% [FACT — adjusted price history]. Over the same window consensus moved the other way: FY26 EPS estimate $60.22 (30d ago) → $73.39 (+22%); FY27 estimate $114.59 → $150.77 (+32%); versus 60–90 days ago the FY27 estimate is up ~47–50% ($100.53–102.58 → $150.77) [FACT — yfinance eps_trend, accessed 2026-07-18].

On FY27 consensus, the market paid ~10.9× at the peak and pays 5.6× now — the forward multiple compressed ~47% in three weeks while the earnings estimate it is applied to rose ~31%. The entire post-print upgrade cycle — the largest positive revision wave in the company’s history — has been absorbed by de-rating, not by price. [Multiple compression is arithmetic FACT; its meaning is INTERPRETATION.]

This is the classic cyclical-top signature the baseline warned about — a cyclical’s forward P/E is lowest at the peak — now observed in real time, with one genuine novelty: the compression has taken MU’s forward multiple below the point where prior-cycle bears said it “should” trade (8–10× forward), while the earnings base keeps rising. At 5.6× FY27 consensus the tape is no longer pricing persistence of peak earnings at all; it is pricing a large, near-term erosion of them. That is materially different from June, when the embedded expectation was persistence. [INTERPRETATION]

© Embedded-expectations arithmetic at $849 — what must normalized EPS be?

The reverse-DCF-style question, restated. Market cap ~$976B, EV ~$952B. Capitalize that at a mid-cycle multiple appropriate for a high-quality but still-cyclical franchise (12–15× normalized net income), and the price requires normalized net income of ~$65–81B — normalized EPS of roughly $57–70 on 1.15B shares. Equivalent perpetuity framing: EV $952B at a 10% discount rate needs ~$95B of normalized annual free cash flow — versus FY26 actual FCF of roughly $38–45B (stockanalysis FY26 FCF estimate $37.8B, 2026-06-18; 9M FY26 operating cash flow less ~$27B capex) and FY27 consensus-implied FCF well above $100B. [INTERPRETATION — author’s arithmetic from verified inputs]

Benchmarks for that $57–70 normalized-EPS requirement:

• Prior-cycle reality: FY2022 (a peak year) EPS ~$7.8; FY2025 $8.29; FY2023 a $5.8B loss. The best single year in company history before this cycle earned ~1/8th of what the price now normalizes. [FACT — filings/stockanalysis]
• FY27 consensus: $150.77. The price at $849 capitalizes ~40–45% of FY27 consensus EPS as the permanent base — the market is underwriting roughly a 55–60% erosion from the FY27 peak, held forever. [INTERPRETATION]
• The baseline’s embedded expectation was “abolition of the memory cycle.” At $849 that is no longer true. The embedded expectation has softened to: the cycle is not abolished, but its floor has moved up several-fold — through-cycle earnings must hold at ~4–8× any prior-cycle peak. [INTERPRETATION]

How much does the SCA floor change the normalized-earnings debate? Since the baseline, the contractual landscape is materially better evidenced: 16 take-or-pay contracts, ~25% of revenue covered, $100B minimum-price cumulative RPO, $22B customer deposits, and management’s claim that contract floors sit at gross margins “well above peak quarterly margins in any past cycle” (prior all-time peak quarterly GM ~47%, FY2018) [FACT — FQ3 call, 2026-06-24; TD Cowen’s $10–12/GB floor estimate vs. mid-$20s current server DRAM is an analyst estimate, unconfirmed]. Rough arithmetic [INTERPRETATION]: if ~25% of FY26-implied revenue (~$129B) ≈ ~$32B/yr is floored at ≥~50% GM, the contracted book alone generates ~$16B+ of gross profit — versus total company opex of ~$6–7B — i.e., the contracted floor plausibly covers opex and tax by itself, putting a rough trough-EPS anchor in the mid-to-high single digits from the contracted quarter of the business alone, before any contribution from the other 75% at whatever mid-cycle pricing then prevails. The baseline’s bear anchor of $8–15 normalized EPS assumed no floor; a defensible floor-raised bear anchor is now ~$15–25. The floor roughly doubles the trough-earnings anchor — but it protects only ~25% of revenue, so the normalized-earnings debate is still, arithmetically, a debate about the uncontracted three-quarters. [INTERPRETATION]

(d) Scenario analysis — updated (illustrative zones, deliberately wide, no point estimate, not price targets).

Same construction as the baseline, inputs updated for: FQ3 actuals; FY26 implied EPS ~$72–73 (34-analyst consensus $73.39); FY27 consensus $150.77; the SCA floor ($100B RPO / $22B deposits / ~25% coverage); 2027 modeled tighter than 2026 (Goldman/UBS/TrendForce, as cited in Appendix B); and a now-dated 2028–29 capacity schedule (Samsung P5 pulled forward to July 2026 groundbreaking; SK hynix Yongin/M15X funded by the July 2026 US listing; CXMT ~350k WSPM).

[INTERPRETATION — author’s scenario construction from filed financials, consensus data, and industry research. These are illustrative valuation zones for thinking about asymmetry — not forecasts, not price targets, and not a recommendation.]

The critical relocation. At $936 the price sat toward the bull end of the baseline distribution. At $849 it sits between base and bull — the de-rating has moved the market’s own pricing from “structural case required” to “structural case partially priced.” Sell-side targets ($1,550–1,750) remain anchored in the bull zone; the tape is now pricing something closer to the base case’s low end on forward earnings and the bear case’s skepticism on duration, simultaneously. [INTERPRETATION]

(e) Comps (2026-07-17 closes; yfinance multiples accessed 2026-07-18 — convenience data, flagged where broken):

[FACT (prices/multiples as sourced) with flagged discrepancies — ROIC MCP 2026-07-17 closes; yfinance accessed 2026-07-18; ROIC aggregator annual EV snapshots for SNDK/WDC predate the boom (FY2025) and were not used.]

Two observations. First, the whole complex now trades on the same peak-EPS discount — SNDK at 6.4× forward, SK hynix at 3.8×, MU at 5.6×: the market is applying single-digit multiples to the entire memory/storage chain’s forward earnings, i.e., it believes none of these earnings persist. Second, MU’s forward P/E now sits ~3× below NVDA’s and AVGO’s while its near-term estimate momentum is the strongest in the group — the market prices MU’s earnings as the least durable in the AI complex. That is precisely the variant-perception question (Section 11). [INTERPRETATION]

(f) What the market is underwriting correctly vs. incorrectly — as of 2026-07-18.

Correctly [INTERPRETATION]:

• That FQ3/FQ4 are peak or near-peak quarters and must not be capitalized at secular multiples. The 5.6× FY27 multiple is the market declining to pay for peak earnings — appropriate cyclical hygiene, and the tape’s −32% on a beat-and-raise is textbook late-cycle behavior.
• That the pricing-momentum data has genuinely decelerated (Q3 contract +13–18% vs ~+50% prior — TrendForce, 2026-07-09; management’s “meaningful moderation in the rate of price increases” — FQ3 call) and that 2028–29 supply is now a dated construction schedule, not a hypothesis.
• That the legal/political overhang (the DRAM price-fixing class action naming MU/Samsung/SK hynix, filed 2026-06-25) is real tail risk to the very pricing mechanisms the bull case celebrates.

Potentially incorrectly [INTERPRETATION]:

• Pricing the SCA floor at zero. The tape is valuing MU as if the 2023 playbook (ASPs −50%+, GM negative) applies symmetrically, while $100B of minimum-price RPO, $22B of forfeitable customer deposits, and take-or-pay terms on ~25% of revenue are contractual facts that did not exist in any prior cycle. The floor may prove softer than advertised — contracts do get renegotiated in gluts — but zero is not the right prior.
• Treating MU as fungible with its own history. Book value is now $63.56/share and compounding at ~$25/quarter; even a full reversion to the historical 2–4× book band no longer reaches the old bear prices — the bear anchor has mechanically risen (hence the scenario-zone revision above).
• Conversely, the sell side is incorrectly symmetric: PTs of $1,550–1,750 capitalize FY27 consensus at ~10–11× as if the floor protected the whole P&L. It protects ~25% of revenue. Both sides are mis-pricing the same 75% — the tape at ~zero terminal value, the Street at ~full.

No price target. No recommendation. The embedded expectation at $849 is stated above in Section 10©: normalized EPS of roughly $57–70, i.e., ~40–45% of FY27 consensus held as the permanent base. Everything in this section is arithmetic and scenario framing around that single number.

