Micron Technology MU

Reference table

1. What the business is

Micron Technology, incorporated 1984 and headquartered in Boise, Idaho, employs 53,000 people under chief executive Sanjay Mehrotra. It manufactures three memory technologies, described in the 10-Q as follows.

• DRAM — volatile, low-latency working memory. This includes High-Bandwidth Memory (HBM), a 3D-stacked DRAM architecture using through-silicon vias, which AI accelerators consume directly and which is the physical bottleneck in the current buildout.
• NAND — non-volatile storage, sold as components, SSDs and managed NAND.
• NOR — non-volatile code storage for automotive, industrial and embedded uses.

The reportable segments changed and the vendor file has not caught up. Note 17 of the FQ3'26 10-Q states the company reports through four business units: CMBU (Cloud Memory — hyperscale cloud, and HBM for all data centre customers), CDBU (Core Data Center — mid-tier cloud, enterprise, OEM data centre, and all data centre storage), MCBU (Mobile and Client), and AEBU (Automotive and Embedded). The vendor's seg_prod block contains none of these: it reports a DRAM/NAND technology split for FY2025 and the superseded CNBU/EBU/MBU/SBU unit names for FY2024 and earlier. The series is therefore discontinuous and off-basis, and is rejected as a segment series (Section 8).

What the filing does give, and the vendor cannot. The dollar tables did not survive text extraction, but the 10-Q states the growth rates directly:

Every unit more than tripled year on year, and the two data centre units led. CDBU's +653% is the standout. The 10-K states that in FY2025 approximately one-half of total revenue was concentrated in the data centre end market and over half of total revenue came from the top ten customers.

Technology mix (FY2025, the only year on a comparable basis): DRAM $28.578B (76.5% of revenue), NAND $8.503B (22.7%), unallocated $297M (0.8%).

Geography, and an asymmetry worth naming. FY2025 revenue by billing location: United States $24.113B (64.5%), Taiwan $5.672B (15.2%), China $2.639B (7.1%), Other Asia-Pacific $1.913B (5.1%), Hong Kong $1.138B (3.0%), Japan $895M (2.4%), Europe $625M (1.7%), other $383M (1.0%). The US share jumped from 52.4% in FY2024. But the 10-Q risk factors state that "a majority of our DRAM production output in 2025 was from our fabrication facilities in Taiwan." Revenue is billed to America; the wafers are made in Taiwan. That asymmetry is the geopolitical exposure and no financial statement captures it.

Customer concentration is falling, not rising. The 10-Q discloses revenue from one customer at 10% of total revenue for 9M FY2026, down from 16% for 9M FY2025, primarily in CMBU.

2. The single most important thing — this is a price event, and the trailing multiple inverts

Everything in this dive follows from one decomposition, and it is the company's own.

The FQ3'26 10-Q states that total revenue rose 346% year on year and 74% sequentially. It then breaks that down:

> "Sales of DRAM products increased 343%, primarily due to a low-260% range increase in average selling prices and a low-20% range increase in bit shipments."

> "Sales of NAND products increased 361%, primarily due to a mid-310% increase in average selling prices and a low-double-digit increase in bit shipments."

And sequentially, more starkly:

> "Sales of DRAM products increased 67%, primarily due to a low-60% range increase in average selling prices and a low-single-digit percentage range increase in bit shipments."

Roughly 80% of the revenue increase is price. Roughly 20% is volume. Gross margin followed exactly as arithmetic requires: 38% (FQ3'25) → 74% (FQ2'26) → 85% (FQ3'26), and 38% → 77% on a nine-month basis. Every one of those figures is stated in the filing and every one reproduces the vendor's series to within rounding.

Micron's own risk factors then supply the historical context, and they are more bearish than any outside commentator:

> "Over the past five fiscal years, annual percentage changes in DRAM average selling prices have ranged from an increase in the low 40% range to a decrease in the high 40% range. DRAM average selling prices increased approximately 140% for the first nine months of 2026... In some prior periods, average selling prices for our products have been below our manufacturing costs, and we may experience such circumstances in the future."

The current nine-month move is roughly three and a half times the largest annual increase of the prior five years, and the historical downside case is a high-40% annual decline. That is the cycle, stated by the company, in the filing.

The historical record is not ambiguous. Micron's six reported fiscal years:

In FY2023 — two years ago — gross margin was negative nine percent and the company lost $5.34 a share. That is the same company, the same fabs, the same three-player industry structure. FY2026E EPS of $73.22 is 9.5x the prior-cycle peak and 24x the six-year average.

The inversion — why 5.8x forward is not cheap

At $892.67 the forward multiples are:

Every screen reads 5.8x and calls it cheap. The correct reading is the opposite, and the arithmetic is one line. A durable earnings stream deserves roughly 15x. So:

> $892.67 ÷ 15 = $59.51 of permanent earnings per share that today's price already capitalises.

At 12x the implied permanent figure is $74.39; at 18x it is $49.59. Take the range as $50-$74 of durable EPS. Now place it:

• It is 32-48% of the FY2027 consensus of $154.67 — so the market believes roughly a third to a half of consensus FY2027 earnings is permanent, and is refusing to pay for the rest. That refusal is the low multiple. The multiple is not a discount to fair value; it is the market's estimate of the transitory fraction.
• It is 6.5x to 9.6x the prior-cycle peak EPS of $7.74.
• It is 16x to 25x the six-year average EPS of $3.03. On that average, the trailing multiple is 295x.

Micron is not statistically cheap. It is a stock in which a permanent eight-fold improvement over the best year in company history is already the base case in the price. A tracked independent claim from 2026-05-12 puts the general form of this better than we can:

> "Don't mistake memory names' single-digit PEs for cheapness — they're the most cyclical names on the planet with wildly swinging operating margins, so they deserve a discount."

> — Josh Brown, 2026-05-12, neutral, conviction 65, compound_and_friends

The honest counterweight, which is genuinely new. Two disclosures in this 10-Q have no precedent in memory and they are the reason this dive is not a Watch:

1. Take-or-pay agreements. "These agreements are structured as take-or-pay agreements, with binding commitments for specific volumes over the multi-year contract terms and include contractually enforceable volumes. Pricing for most agreements is either fixed, or is subject to minimum and maximum pricing. The largest agreements generally have a ceiling price for existing products that approximates the market price in the second calendar quarter of 2026, and a floor price through the term of the agreement."

2. Customer prepayments. "In connection with these strategic customer agreements, we expect to receive cash deposits and related financial commitments of $22 billion for agreements concluded to date. Approximately $18 billion of these commitments will be in the form of cash deposits."

At FY2025 year-end, the same disclosure read that remaining performance obligations "were not material." The entire structure appeared in one quarter. Customers do not prepay eighteen billion dollars for a commodity they expect to buy cheaply in eighteen months.

But read the ceiling clause again. The largest agreements cap the price at roughly the second-calendar-quarter-2026 market level — which is to say, at approximately today's peak. Micron has traded away its upside above current pricing in exchange for a floor. That is a rational trade for the company and it does de-risk the trough, but it also means the FY2027 consensus of $154.67, which implicitly assumes further price appreciation, is partly capped by contract on the largest volumes. This cuts both ways and almost nobody is saying so.

And check the size. The 10-Q states: "As of May 28, 2026, the transaction price allocated to our remaining performance obligations was approximately $5 billion, of which $422 million has been recognized as contract liabilities... Approximately one-third of the remaining performance obligations as of May 28, 2026 are expected to be recognized as revenue over the next twelve months." Management qualifies it — the figure uses minimum committed volumes and minimum pricing, excludes contracts without price bands, and is "not expected to be indicative of future revenue." Fair. But the contractually disclosed backlog is roughly 12% of a single quarter's revenue and roughly 5.5% of trailing twelve-month revenue. The floor under this business is thinner than the narrative implies.