11. Variant Perception

Consensus belief now — and the story is that there are two of them. Published consensus (the sell side) is uniformly bullish: 89% of July ratings Buy/Strong Buy, zero Sells, PTs cascading to $1,550–1,750 post-print (BofA $1,550; HSBC $1,700; TD Cowen $1,600; KeyBanc $1,750) — targets ~2× spot — underwritten by FY27 consensus EPS of $150.77, 2027 modeled tighter than 2026, and HBM sold out [FACT — Fool / Finbold, 2026-07-11/14; yfinance eps_trend, 2026-07-18]. Tape consensus (the marginal dollar) has voted the opposite way for three straight weeks: −32% from the post-print ATH, every rally sold, the stock now at 5.6× the very FY27 estimate the sell side publishes [FACT — adjusted price history; yfinance]. The same number — $150.77 — is being capitalized at ~10–11× by research departments and at 5.6× by the market. That divergence, not the demand debate (settled by FQ3), is the current variant-perception battleground. Michael Burry’s disclosed short (“destroyer of capital,” 42-year median ROIC ~4% — Benzinga, 2026-07-03) is the intellectual anchor of the tape’s side. [INTERPRETATION]

Strongest bull case. The bear case is now fighting a contract, not just a price. $100B minimum-price RPO, $22B deposits, take-or-pay terms, floors set above any prior cycle’s peak margins — on ~25% of revenue — convert the oldest cyclical critique (“peak EPS is fake”) into a question with a contractual answer for a quarter of the P&L. Meanwhile 2027 is modeled tighter than 2026 by Goldman/UBS/TrendForce, HBM4 is priced >50% above HBM3E, FY27 consensus has been revised up ~50% in 90 days, and the stock trades at 5.6× that number — cheaper than the HDD makers (WDC 25.7×, STX 27.7× forward) on earnings with better visibility. If the floor holds through the first down-leg, every bear model built on pre-2026 normalized EPS is wrong by 2–3×.

Strongest bear case. The capital cycle does not care about contracts. Burry’s 42-year median ROIC of ~4% is the base rate; ~$976B of market cap requires that base rate to be permanently repealed. The floor protects 25% of revenue — the other 75% is exactly as cyclical as ever, and 2023 already demonstrated that memory customers walk away from LTAs when spot collapses (deposits get forfeited; forfeiture is cheaper than taking delivery at 2× spot). Supply is now a dated schedule: Samsung pulled P5 Fab 2 forward six months, SK hynix just raised ~$26.5B in the US to fund Yongin, CXMT is at ~350k WSPM, and Micron itself will spend >$40B in FY27. And the tape — which led fundamentals at every prior memory top — is behaving as it does at peaks: down 32% on the best quarter in company history, with a price-fixing class action now hanging over the pricing mechanism itself. A single-digit forward P/E on a memory stock has never once been cheap.

The crux has narrowed since the baseline. In June, bull and bear disagreed about persistence of demand. FQ3 settled demand (revenue +74% QoQ, FQ4 guide ~15% above the Street). The disagreement is now singular and arithmetical: the through-cycle value of the uncontracted 75% of revenue. The tape prices it at ~zero terminal value; the sell side prices it at ~10× FY27 earnings; the truth is gated on whether the 2028–29 capacity wave arrives into still-growing AI bit demand (floors ratchet up) or into digestion (floors get tested and renegotiated). [INTERPRETATION]

The five assumptions that matter, with falsifiers (updated from baseline):

What MU is in factor space (FactorsToday, model date 2026-07-17). The factor fingerprint confirms the two-consensus split is not just narrative [FACT — FactorsToday stock-loadings, All-Factors model]: MU is a high-beta semiconductor-momentum stock — largest loadings Semiconductors +1.27, Market +1.21, Technology +1.01, Country: South Korea +0.73 (statistically, MU co-moves with the Korean memory complex: the factor-space signature of the Samsung/SK hynix linkage that has been importing KOSPI volatility into this name all month), Momentum +0.49, Quality +0.47, Growth +0.44 — and a very large negative LowVolatility loading of −1.61, one of the most anti-low-vol names in the universe. Model R² = 0.66: two-thirds of variance is factor-explained, yet residual idiosyncratic vol is still ~43% annualized. The risk-adjusted record is extreme in both directions [FACT — FactorsToday leaderboard, reconciled to the price CSV]: 1-year Sharpe 8.5 on +651% — with a live −30.3% drawdown that is the current one; lifetime Sharpe 0.38 on a 20-year max drawdown of −90.9%. Risk-adjusted brilliance, absolute-risk brutality.

The regime has turned against exactly this profile [FACT — FactorsToday factor returns]: over 126–252d MU’s whole stack was strongly in favor (Semiconductors +22.2%, z +2.4; Technology +20.2%, z +2.1), but over the trailing 21 days every leg has cooled or reversed (Technology −1.9%; Momentum −1.0%, 5-day z −1.3; Growth −1.6%), while LowVolatility (+2.8%, z +1.8) and Value (+2.5%, z +1.7) led the last week — precisely the factors MU is short. Positioning offers no cushion either way: short interest is trivial — ~31.7M shares, 2.8% of float, 0.55 days-to-cover, down from 37.5M the prior month — the shorts have been covering into the drawdown, so there is no squeeze fuel and no crowded-short support [FACT — Yahoo Finance via yfinance, accessed 2026-07-18]. Read: part of this drawdown is a momentum/quality-to-value/low-vol rotation hitting the most anti-low-vol large-cap in the market, not only company-specific news — and the +0.73 Korea loading means “imported volatility” from Seoul is structural, not incidental. Neither factor observation answers whether this is a reset or the start of cyclical mean reversion; they do say the drawdown will not be arrested by positioning mechanics. [INTERPRETATION]

Updated vs. baseline. The baseline’s falsification architecture (“first sequential ASP decline or major capacity announcement breaks the bull”) has partially fired on the capacity leg — Samsung’s P5 pull-forward is the amber flag the baseline pre-specified — yet the stock de-rated without the ASP leg firing. The mechanical tripwires now sit in different places than June: (i) Q4 contract pricing (does +13–18% hold, decelerate, or turn negative); (ii) the first quarterly disclosure of SCA RPO conversion and deposit movement (does the floor transact); (iii) any 2028 bit-supply revision from Samsung/SK hynix/CXMT. The bear no longer needs to prove demand fails — only that the uncontracted 75% mean-reverts. The bull no longer needs demand to accelerate — only that the floor transacts as written. [INTERPRETATION]

The single sharpest variant-perception insight. For the first time in this cycle, the market’s embedded expectation is more conservative than published consensus: the tape de-rated MU 32% while raising its own FY27 estimate ~32%, landing at 5.6× — a multiple that prices ~55–60% permanent erosion of earnings the sell side treats as durable. When the tape refuses to pay for earnings it can already see, the debate is no longer about the estimate. It is a referendum on the SCA floor — and the falsifier is no longer the stock price but the first weak-quarter 10-Q showing whether $22B of customer deposits convert to floor-priced revenue or get renegotiated away. [INTERPRETATION]

No price target. No recommendation.

12. Fact vs. Interpretation

13. Open Questions

1. Do the SCA floors transact? The whole post-FQ3 thesis reduces to this. Watch: the first quarterly disclosure of RPO conversion and deposit movement (FQ4’26 10-K), any renegotiation/forfeiture language, and coverage expansion toward management’s ~50%-of-revenue target. The 2023 LTA-abandonment precedent is the base rate.
2. What are the actual SCA floor prices? TD Cowen’s estimate (~$10–12/GB floor vs. mid-$20s current server DRAM) is unconfirmed and management declined specifics. The floor’s distance from spot determines how violent the first down-leg is before the contract bites.
3. Q4-CY2026 contract pricing (October negotiations): does the +13–18% pace hold, decelerate to single digits, or print the first flat/negative quarter? This is the bull’s master falsifier, live within 90 days.
4. Standalone HBM revenue and margin — still undisclosed (inside CMBU). With Samsung re-entering and Micron’s HBM4 share assumption diluted to ~10% (UBS), the size and profitability of the HBM franchise is a material unknown inside the record numbers.
5. The antitrust class action — N.D. Cal. allegations of coordinated DRAM output restriction. The 2018 analog was dismissed (affirmed 9th Cir. 2022), but the suit hangs over the very pricing mechanism the bull case celebrates, and invites regulatory attention to SCAs themselves.
6. HBM export controls — reported (unconfirmed) tighter unilateral US curbs on HBM; Micron’s China exposure is modest post-CAC-ban but a formal curb would be a new revenue headwind.
7. FY2027 capex and its ROIC — quarterly capex above FQ4’s ~$10B implies a >$40B FY27; the return on the >$250B program depends on where pricing is when the bits land (2028–29). No ROIC gate exists in executive comp to discipline it.
8. The −$321M FQ3 other non-operating expense — not disaggregated in the 10-Q; partially consistent with debt-prepayment costs, composition unconfirmed.
9. Customer-inventory trajectory — ~7–9 weeks (up from 2–4 in February) is below the ~10-week alarm but rising; the direction matters more than the level.
10. Micron’s equity purchase in “a leading AI company” (subsequent event, amount undisclosed) — size, identity, and strategic intent.