3. Growth and margin trajectory

Verification of the FQ3'26 quarter, since the brief required it and the earnings-release exhibit is absent from the 8-K. The 8-K of 2026-06-24 is a two-page cover document; Exhibit 99.1, the press release carrying the financials, is not in the extracted text, and the 10-Q's financial statement tables did not survive extraction either. The quarter is nonetheless confirmed by four independent percentage disclosures in the 10-Q MD&A, each of which reproduces the vendor series:

1. Revenue +346% year on year: $9.301B × 4.46 = $41.48B ✓

2. Revenue +74% sequentially: $23.860B × 1.74 = $41.52B ✓

3. Nine-month revenue +203%: ($13.643 + $23.860 + $41.456) ÷ ($8.709 + $8.053 + $9.301) = 78.959 ÷ 26.063 = +202.9% ✓

4. Gross margin 85% / 74% / 38% and 77% / 38% nine-month: vendor 84.6% / 74.4% / 37.7%, and 60.457 ÷ 78.959 = 76.6% ✓

The $41.456B quarter is confirmed. So is the gross margin. The $25.11 EPS is not directly confirmable and is a different basis from the $24.67 in the income statement — see Section 8, item 2.

Beat history — five consecutive beats, accelerating on revenue:

Analysts have been unable to model the price move for five straight quarters. That is a real and repeated pattern, and it is the strongest argument for taking the over on FQ4. It is also exactly what the last three quarters of every prior memory cycle looked like.

Operating leverage is now essentially total. Revenue rose $17.596B sequentially in FQ3'26 and operating income rose $17.198B — a 97.7% incremental operating margin. When price is the driver and volume is flat, nearly every incremental dollar of revenue is an incremental dollar of profit. The same arithmetic runs in reverse on the way down, which is why memory operating margins swing from +80% to −37% (FY2023: operating loss of $5.745B on $15.540B of revenue) within two years.

Operating expenses are rising fast and permanently. R&D was $3.798B in FY2025 and the 10-Q states R&D expense rose 36% year on year in both FQ3 and the nine months, "as we ramp R&D investments in support of long-term opportunities." SG&A rose 28% and 22%. These are fixed costs being reset upward at the cycle peak — they will not come back down when price does.

4. Balance sheet and the enterprise-value rebuild

The vendor's balance sheet is FY2025 (2025-08-28) — nine months stale, and it predates the entire boom. It shows $10.307B of cash and short-term investments and $15.278B of total debt. Both figures are now wrong by an order that matters. The 10-Q settles it:

Note that even at the same date the filing and the vendor disagree. The 10-Q states cash and marketable investments were $11.94B at 2025-08-28; the vendor's cash plus short-term investments for that date is $10.307B. The $1.63B gap is longer-dated marketable securities that the vendor's "cash + short-term investments" excludes. The filing wins; we use $30.13B for the current date.

Debt, rebuilt from the filing. The debt footnote tables did not survive extraction, but two filing disclosures bracket it. The 10-Q's financing-activities discussion states $9.38 billion of debt repayments in the first nine months of FY2026, comprising "the prepayment in full of the 2028 Notes, 2029 Term Loan A, 2029 A Notes, 2029 B Notes, and 2030 Notes and the partial prepayments of the 2031 Notes, 2032 Notes, 2033 A Notes, 2033 B Notes, 2035 A Notes, and 2035 B Notes." The 8-K of 2026-04-01 independently confirms the mechanism — cash tender offers for any and all of six series (5.300% 2031, 5.650% 2032, 5.875% 2033A, 5.875% 2033B, 5.800% 2035A, 6.050% 2035B). $15.278B less $9.38B is $5.90B, plus finance-lease additions (the 10-K flagged $1.16B of executed-but-not-commenced finance leases at FY2025 year-end). The vendor's own trailing ratios triangulate to the same place: debtToAssetsRatioTTM 0.0475 against financialLeverageRatioTTM 1.3315 on equity of $100.9B implies assets of ~$134.3B and total debt of ~$6.4B. Both routes agree.

The corrected enterprise value:

> Market cap $1,008.173B + debt ~$6.4B − cash and marketable investments $30.13B = ~$984.4B

The vendor reports enterpriseValueTTM of $989.554B, implying net cash of $18.619B against our corrected $23.7B. The vendor understates net cash by ~$5.1B (27.6%) and overstates enterprise value by ~$5.1B — the familiar marketable-securities omission. In direction it is the textbook defect. In magnitude it is 0.52% of enterprise value, and it changes no conclusion in this dive. We say so rather than inflating it; honesty runs both ways.

Micron has deleveraged violently at the top of the cycle, which is exactly right and exactly what it failed to do in prior cycles. Interest coverage is 258x. Debt to equity is 0.063. debtToMarketCapTTM is 0.51%. This is the strongest balance sheet Micron has ever carried into a potential downturn, and it is the principal reason the risk score is 8 rather than 9.

Inventory — the most encouraging number in the file. Inventory was $8.387B (FY2023), $8.875B (FY2024), $8.355B (FY2025), and averageInventoryTTM is $8.417B. It is flat in dollars while revenue quadrupled. Measured against revenue, inventory has fallen from roughly 197 days (FY2023) to roughly 34 days of trailing revenue. Days of inventory on a cost basis reads 126 days, but that is an artifact of cost of goods sold collapsing to 27% of revenue at an 85% margin. There is no inventory build. This is a genuine shortage, not channel stuffing, and it is the single cleanest piece of evidence for the bull case anywhere in the data.

Receivables — the least encouraging. averageReceivablesTTM is $24.169B and days of sales outstanding read 98 days on a trailing-revenue basis and 125 days on the vendor's own turnover calculation (the two are internally inconsistent — Section 8). The 10-Q confirms the direction: operating cash flow was held back by "a significant increase in receivables due to higher revenue." Twenty-four billion dollars of receivables against customers concentrated in a single end market is a real credit exposure if demand air-pockets, and the filing's own risk factors contemplate exactly that: "we may experience a decrease in short-term and/or long-term demand... Any inability of our current or potential future customers to pay us for our products may adversely affect our earnings and cash flow."

5. Cash flow and capital returns — where the money actually goes

Trailing twelve months, rebuilt:

The vendor's trailing capex ties to the filing to within $4 million, which is worth saying because it validates the rest. The 10-Q discloses $19.60B of PP&E expenditure in 9M FY2026 and $10.20B in 9M FY2025; the vendor's FY2025 full-year capex is $15.857B, implying FQ4'25 of $5.657B. $5.657B + $19.60B = $25.257B against the vendor's derived $25.261B. Confirmed — and it establishes that the vendor's capex is gross PP&E expenditure, excluding the government-incentive offset.

Which means the vendor understates free cash flow. The 10-Q reports $2.99B of "proceeds from government incentives to offset capital expenditures" in 9M FY2026 (against $1.29B in 9M FY2025), booked as a separate investing line. These are a genuine capex offset. Corrected trailing free cash flow is approximately $29.2B, not $26.2B, and the free-cash-flow yield is ~2.9% rather than the vendor's 2.60%.

Either way, the number is the point. At the greatest earnings peak in the history of this industry — $50.5B of trailing net income, a 55.9% net margin, an 85% gross margin — the free-cash-flow yield on a $1.008 trillion market capitalisation is 2.6% to 2.9%. Capital expenditure consumes 49.1% of operating cash flow. That is the cyclicality argument restated in cash: peak accounting earnings are not converting to owner cash, because they are being reinvested into the capacity that will eventually end the shortage.

And capex is still rising. The FQ2 10-Q (2026-03-19) guided FY2026 capital expenditure to "above $25 billion" net of government incentives. The FQ3 10-Q (2026-06-25) raised it to "approximately $27 billion." Against 9M net capex of $16.61B ($19.60B gross less $2.99B incentives), that implies roughly $10.4B of net capital expenditure in FQ4 alone — a 14-week quarter. Purchase obligations for PP&E stood at $2.93B at 2026-05-28, up from $2.10B at 2026-02-26.

Government incentives materially distort both capex and tax, and must be named. The 10-Q discloses: CHIPS Act direct funding agreements for up to $6.1B (Boise plus Clay, New York) and a further up to $275M for Manassas, Virginia — $6.4B of grants in total; a 35% investment tax credit on qualified US semiconductor investments; and a non-binding term sheet with New York State for up to $5.5B over 20-plus years. Micron also acquired a wafer fab at Tongluo, Taiwan from Powerchip in March 2026 for $1.8B cash. All incentives are "conditioned upon achieving or maintaining certain outcomes and satisfying compliance requirements and are subject to reduction, termination, or clawback."

Capital returns are, at the peak of the greatest cycle in company history, essentially nil.