14. What Must Be True

Bull case — what must be true, and its falsification test. For the bull (higher-floor supercycle) to be right, the SCA contract layer must transact as written: deposits convert to floor-priced revenue through the first soft quarter without renegotiation, coverage expands toward ~50% of revenue, 2027 stays as tight as Goldman/UBS/TrendForce model, and the 2028–29 capacity wave is absorbed by AI bit demand (HBM4E, Rubin-class accelerators) rather than landing as a glut. Through-cycle EPS then holds at $50–100+ and 5.6× forward is absurd.

• Falsification test: (i) the first quarter of sequential DRAM ASP decline (Q4-CY2026 contract round is the live checkpoint); (ii) any disclosed SCA renegotiation, floor waiver, or deposit forfeiture in a weak quarter — the 2023 LTA playbook; (iii) a 2028 bit-supply revision upward from Samsung (P5 ahead of schedule), SK hynix, or CXMT; (iv) customer inventories through ~10 weeks. Any one breaks the higher-floor thesis.

Bear case — what must be true, and its falsification test. For the bear (cyclical top, mean reversion) to be right, the uncontracted ~75% of revenue must behave like every prior cycle: spot rolls through the floors, floors get renegotiated or deposits forfeited, margins round-trip, and the market re-rates Micron from 13.4× book toward 2–4× a book that keeps compounding. The 42-year median ROIC of ~4% (Burry’s base rate) reasserts.

• Falsification test: sustained sequential ASP strength with supply discipline through FY2027 (already tracking), SCA deposits demonstrably converting to floor-priced revenue (watch the FQ4 10-K RPO roll-forward), coverage rising above ~35–40% of revenue, and 2028 bit-supply delays. If the floor holds through the first down-leg, every pre-2026 normalized-earnings model is wrong by 2–3×.

The update in one paragraph: both tests still key off the same observables — sequential ASPs and capacity — but the SCA layer has inserted a third observable between them: whether a contract can hold a price the spot market has left. The prior report said the thesis would resolve in the data, quarter by quarter. Five weeks later it is resolving exactly as advertised: the ASP data still says up-cycle, the capacity data says the end is scheduled, and the tape has voted for the end before the data has. The next two quarterly prints (FQ4’26 in late September, FQ1’27 in December) carry the two pieces of evidence that matter most: the first SCA conversion disclosure and the Q4-CY2026 contract round.

15. Source Appendix

See the Diligence Questionnaire (Appendix A) and Source Appendix (Appendix B) below for the full diligence answers and citation list. Primary sources relied upon for this update: Micron FQ3 FY2026 Form 10-Q (filed 2026-06-25) and earnings 8-K (2026-06-24); FQ3 FY2026 earnings-call transcript (2026-06-24); Form 3/4 insider filings (2026-06-11 → 2026-07-17); the FY2021–FY2025 10-K corpus and prior 10-Qs (SEC EDGAR); TrendForce/DigiTimes/Counterpoint industry pricing and capacity data (via public press, cited with dates); contemporaneous financial press (Reuters, CNBC, Barron’s, MarketWatch, Benzinga, Fool, Finbold, 247wallst, Invezz); sell-side actions as reported (BofA, HSBC, TD Cowen, KeyBanc, UBS, Goldman); AZI valuation_index and adjusted price history (2026-07-17/18); FactorsToday factor model (2026-07-17); yfinance consensus and comp data (accessed 2026-07-18); and the prior report of 2026-06-09 as the baseline. Third-party aggregator data was reconciled to filings throughout; where they disagreed, the filing won and the discrepancy is noted.

No buy/sell recommendation and no price target appears anywhere in the analytical body of this report; the only position is the clearly-labeled Kimi’s Take block, which is the author’s own view.

APPENDIX A — Standard Diligence Questionnaire

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

Supplemental to the research report. Answers grounded in primary filings and the sources cited in Appendix B; Fact / Interpretation / Assumption labels applied where it matters. As of 2026-07-18.

General

What thoughtful questions have other investors asked about this company? The June question — “are these peak earnings?” — has been answered by the tape before the data: the stock made its all-time high ($1,255 intraday) the day after the best quarter in company history and then fell 32% in three weeks. The questions serious investors now ask:

1. Do the SCA floors transact? Sixteen take-or-pay Strategic Customer Agreements cover ~25% of revenue, with ~$100B of minimum-price cumulative RPO and ~$22B of customer deposits/commitments (~$18B cash), and management claims floor-price gross margins “well above our peak quarterly margins in any past cycle” (FQ3’26 call, 2026-06-24). If the floors hold when spot falls through them, 40 years of memory boom-bust arithmetic changes; if they renegotiate the way 2023’s LTAs did (deposits forfeited, volumes abandoned), they are marketing. TD Cowen’s estimate of the band — ~$10–12/GB floor vs. mid-$20s current server DRAM — is unconfirmed (analyst estimate, not disclosure). This is now the question of the stock.
2. Why does the same estimate get 5.6× and ~10–11×? FY2027 consensus EPS is $150.77 (yfinance eps_trend, 2026-07-18). The market capitalizes it at 5.6× ($848.95, 2026-07-17); the sell side — 89% Buy, zero Sells, targets $1,550–1,750 — at ~10–11×. One side is wrong about the through-cycle value of the uncontracted 75% of revenue.
3. How bad is the 2028 supply wave? Samsung pulled its ~$82.7B P5 twin-fab groundbreaking forward six months to July 2026 (output ~2028); SK hynix raised ~$26.5B in a US listing (2026-07-10) to fund Yongin/M15X; Micron itself guided FY27 capex above a $40B run-rate inside a >$250B program (2026-07-09); CXMT approaches ~350k WSPM by end-2026 (~91% of Micron’s wafer capacity; Tom’s Hardware, 2026-07-16). The glut is scheduled, not present — every third-party model still has 2027 tighter than 2026 (Goldman/UBS/TrendForce, June–July 2026).
4. Is Micron’s HBM share durable? Samsung re-entered: fully qualified on HBM4 at Nvidia/AMD and taking ~30% of Rubin per UBS, vs. SK hynix ~60% and Micron ~10% (UBS, 2026-07). Management says it deliberately holds HBM share near its DRAM share because of the HBM die trade-ratio (FQ3’26 call) — strategy, not loss — but the HBM4-generation dilution is real.

Cyclicality & Earnings Nature

Are earnings at a cyclical high or low? [FACT/INTERPRETATION] A cyclical high, more extreme than at the June baseline. FQ3’26 (ended 2026-05-28): gross margin 84.6%, operating margin 80.4%, net margin 68.1%, diluted EPS $24.67 (10-Q filed 2026-06-25). FQ4 guided to ~86% GM and ~$31 EPS (call, 2026-06-24). Capitalizing these figures at any multiple is the canonical cyclical error; note also FQ4’26 has 14 weeks (53-week year), flattering sequential optics ~8%.

Driven by the external environment or the company’s own actions? Overwhelmingly external. FQ3 revenue +73.8% QoQ on DRAM ASPs +low-60s% and bit shipments up only low-single-digit % (10-Q MD&A). COGS was essentially flat ($6.1B → $6.4B) while gross profit doubled. Company execution (HBM4 ramp “twice as fast as HBM3E,” 1γ DRAM) is genuine but secondary to price.

How stable are revenues? Historically not at all: $30.8B (FY2022) → $15.5B (FY2023) → $37.4B (FY2025) → ~$129B implied FY2026. What is new is a contracted ballast: SCAs cover ~25% of revenue (~20% of DRAM, ~30% of NAND volume) with take-or-pay terms, and management targets ~50% coverage (FQ3’26 call). [INTERPRETATION] A quarter of the P&L now has a contractual floor claim; three-quarters re-prices at market as always. The stability question has narrowed to whether the floor transacts.

Outlook for products/services? Demand is settled, price is the question. CY2026 industry bit growth guided DRAM low-to-mid-20s%, NAND ~20% (call); server unit growth raised to high-teens. Management’s duration claim strengthened: tightness “beyond calendar 2027,” no line of sight to supply catching demand (Mehrotra, 2026-06-24). Against that, management itself flagged “meaningful moderation in the rate of price increases,” and Q3-CY2026 contract forecasts decelerated to +13–18% DRAM / +10–15% NAND (TrendForce, 2026-07-04/09) — deceleration from buyer resistance and LTA lock-ins, not new supply.