• Buyback: $10B authorised, $7.84B cumulatively used through 2026-05-28, leaving ~$2.16B (0.21% of market cap). In 9M FY2026 Micron repurchased 2.5 million shares for $650M. In FQ3'26 — the $41.5B quarter — the 10-Q states plainly: "No shares were repurchased in the third quarter of 2026." The authorisation is "subject to... restrictions applicable under our CHIPS Act direct funding agreements."
• Dividend: $0.115 per quarter in FQ1 and FQ2 FY2026, raised to $0.15 in FQ3, and $0.15 declared 2026-06-24 for payment 2026-07-21. Annualised $0.60 is a 0.067% yield.
• Total 9M returns: $650M of buyback plus $437M of dividends = $1.087B against a $1.008T market capitalisation. A further $762M went to share repurchases for tax withholding on employee equity, which is not a return to shareholders.

A company that believed its shares were cheap, holding $30.13B of cash and generating $29B of free cash flow, did not buy a single share in its best quarter ever. That may be CHIPS-agreement restriction rather than judgement, and the filing supports that reading. It is nonetheless a fact, and it belongs in the verdict.

6. Valuation — priced in or room?

The estimate dispersion is the cycle debate, quantified. FY2026 EPS spans $72.13 to $79.46 — a 10% band, because the year ends in 24 days. FY2027 spans $129.31 to $216.83 on 23 analysts, a 57% band. FY2028 spans $82.29 to $200.85 on only 12 analysts — the low estimate is 49% of the mean. By FY2029 there are five analysts and by FY2030 seven. No conclusion in this dive rests on any estimate beyond FY2027, and FY2027 is used with its dispersion visible.

A basis note that matters. The consensus series is non-GAAP. The reported GAAP diluted EPS for FQ3'26 is $24.67 ($28.243B ÷ 1.145B shares, which ties exactly); the earn_cal actual and the basis on which the estimates are struck is $25.11, a difference of ~$504M. All forward multiples in this dive are non-GAAP-to-non-GAAP; the trailing 20.2x is GAAP.

6a. What today's price assumes (the inversion)

At $892.67 — 12.19x FY2026E and 5.77x FY2027E — the price embeds the following falsifiable claims. Each is labelled by source.

• That roughly $50-$74 of EPS is permanent, and the rest of FY2027 consensus is not. (Our arithmetic on consensus inputs: $892.67 ÷ a 12x-18x durable multiple.) That is 32-48% of the $154.67 FY2027 consensus, and simultaneously 6.5x-9.6x the prior-cycle peak of $7.74. This is the central claim and the most important number on this page. It is falsified upward if FY2028 EPS prints above ~$120 with gross margin above 60%; falsified downward if FY2028 EPS prints below $40.
• That gross margin falls from 85% but not below roughly 55-60% on a through-cycle basis. (Our derivation from the implied durable EPS.) A tracked independent claim from 2026-06-25 (Angelo Zino, conviction 60) models margins levelling in the mid-80s for a couple of quarters and normalising toward the mid-70s — above our implied assumption, which is to say the specialist consensus is more bullish than the price. The most fragile assumption, and named as such: that the 85% margin normalises rather than reverts. Micron's own five-year history contains a −9.1% gross margin year.
• That the FQ4'26 print on 2026-09-22 lands near the $31.33 / $50.452B consensus. (Consensus-derived.) Adjust it for the 14-week quarter the 10-Q discloses: +21.7% sequential revenue growth is +13.0% per week. A quarter that beats on the headline but misses per week would be a genuine deceleration signal that most reporting will not catch.
• That capital expenditure of approximately $27 billion net does not destroy value. (Filing-derived guidance, half weight as management source.) At a ~2.9% free-cash-flow yield, the market is currently paying for growth it is not receiving in cash, and the payback on the New York fabs begins "in 2030 and beyond."
• That no de-rating occurs, i.e. the market keeps paying ~5.8x forward. This is the assumption most likely to be wrong in either direction, because 5.8x is not an equilibrium multiple for anything — it is either far too low for a structurally changed business or roughly right for a peak-cycle commodity. There is no scenario in which 5.8x forward is a stable resting place.