How big will this market be — growing, shrinking, domestic or international? [FACT] Growing and global. Total memory ~$552B (2026E) → ~$843B (2027E); HBM TAM ~$100B by 2028 (management, Q1’26 call). Goldman/UBS/TrendForce all model 2027 tighter than 2026 across DRAM, NAND and HBM; server RDIMM 2027 bit supply grows only ~15–20% against demand (TrendForce, 2026-07-09). The constraint is capacity, not demand — which is precisely why the 2028 supply response is the risk.

Business Quality & Competitive Moat

Is the industry getting more or less competitive? [INTERPRETATION] More rational at the top, more contested at the bottom — and both trends accelerated since June. The three-player DRAM oligopoly (Samsung ~38%, SK hynix ~29%, Micron ~22%; Q1-2026) now signs 5-year take-or-pay LTAs with 60–70% of volume/price locked (UBS on SK hynix, 2026-07-07) — conduct the industry has never shown. Samsung’s HBM re-entry (qualified, shipping, ~30% Rubin share) re-contests the premium tier. CXMT (~8% DRAM share, ~350k WSPM, ~$8.5B STAR IPO filed) commoditizes the tail the Big 3 are abandoning anyway.

How profitable is the business (ROIC, ROE)? [FACT] Peak: 9M FY26 net income $47.3B — more than the prior decade combined. Through-cycle: poor. Burry’s disclosed short (2026-07-01) rests on a ~4% 42-year median ROIC — the correct base rate. Greenwald’s two moat tests both fail: share stability exists (three players for a decade) but on a commodity with no pricing power through the cycle, and stable-share-plus-superior-ROIC is absent — ROIC has historically run around or below the cost of capital. The SCA floor is a direct attempt to repeal that base rate for 25% of revenue; unproven through a down-leg.

How profitable is the industry — how many competitors, what barriers to entry? Three scaled DRAM producers; barriers are capital (> $40B/yr for Micron alone in FY27) and EUV/process know-how. But barriers that protect a commodity oligopoly produce peak-trough swings, not franchise returns — Marathon’s capital-cycle point: high returns are now attracting exactly the capital inflow (P5, Yongin, NY/Idaho, CXMT) that will end them, on a publicly dated 2028 schedule.

Can the business be easily understood? Yes — sell bits at a spread to cost. The difficulty is cycle timing, and now, contract enforceability.

Can it be undermined by foreign low-cost labor? Not labor — it is capital-intensive — but by foreign state capital: CXMT (~350k WSPM, Q1-2026 revenue ~$7.3B +700% YoY) is indifferent to ROIC, and Apple is lobbying for clearance to source from CXMT/YMTC (reports, 2026-06/07). Mitigants: CXMT is not yet in qualified HBM (HBM3 MP targeted end-2026, unproven) and its share is concentrated in commodity/consumer DRAM.

Do brands matter? No. B2B qualification, performance-per-watt, and price decide.

What is the nature of competition? Cost-per-bit, node timing, HBM qualification — a capital and engineering race re-run each generation, now with a contract-terms dimension (SCAs vs. SK hynix’s hyperscaler LTAs).

Customers’ switching costs? Real but generation-bound qualification lock-in — plus, new this cycle, take-or-pay contracts with forfeitable deposits (~$22B) that raise the financial cost of switching for ~25% of revenue. Not a compounding moat: customers re-qualify each generation, and 2023 proved customers walk from LTAs when spot collapses.

Financial Condition & Balance Sheet

Assets not fully recognized on the balance sheet? [INTERPRETATION] Process IP, HBM qualification positions, and CHIPS/ITC-subsidized US fab optionality arguably exceed book in the bull case; conversely the fab base (PP&E $56.4B net, $125.0B gross, $10.9B construction-in-progress) is carried at cost and would be impaired in a deep downturn. Book value $63.56/share is compounding ~$25/quarter — the bear’s reversion anchor keeps rising mechanically.

Off-balance-sheet liabilities? [FACT, FQ3’26 10-Q] $1.34B of Lease-SPE finance-lease liabilities off balance sheet (Note 3); PP&E purchase obligations $2.93B (up from $2.10B in February); CHIPS direct-funding covenants, including the buyback restriction running to 2026-12-09; SCA-related commitments. No hidden leverage: net cash ~$24.4B.

How conservative is the accounting? [INTERPRETATION] Reasonably conservative, with two watch items. Trade-AR DSO held at ~59 days (≈58 at FYE25, ≈59 at FQ2) — the $20B receivables build is the mechanical arithmetic of revenue nearly doubling, not channel stuffing; finished-goods inventory fell 43% since FYE25 to $621M. Consideration payable to customers (pricing adjustments/returns) rose to $3.32B from $1.19B at FYE25 — reported revenue is already net of large give-backs; benign now, a margin cushion in reverse later. FY2023’s prompt $1.83B NRV write-down remains the template for honest trough accounting.

How CapEx-hungry is the business? [FACT] Extremely, and more so than in June: FY26 capex guided to ~$27B net of incentives (FQ3 10-Q MD&A), FQ4 alone ~$10B, and FY27 quarterly capex guided above FQ4 — a >$40B run-rate (call, 2026-06-24). The US program was raised to >$250B through 2035 (2026-07-09), plus a new up-to-$3B supply-chain commitment ($500M to GlobalWafers). What de-risks it: ~$22B of customer deposits/commitments, ~$3B/yr government incentives, up to $6.4B federal CHIPS grants, 35% ITC, up to $5.5B NY State incentives — the supercycle is substantially pre-funded by customers and governments, a structure no prior memory upcycle had.

Capital Allocation & Management

How much FCF does the business generate, how does management use it, what is the philosophy? [FACT] FQ3’26 FCF ~$17.6B (OCF $25.4B less capex $7.8B); 9M FCF $26.1B. Use of the windfall in 9M: $9.38B of debt repaid/prepaid (debt $14.6B → $5.7B; net cash ~$24.4B), capex $19.6B gross, $650M buybacks (all H1 at ~$260/share), $437M dividends. Philosophy: deleverage into strength, build for AI demand, let customers and governments pre-fund the fabs. The conservative-financier pattern asserted itself at the peak — a genuine positive delta since June.

Significant acquisitions recently? Small. Tongluo (Taiwan) fab from Powerchip for $1.8B cash (closed March 2026; shipments pulled forward to mid-CY2027). New: an undisclosed purchase of non-marketable equity in “a leading AI company” (subsequent event, 10-Q Note 4 — size and identity open questions). No empire-building.

Buying back shares? Zero in FQ3’26 — no peak-chasing at $1,100+. The H1 buys at ~$260 look well-timed in hindsight. Buybacks are CHIPS-gated until 2026-12-09 (second anniversary of the definitive agreements); from that date management “intend[s] to increase capital return” and, over time, return “100% of excess cash” (Murphy, FQ3 call). [INTERPRETATION] Intent, not commitment; the honest read is buybacks remain the residual claimant. $2.16B remains on the $10B authorization.

Issuing large amounts of new shares to insiders? No — but dilution persists: share count rose 1,122M → 1,129M in 9M despite the buyback; RSUs granted at $227 weighted-average FV vs. $100 a year ago. The token buyback offsets barely a quarter of equity-comp/ESPP issuance.

Compensation policy of directors/management? [FACT] Unchanged since the 2025-11-25 proxy — and still the critical gap: STI on profitability/strategic goals, LTI on relative TSR and share metrics (HBM, DC-SSD), no ROIC or return-on-capital gate anywhere. In a business deploying >$40B/yr into a dated 2028 supply window, the people choosing the capex are paid on bit growth and share, not on what the capital earns. Say-on-pay 84%. Board: Alexis Black Björlin added 2026-06-09 (nine directors, eight independent).

Motivations of management? [FACT/INTERPRETATION] Insiders own <1% (0.25%). Since the baseline: zero open-market buys; ~$91M sold in five weeks at $983–$1,192 — CEO Mehrotra 40,000 sh/$46.3M (10b5-1 adopted 2026-01-30), EVP Arnzen 40,000 sh/$43.4M (~29% of her direct holdings), and director Dugle $1.5M with no 10b5-1 footnote — a discretionary sale (Form 4s, 2026-06-26 → 07-15). The asymmetry is the tell: the company last bought stock at ~$260; the insiders sell at $1,000+. Aligned to grow the franchise; not aligned to per-share compounding.

Valuation & Market Data

Is the stock an ADR, MLP, or K-1 issuer? No — common stock of a US C-corporation (NASDAQ: MU). Standard 1099. (Note: SK hynix’s new Nasdaq ADR, listed 2026-07-10, now gives US investors a direct memory-complex alternative at 3.8× forward vs. MU’s 5.6×.)

Dividend policy? [FACT] Token. $0.15/quarter (raised +30% March 2026; re-declared 2026-06-24, payable 2026-07-21); ~0.07% yield at $849. Never cut, but immaterial to total return — a signaling device.

How profitable is the business? At present, extraordinarily (84.6% GM, 68.1% net margin FQ3’26); through a cycle, modestly to poorly (FY2023: $5.8B loss). The SCA floor is management’s attempt to convert the first sentence into a permanent condition for ~25% of revenue.

Is net income diverging from cash from operations? [FACT] No. 9M FY26 OCF/net income = 0.97; FQ3 = 0.90. The gap is the mechanical $20B receivables build against $47B of net income with DSO flat at ~59 days — clean conversion, not a red flag. Watch: any DSO creep above ~60 days while ASPs stall would be a falsification-grade signal.

Risks & Downside

What factors would cause the stock to decline? [FACT/INTERPRETATION] Ranked by signal value: (1) the first sequential DRAM ASP decline — Q3-CY26 contracts still +13–18%, so the Q4 contract round (October) is the live checkpoint; (2) the 2028 supply wave being pulled forward further (Samsung P5 amber flag already flying); (3) SCA renegotiation/floor waiver/deposit forfeiture in a soft quarter — the 2023 LTA playbook; (4) the DRAM price-fixing class action (N.D. Cal., filed 2026-06-25, naming all three producers — a 2018 analog was dismissed, but it hangs over the pricing mechanism itself); (5) CXMT commodity-price pressure plus Apple legitimation; (6) tighter unilateral US HBM export curbs (reported, unconfirmed, 2026-07-15); (7) AI-capex digestion (Meta capacity-leasing and CoreWeave memory-hedging reports — single-sourced, flagged). The July tape (−32% on a beat-and-raise) shows the market pre-trading this list.

Risk of a catastrophic loss? [INTERPRETATION] Deep drawdown risk: high — this is a 13.4× book, 98th-percentile cyclical with a −90% lifetime max drawdown in its own history. Permanent impairment: very low — $24.4B net cash, IG ratings (upgraded by all three agencies), an essential product, and a capex program now substantially pre-funded by third parties. The downside is a cyclical round-trip, not a zero.

Chance of a total loss? Negligible. Even FY2023’s trough left Micron solvent; today it carries more net cash than its entire debt stack at the prior peak.

Recent News & Events

Has the business environment changed recently? [FACT] Yes, in both directions, in five weeks. Better: the FQ3 print (2026-06-24) beat by ~16% with FQ4 guided ~15% above the Street; the SCA layer was disclosed (16 agreements, ~$100B RPO, $22B deposits); Q3 contract pricing still rising; 2027 modeled tighter than 2026. Worse: the stock de-rated 32% from its print-day ATH on the best quarter ever; the DRAM antitrust suit was filed (6/25); Burry disclosed his short (7/1); SK hynix’s $26.5B US listing (7/10) funds competitor capacity; Samsung pulled P5 forward (July); CXMT filed a ~$8.5B IPO; HBM export-curb reports surfaced (7/15). The debate migrated from demand (settled) to supply (scheduled), law, and positioning.

Significant acquisitions? No — Tongluo ($1.8B, March 2026) is the only one; the unnamed AI-company equity stake is new but undisclosed in size.

Change in accounting policies? None material. Note: FY2026 is a 53-week year (FQ4 has 14 weeks) — sequential comparisons flatter momentum by ~8%. Securities class action dismissed (Feb 2026; voluntarily dismissed Apr 3, 2026); Netlist $445M verdict under appeal with both patents found unpatentable at PTAB.

Recent changes — new markets, facilities, management? [FACT] First concrete poured at the Clay, NY megafab 2026-07-09 (a quarter+ ahead of schedule); US program raised to >$250B; up-to-$3B supply-chain program ($500M GlobalWafers); Boise ID1 first wafers mid-CY2027, ID2 late CY2028; Tongluo shipments pulled forward to mid-CY2027; Singapore HBM packaging contributing 1H-CY2027; SCAs extended into automotive (GM 7/1, Ford 7/6, Qualcomm/Harman 7/16). Management stable (Mehrotra/Murphy); board refresh with Björlin added June 2026.

No buy/sell recommendation and no price target appears in this appendix.

APPENDIX B — Source Appendix

Source Appendix — Micron Technology, Inc. (NASDAQ: MU), 2026-07-18 Update

Compiled from the sources listed below. Update-mode report; baseline = the prior report of 2026-06-09. Every entry is dated; anything that could not be verified against a primary or named secondary source is marked UNVERIFIED rather than dropped. All prices are 2026-07-17 closes (2026-07-18 is a Saturday; no live trading).

Tier 1 — Primary: SEC filings

All filings below were retrieved from SEC EDGAR.

Current-quarter filings (post-baseline):

• MU 10-Q, FQ3 FY2026 (period ended 2026-05-28), filed 2026-06-25 (SEC EDGAR). Primary source for all FQ3/9M financials: Statements of Operations p.5, Balance Sheets p.7, Cash Flows p.10, Notes 3–17 pp.11–25 (Note 9 Debt, Note 10 Contingencies, Note 11 Equity, Note 14 Revenue/SCAs), MD&A pp.26–34 (DRAM/NAND ASP & bit commentary, liquidity, CHIPS), Item 1A Risk Factors.
• MU 8-K (FQ3 FY2026 earnings + FQ4’26 guidance), filed 2026-06-24 — SEC EDGAR acc. 0000723125-26-000013, Exhibit 99.1 a2026q3ex991-pressrelease.htm. (GAAP diluted EPS guide $30.73 ± $1.00; non-GAAP reconciliation per Ex-99.1.)
• MU 8-K Item 5.02 (appointment of director Alexis Black Björlin; board to nine, eight independent), filed 2026-06-09 — SEC EDGAR acc. 0001104659-26-071845. Resolves the baseline’s open question on the 2026-06-09 8-K.
• MU 10-Q, FQ2 FY2026 (period ended 2026-02-26), filed 2026-03-19 (H1 FY26 cash flow used to derive FQ3-only cash flow; Note 14 technology revenue split).
• DEF 14A (proxy), filed 2025-11-25 — latest on file; no new proxy since (comp/incentive read unchanged: no ROIC gate; say-on-pay 84%).

Insider-filing corpus, 2026-06-11 → 2026-07-17 (all via SEC EDGAR):

Form 4 EDGAR accessions: 000179875726000005/6, 000124265426000011/12, 000205876926000004, 000145336826000001, 000121836326000003, 000163206326000003, 000165214926000004. Zero code-P open-market purchases in the window.

FY2021–FY2025 10-Ks (SEC EDGAR): filed 2021-10-08 (FY2021), 2022-10-07 (FY2022), 2023-10-06 (FY2023), 2024-10-04 (FY2024), 2025-10-03 (FY2025).

Prior 10-Qs (SEC EDGAR): 15 quarters, 2021-07-01 (FQ3’21) through 2026-03-19 (FQ2’26), plus the current 2026-06-25 filing.

Historical 8-Ks (SEC EDGAR): 2021-06-30 → 2026-06-24 (~60 filings; earnings 8-Ks, Item 5.02/5.07/8.01 events).

Tier 2 — Primary: company communications

• FQ3 FY2026 earnings call, 2026-06-24 — transcript via the ROIC (roic.ai) earnings-call corpus, accessed 2026-07-18. Caveat: machine-transcription artifacts flagged (“TMBU”=CMBU, “SEAs”=SCAs, “HPN”=HBM, etc.); prepared-remarks cross-check against Micron’s posted version recommended before publication.
• FQ3 FY2026 earnings press release, GlobeNewswire, 2026-06-24 — “Micron Technology, Inc. Reports Record Results for the Third Quarter of Fiscal 2026” — https://www.globenewswire.com/news-release/2026/06/24/3317151/14450/en/Micron-Technology-Inc-Reports-Record-Results-for-the-Third-Quarter-of-Fiscal-2026.html — VERIFIED.
• SCA / automotive press releases: GM strategic memory supply agreement, GlobeNewswire, 2026-07-01 — https://www.globenewswire.com/news-release/2026/07/01/3320600/14450/en/ — VERIFIED; Ford SCA, GlobeNewswire, 2026-07-06 — https://www.globenewswire.com/news-release/2026/07/06/3322416/14450/en/ — VERIFIED; Qualcomm/Harman/Tier-1 auto AI-vehicle memory agreements, 2026-07-16 — corroborated by Reuters, https://www.reuters.com/business/micron-signs-deals-with-qualcomm-others-ai-powered-automobile-chip-components-2026-07-16/.
• “Micron Accelerates U.S. Investments, Pours First Concrete at New York Fab” (>$250B US investment through 2035; Clay, NY first concrete; Commerce Sec. Lutnick, Gov. Hochul), GlobeNewswire, 2026-07-09 — https://www.globenewswire.com/news-release/2026/07/09/3324807/14450/en/Micron-Accelerates-U-S-Investments-Pours-First-Concrete-at-New-York-Fab.html — VERIFIED.
• Up-to-$3B US supply-chain investment incl. $500M GlobalWafers (Sherman, TX 300mm wafers) + 10-yr supply agreement, Micron IR press release, investors.micron.com, 2026-07-09 — not yet SEC-filed; definitive agreements pending (watch for 8-K).
• “Trump Accounts” $250M commitment, GlobeNewswire, 2026-06-30 — https://www.globenewswire.com/news-release/2026/06/30/3319801/14450/en/ — VERIFIED (low materiality).

Tier 3 — Industry data

TrendForce (contract-price forecasts, as cited):

• Q3-2026 conventional DRAM contract +13–18% QoQ — via EE Times China weekly, 2026-07-04 — https://www.eet-china.com/mp/a507373.html
• Q3-2026 server DRAM contract +13–18% QoQ (after ~+50% Q2); 2027 server RDIMM bit supply +15–20% — via Sina/IT之家, 2026-07-09 — https://k.sina.com.cn/article_5953190046_162d6789e06703h5cs.html
• Q3-2026 NAND +10–15% QoQ — igor’sLAB summary of TrendForce Q3-2026 report, 2026-07-07 — https://www.igorslab.de/en/dram-and-nand-remain-more-expensive-trendforce-sees-slowing-but-still-rising-memory-prices-q3-2026/
• SLC NAND +120–170% 2H26 — TrendForce press center, 2026-07-13 — https://www.trendforce.com/presscenter/news
• SK hynix ~70% HBM revenue share Q1-2026 (TrendForce est., via AInvest, 2026-07-10, below)
• Note: TrendForce’s paywalled contract matrices and current-week DRAMeXchange spot prints not directly verifiable; direction/ranges taken from press summaries and secondary citations.

DigiTimes: Q3-2026 DDR4 8Gb contract >+50% QoQ — via Gizmochina, 2026-07-08 — https://www.gizmochina.com/2026/07/08/severe-memory-shortages-drive-ddr4-price-over-50-in-q3-2026-ddr3-also-gets-costlier/; SK hynix Yongin cleanroom 2027 + ~$29B ADR plan — DigiTimes tag feed, 2026-06-26 — https://www.digitimes.com/tag/cleanroom/0012814.html; Samsung HBM4 ramp at P4 — DigiTimes, 2026-05-26 (same tag feed).

Counterpoint (as cited in press): CXMT DRAM share ~8% (roughly tripled YoY) vs MU ~22% — via 247wallst, 2026-07-15 (Tier 6); Q1-2026 conventional DRAM shares Samsung 38% / SK hynix 29% — via AInvest, 2026-07-10 — https://www.ainvest.com/news/sk-hynix-2026-surge-supply-discipline-ai-demand-distinction-determines-2027-2607/

Sell-side research (as cited in press — originals not reviewed):

• UBS: SK hynix >5-yr LTAs with 60–70% volume/price locked; HBM4 Rubin shipments stepped up from Q2-2026 — AASTOCKS, 2026-07-07 — https://www.aastocks.com/en/(Z('ontesting%3D'zzzzz`9769455815`'))/mobile/News.aspx?NewsID=NOW.1532031&NewsSource=HK6&NewsType=; UBS HBM4/Rubin share assumptions SK hynix 60 / Samsung 30 / Micron 10 — wallstreetcn, 2026-04-08 — https://wallstreetcn.com/articles/3769509; UBS server DDR ~11 weeks / SSD ~8 weeks inventory (Dec-2025, latest hard numbers found).
• Goldman Sachs: 2027 modeled tighter than 2026 across DRAM/NAND/HBM — AASTOCKS, 2026-06-05 — https://www.aastocks.com/en/stocks/news/aafn-con/NOW.1526920/
• Citi: Q2-2026 DRAM ASP +44% / NAND +53% QoQ; spot DRAM ~+700% since start of 2025 — via Sahi, 2026-07-13 — https://www.sahi.com/blogs/why-samsung-stock-fell-despite-record-profit-2026; Citi 90-day positive catalyst watch on MU, opened 2026-07-06 (via Tier 6 press).
• KeyBanc (John Vinh): PT to $1,750 Overweight after Asia supply-chain trip; double-digit DRAM/NAND price rises through year-end; HBM prices doubling over the coming year — via Motley Fool, 2026-07-14 (Tier 6).
• TD Cowen (Krish Sankar): SCA floor estimate $10–12/GB vs mid-$20s/GB current server DRAM — analyst estimate, UNCONFIRMED by management; Buy reiterated, $1,600 PT, 2026-07-11 (via Finbold, Tier 6).
• KIS (Korea Investment & Securities): SK hynix Q2 profit estimate ~8% below consensus on slower HBM4 ramp — via 247wallst, 2026-07-13/17 (Tier 6).
• Raymond James “DRAM ASPs peak by mid-2026” warning — carried from baseline (2026-06-09 report); falsification-window input.

SEMI / WFE: 300mm memory fab equipment spending $52B in 2026 (+29% YoY), $57B in 2027 (+11%) — via Futunn, 2026-07-09 — https://news.futunn.com/en/post/75757055/storage-giants-are-raising-substantial-capital-to-stockpile-financial-ammunition

Other industry sources: DRAMeXchange DDR5 spot new high early July 2026 — Open Source Securities via Futunn, 2026-07-09 (above); SemiAnalysis/Citrini CXMT ~350k WSPM by end-2026 (~91% of Micron’s ~385k) — via Tom’s Hardware, 2026-07-16 (Tier 6); Omdia CXMT ~8% DRAM share — cited via Crypto Briefing, 2026-07-13 (Tier 6, attribution chain not fully verified); Samsung P5 Fab 2 pull-forward (₩120T / ~$82.7B twin-fab program) — NineScrolls, 2026-05-13 — https://ninescrolls.com/news/samsung-pulls-forward-p5-fab-2-by-six-months-commits-82-7-billion-to-a-600-000; Samsung 70+ litho tools for P5 PH1 — IT之家/Sedaily, 2026-04-07 — https://www.ithome.com/0/936/362.htm; SK hynix Q1-2026 record / >3-yr HBM demand coverage — NineScrolls, 2026-04-24 — https://ninescrolls.com/news/sk-hynix-q1-2026-record-52-6-trillion-won-revenue-72-operating-margin-m15x-fab; SK hynix 1c ramp 20k→160–190k wpm — AInvest, 2026-07-10 (above) + Korea IT Times, 2026-04-23 — https://www.koreaittimes.com/news/articleView.html?idxno=153265; customer-side DRAM inventories ~7–9 weeks + cycle-trigger dashboard — Luminix, 2026-07-05 — https://www.useluminix.com/reports/industry-analysis/dram-cycle-position-analysis-peak-timing-indicators (research aggregator, medium confidence); hoarding-screening behavior — Sourceability/TrendForce, 2026-01-30 — https://sourceability.com/post/suppliers-work-to-curb-hoarding-as-memory-supply-tightens.

Tier 4 — Transcripts & calls

• ROIC (roic.ai) earnings-call transcript corpus, accessed 2026-07-18 — source of the FQ3 FY2026 (2026-06-24) call transcript.
• Prior-call archive consulted: Micron earnings calls and conference presentations, 2011-06-23 → 2026-06-24 (most recent prior items: Q2-FY26 call 2026-03-18, JPMorgan conference 2026-05-20, Wolfe auto conference 2026-02-11).

Tier 5 — Quantitative / third-party (reconciled, non-authoritative)

• AZI valuation_index, pulled 2026-07-18; index timestamp 2026-07-17 19:00:06 — composite 94.2nd pct, P/B 13.3566× / 98.5th, P/S 16.5669× / 98.5th, P/E 40.005× / 85.6th, price $848.95. KNOWN DISCREPANCY: AZI TTM denominators are stale (pre-FQ3 print) — TTM EPS $21.22 vs yfinance ~$44.23; TTM sales ~$59B vs ~$90B. AZI’s P/E and P/S percentiles therefore overstate expensiveness; P/B percentile unaffected. Refreshed multiples (yfinance): trailing P/E ~19.2×, TTM P/S ~10.6×, EV/EBITDA ~13.8–14.0×.
• Adjusted price history, accessed 2026-07-18 — basis for all price-action stats (2026-07-17 close $848.95; ATH $1,254.81 intraday 2026-06-25; 52-wk low $103.21 2025-08-01).
• FactorsToday factor model, model date 2026-07-17, fetched 2026-07-18 — six endpoints (stock-loadings, stock-info, stock-specific-vol, leaderboard, factor-returns-historic, related-stocks). All-Factors model R²=0.66. Caveat: leaderboard m6 de-annualized return overstates CSV 6-mo move (window alignment approximate; m3/y1 reconcile within 0.2pp).
• yfinance (Yahoo Finance data), accessed 2026-07-18 — FY26 consensus EPS $73.39 (34 analysts), FY27 $150.77 (37 analysts, range $70.77–$221.27), FQ4’26 $31.33; TTM EPS $44.23; comps multiples (SNDK/WDC/STX/NVDA/AVGO/SKHY); short interest ~31.7M sh (~2.8% float, ~6/30 settlement); institutional ~82% / insider ~0.25% ownership. Convenience/aggregator data — non-authoritative.
• ROIC (roic.ai) fundamentals, pulled 2026-07-18 — cross-check only; ties to filings except known label discrepancies: ROIC folds finance leases into LT borrowings (debt 5,794 vs filing 5,140) and labels OCF as “free cash flow” — the filing wins. ROIC 2026-07-17 closes: MU $848.95, SNDK $1,354.38, WDC $479.34, STX $789.20, NVDA $205.58, AVGO $373.41, SKHY $165.69. ROIC annual EV snapshots for SNDK/WDC/STX are FY2025-vintage (pre-boom) and were NOT used.
• SKHY (SK hynix ADR) data-quality flag: yfinance feed partially broken for the new ADR (listed 2026-07-10) — EV/EBITDA −0.34× and P/S 0.008× are ADR-ratio artifacts; market cap $1.09T unverified; yfinance price $154.03 vs ROIC close $165.69 (likely one-session lag). Only trailing/forward P/E used, with caveats; ADR ratio unknown.
• stockanalysis.com MU forecast page — page served a 2026-06-18 cache (STALE, pre-FQ3: FY26 $61.01 / FY27 $117.95); used only to evidence the pre-print consensus level and the size of the post-print revision wave; superseded by yfinance trend.

Tier 6 — Press

Wires / majors

• Reuters, 2026-06-25 — “Micron overtakes Meta in market value amid relentless AI infrastructure demand” — https://www.reuters.com/business/micron-overtakes-meta-market-value-amid-relentless-ai-infrastructure-demand-2026-06-25/
• Reuters, 2026-07-06 — Ford SCA (headline-level, corroborated by company PR)
• Reuters, 2026-07-09 — “Micron to invest up to $3 billion in US chip supply chain” — https://www.reuters.com/business/micron-invest-up-3-billion-us-chip-supply-chain-2026-07-09/
• Reuters, 2026-07-16 — “Micron signs deals with Qualcomm, others for AI-powered automobile chip components” — https://www.reuters.com/business/micron-signs-deals-with-qualcomm-others-ai-powered-automobile-chip-components-2026-07-16/
• CNBC, 2026-06-24 — “Micron is tech’s margin king: memory crisis pushes it past Nvidia, Meta” — https://www.cnbc.com/2026/06/24/micron-is-techs-margin-king-memory-crisis-pushes-it-past-nvidia-meta.html
• CNBC, 2026-07-01 — “Chip stocks notched record rallies in second quarter, start Q3 with dud” — https://www.cnbc.com/2026/07/01/chip-stocks-notched-record-rallies-in-second-quarter-start-q3-with-dud.html
• CNBC, 2026-07-09 — “Micron stock, US chipmaking” (Clay NY concrete pour) — https://www.cnbc.com/2026/07/09/micron-stock-us-chipmaking.html
• CNBC Fast Money, 2026-06-29 — panel discussion of Apple-CXMT sourcing report (no stable URL; via 247wallst 2026-07-02 below) — Apple has not confirmed.
• Barron’s, 2026-07-03 — “Micron stock, Fourth of July” — https://www.barrons.com/articles/micron-stock-fourth-of-july-738609dd
• Barron’s, 2026-07-17 — Micron stock price note — https://www.barrons.com/articles/micron-stock-price-2acd8eb5
• MarketWatch, 2026-07-07 — “Micron’s stock falls as investors wonder if the memory market is near the top” — https://www.marketwatch.com/story/microns-stock-falls-as-investors-wonder-if-the-memory-market-is-near-the-top-ab2d2feb
• MarketWatch, 2026-07-13 — “Micron and other chip stocks feel the pain of imported volatility — blame SK Hynix” — https://www.marketwatch.com/story/micron-and-other-chip-stocks-feel-the-pain-of-imported-volatility-blame-sk-hynix-f705a0fd
• Bloomberg — Apple Mac/iPad price increases citing memory cost (2026-06-25), cited second-hand via TFTC, 2026-07-07 — https://www.tftc.io/apple-cxmt-ymtc-blacklisted-chinese-chips-ai-memory-shortage (original Bloomberg article not directly fetched)
• New York Times, 2026-07-01 — “Micron $250 million Trump Accounts” — https://www.nytimes.com/2026/07/01/business/economy/micron-250-million-dollars-trump-accounts.html
• Seattle Times / AP, 2026-06-23 — “How major US stock indexes fared Tuesday 6/23/2026” (MU −13.2%) — https://www.seattletimes.com/business/how-major-us-stock-indexes-fared-tuesday-6-23-2026/
• Investor’s Business Daily, 2026-07-09 — “SK Hynix stock raises $26.5 billion in U.S. listing” — https://www.investors.com/news/technology/sk-hynix-stock-raises-26-5-billion-u-s-listing-memory-giants-micron-sandisk-rise/
• Investopedia, 2026-06-11 — markets wrap (tech rebound; Iran strikes called off) — https://www.investopedia.com/stock-market-today-dow-jones-s-and-p-500-06112026-11995532
• BusinessWire, 2026-07-06 — Scott+Scott fiduciary-duty investigation of Micron — https://www.businesswire.com/news/home/20260706993537/en/

Trade / specialist

• Tom’s Hardware, 2026-07-16 — “CXMT close to matching Micron’s memory capacity in 2026, research claims” — https://www.tomshardware.com/pc-components/dram/cxmt-close-to-matching-microns-memory-capacity-in-2026-research-claims-would-put-china-on-track-to-become-worlds-second-largest-dram-producer
• EE Times China, 2026-07-04 — TrendForce Q3-2026 weekly (Tier 3 URL above)
• igor’sLAB, 2026-07-07 — TrendForce Q3-2026 summary (Tier 3 URL above)
• DigiTimes, 2026-05-26 / 2026-06-26 — cleanroom/Yongin tag feed (Tier 3 URL above)
• TheElec-class Korea/China press used via citation chains: Korea IT Times, 2026-04-23; IT之家/Sedaily, 2026-04-07; Sina/IT之家, 2026-07-09; wallstreetcn, 2026-04-08; NineScrolls, 2026-04-24 / 2026-05-13; AASTOCKS, 2026-06-05 / 2026-07-07 (URLs in Tier 3); Witology semiconductor daily, 2026-06-18 — https://witology.com.tw/daily-semiconductor/2026-06-18/; Korea Market Morning Brief, 2026-06-16 — https://bigbloo.tistory.com/359
• Sourceability, 2026-01-30 — hoarding screening (Tier 3 URL above)

Aggregators / blogs (lower evidentiary weight — used for event detection, verified against originals where noted)

• Motley Fool, 2026-06-29 — “Apple Just Raised Prices Because of a Chip Shortage” — https://www.fool.com/investing/2026/06/29/apple-just-raised-prices-because-of-a-chip-shortag/
• Motley Fool, 2026-07-07 — “Samsung’s Preliminary Quarterly Profit Just Jumped 19-Fold — and Micron Stock Fell” — https://www.fool.com/investing/2026/07/07/samsungs-preliminary-quarterly-profit-just-jumped/ — VERIFIED
• Motley Fool, 2026-07-14 — “Why Micron Technology Stock Rallied on Tuesday” (KeyBanc $1,750; 89% Buy skew) — https://www.fool.com/investing/2026/07/14/why-micron-technology-stock-rallied-on-tuesday/ — VERIFIED
• Finbold, 2026-06-26 — “Here’s Wall Street’s average price target for Micron stock” — https://finbold.com/heres-wall-streets-average-price-target-for-micron-stock/
• Finbold, 2026-07-02 — “Monster insider trading alert for Micron stock” (Mehrotra sale; printed “4,000 shares / $46M” — mutually inconsistent figures, superseded by Form 4) — https://finbold.com/monster-insider-trading-alert-for-micron-stock/
• Finbold, 2026-07-07 — “Monster insider trading alert for Micron stock (2)” (Arnzen 40,000 sh / $43.3M) — https://finbold.com/monster-insider-trading-alert-for-micron-stock-2/ — VERIFIED
• Finbold, 2026-07-11 — “Wall Street analyst sets Micron stock price target” (TD Cowen $1,600) — https://finbold.com/wall-street-analyst-sets-micron-stock-price-target/
• 247wallst, 2026-06-24 — “$900 billion just vanished as Korean chip stocks crater” — https://247wallst.com/investing/2026/06/24/900-billion-just-vanished-as-korean-chip-stocks-crater-but-the-squawk-crew-isnt-sold-on-an-ai-breakdown/
• 247wallst, 2026-06-25 — “Micron just locked in $100 billion in sales” — https://247wallst.com/investing/2026/06/25/micron-just-locked-in-100-billion-in-sales-and-wall-street-thinks-the-boom-bust-chip-cycle-is-dead/
• 247wallst, 2026-07-02 — “Apple’s rumored China chip deal could blow a hole in Micron” — https://247wallst.com/investing/2026/07/02/apples-rumored-china-chip-deal-could-blow-a-hole-in-micron-and-trap-the-us-for-years/
• 247wallst, 2026-07-07 — “Insiders are dumping Micron at the highest rate since 2010” — https://247wallst.com/investing/2026/07/07/insiders-are-dumping-micron-at-the-highest-rate-since-2010-is-it-time-to-get-out/
• 247wallst, 2026-07-13/17 — KIS SK-hynix estimate detail (via 2026-07-17 article below)
• 247wallst, 2026-07-15 — “$8.5 Billion Says This Micron Threat Is About to Get a Whole Lot Bigger” (CXMT IPO; Counterpoint share data) — https://247wallst.com/investing/2026/07/15/8-5-billion-says-this-micron-threat-is-about-to-get-a-whole-lot-bigger/ — VERIFIED
• 247wallst, 2026-07-17 — “Micron Technology: Record DRAM Pricing Meets A Stock In Retreat” (CoreWeave hedging claim single-sourced) — https://247wallst.com/technology-3/2026/07/17/micron-technology-record-dram-pricing-meets-a-stock-in-retreat/ — ASP data VERIFIED
• Invezz, 2026-07-01 — “Samsung, SK Hynix, Micron sued over DRAM prices: what’s at stake” (Garciaguirre v Samsung et al, N.D. Cal.) — https://invezz.com/news/2026/07/01/samsung-sk-hynix-micron-sued-over-dram-prices-whats-at-stake/ — VERIFIED
• Invezz, 2026-07-03 — Burry short detail (short at $1,051.87 on 07-01)
• Invezz, 2026-07-10 — “Is SK Hynix stock a better pick to play AI memory market than Micron?” (IPO $149, ~$26.5B, 56.4% HBM share) — https://invezz.com/news/2026/07/10/is-sk-hynix-stock-a-better-pick-to-play-ai-memory-market-than-micron/ — VERIFIED
• Invezz, 2026-07-16 — “Micron, SK Hynix, other memory stocks fall as TSMC outlook sparks chip selloff” — https://invezz.com/news/2026/07/16/micron-sk-hynix-other-memory-stocks-fall-as-tsmc-outlook-sparks-chip-selloff/ — VERIFIED
• Benzinga, 2026-07-03 — “Michael Burry Is Shorting Micron as a ‘Destroyer Of Capital’…” — https://www.benzinga.com/trading-ideas/short-ideas/26/07/60265248/ — VERIFIED
• Proactive Investors, 2026-06-25 — post-print analyst roundup — https://www.proactiveinvestors.com/companies/news/1094520
• Proactive Investors, 2026-07-15 — “Dell, Micron, SanDisk tumble as AI hardware rally hits reversal” (Meta leasing report; HBM export-restriction report) — https://www.proactiveinvestors.com/companies/news/1095510/ — VERIFIED as to what was reported; underlying claims UNVERIFIED (Tier 8)
• FXLeaders, 2026-07-01 — “MU stock heads under $1,000 as the Micron selloff deepens on DRAM lawsuit, oversupply concerns” — https://www.fxleaders.com/news/2026/07/01/mu-stock-heads-under-1000-as-the-micron-selloff-deepens-on-dram-lawsuit-oversupply-concerns/
• MarketBeat, 2026-06-24 — “Micron’s sudden plunge may be an AI buying chance” — https://www.marketbeat.com/articles/microns-sudden-plunge-may-be-an-ai-buying-chance/
• GuruFocus, 2026-06-25/29 — analyst-raise roundup (no stable URL captured)
• AInvest, 2026-07-10 — SK hynix supply-discipline note (Tier 3 URL above)
• Sahi, 2026-07-13 — “Why Samsung stock fell despite record profit” (Citi ASP data; Tier 3 URL above)
• Quantel market summary, 2026-06-23 — KOSPI −10%, Samsung −12.3% — https://resources.quantel.ai/daily-market-summary/market-summary-june-23-2026-tech-stocks-hit-hard
• Matterfact AI Capex Tracker, 2026-06-23 — DeepSeek/Microsoft demand-crack narrative — https://www.matterfact.com/newsletter/2026-06-23-ai-capex-microsoft-deepseek-micron; Matterfact, 2026-07-15 — BofA “positioning/valuation adjustment” characterization — https://www.matterfact.com/newsletter/2026-07-09-ai-capex-nvidia-gives-back-a-trillion
• SIAM market note, 2026-07-07 — “Chip Stocks Slide as Samsung Earnings Fail to Meet Sky-High AI Expectations” — https://www.siam.in/aticles-market/Chip-Stocks-Slide-as-Samsung-Earnings-Fail-to-Meet-SkyHigh-AI-Expectations-36-1774
• Vittarthi market wrap, 2026-06-25 — post-print semi rally — https://vittarthi.com/resources/news/us-stock-market-today-25-june-2026
• Newsquawk US Market Wrap, 2026-07-15 — ASML EUV guide, semis weighed — https://www.newsquawk.com/headlines/newsquawk-us-market-wrap---15th-july-2026-stocks-mixed-on-asml-earnings-and-soft-ppi
• Univest, 2026-07-14 — US–Iran escalation risk-off framing — https://univest.in/blogs/stock-market-prediction-for-tomorrow-15-july-2026
• Pluang news feed, 2026-06/07 — Micron FQ3 record results — https://pluang.com/en/news-feed/micron-tinjauan-memori-menjelang-laporan-keuangan; Pluang, 2026-07 — Burry short note — https://pluang.com/en/news-feed/peringatan-perdagangan-orang-dalam-micron
• Gizmochina, 2026-07-08 — DDR4 >+50% Q3 (Tier 3 URL above)
• Crypto Briefing, 2026-07-13 — CXMT revenue/export-control challenges — https://cryptobriefing.com/cxmt-us-export-control-challenges/; Crypto Briefing, 2026-07-16 — “US lawmakers [press to] ban Chinese memory chips” — https://cryptobriefing.com/us-lawmakers-ban-chinese-memory-chips/
• Tech Wire Asia, 2026-06-30 — “Apple China memory chips CXMT” — https://techwireasia.com/2026/06/apple-china-memory-chips-cxmt/
• Startup Fortune, 2026-05-23 — “China’s memory push could turn today’s chip shortage into a price break” — https://startupfortune.com/chinas-memory-push-could-turn-todays-chip-shortage-into-a-price-break/
• Luminix DRAM cycle tracker, 2026-07-05 (Tier 3 URL above)
• Studio Global, 2026-06-24 — Samsung HBM4 fact-check compilation — https://www.studioglobal.ai/discover/answers/search-fact-check-with-cited-sources-for-6a3b5f5042c5a4f0357b79d9 (weak source — flagged)
• Futunn news, 2026-07-09 — SEMI WFE / capital-raising roundup (Tier 3 URL above)

Tier 8 — Unverified / single-sourced items register