6b. The return bridge (why the multiple moves)

> Expected return ≈ through-cycle EPS growth + multiple re-rating from cyclical to structural + shareholder yield (~0.2%, effectively zero)

Our base case explicitly assumes multiple EXPANSION, and that is the fragile leg — but expansion measured against durable earnings, not against consensus. The base of $1,050 is 15x an assumed through-cycle EPS of $70. Against the reported optics it is 14.3x FY2026E, 6.8x FY2027E and 11.8x book value. So on the numbers the market currently looks at, our base implies almost no re-rating at all; on the number we think matters, it implies the market accepting that $70 rather than $59.51 is the durable figure.

Almost none of the expected return is shareholder yield. The dividend is 0.067% and the buyback has $2.16B of authorisation left, of which zero was used last quarter. This is a pure re-rating and earnings-durability bet, with no income floor whatsoever — which is unusual for a $1 trillion company and should be stated plainly.

The bull case at $1,700 requires the market to accept ~$113 of durable EPS (73% of the FY2027 consensus) at 15x — that is, to reclassify Micron from a cyclical to a structurally advantaged oligopolist. That reclassification is precisely what the take-or-pay agreements and the $18B of customer deposits are evidence for, and it is what the most credible bull in the knowledge base is arguing. It is not the base case and the dive does not need it.

The bear at $450 needs no re-rating at all — only reversion. Through-cycle EPS of ~$30 at 15x. Given that this company earned −$5.34 two years ago, a $30 through-cycle assumption is not a punitive scenario; it is a moderate one.

6c. Variant perception (where we differ, what would surprise)

• We differ from the street on the meaning of the low multiple, not on the earnings. The street's own target consensus of $1,575.91 is 10.2x the FY2027 consensus — meaning that even at their targets, analysts are not paying a normal multiple for FY2027 earnings either. The entire $1,575.91-versus-$892.67 gap is a multiple argument (10.2x versus 5.8x), not an earnings argument. Almost no commentary frames it that way. Watchable: whether targets are revised after the 2026-09-22 print, and in which direction.
• We differ on the ceiling clause, which we have seen discussed nowhere. The 10-Q states the largest strategic agreements "generally have a ceiling price for existing products that approximates the market price in the second calendar quarter of 2026." Micron has contractually capped its participation in further price appreciation on its largest volumes at approximately today's level. The market is treating the agreements as pure upside; they are a collar. Watchable: the proportion of revenue under agreement — a tracked claim from 2026-06-28 asserts the agreements will be "~half of revenue," which if true makes the cap material to FY2027.
• We differ on the free-cash-flow story. At the top of the cycle, with a 55.9% net margin, the corrected free-cash-flow yield is ~2.9% and capex consumes 49.1% of operating cash flow. The bull case is generally argued on earnings; it should be argued on cash, and on cash it is much weaker. Watchable number: FY2027 capital expenditure guidance at the September print.
• We do NOT differ on the demand. The shortage is real, the filing documents it, inventory is flat in dollars while revenue quadrupled, bit shipments are supply-constrained at ~20% growth in the middle of the largest compute buildout in history, and no new capacity arrives before 2028. We are not forecasting a demand collapse and the base case does not need one.
• Positive surprise that would force a re-rate upward: the total customer-deposit figure rising materially above $22 billion, disclosed remaining performance obligations rising from ~$5B toward tens of billions, or a FQ4 print with bit shipment growth accelerating while price holds — that combination would be the first genuine evidence of volume-led rather than price-led growth. Watchable: the RPO disclosure in the FQ4 10-Q.
• Negative surprise that would break the thesis: DRAM spot prices rolling over, a CXMT or YMTC capacity announcement in leading-edge DRAM, an inventory build, or — the tail risk almost nobody prices — an injunction in the YMTC litigation. The 10-Q states the YMTC, AMT and Nextech suits "pertain to substantially all of our DRAM, NAND, and other memory and storage products we manufacture, which account for substantially all of our revenue," and each YMTC complaint "seeks an injunction." Proceedings are live in the Unified Patent Court in Dusseldorf, Munich Regional Court, the London Chancery Division and E.D. Texas. Micron states it "cannot make a reasonable estimate of the potential loss."

Synthos fair values

All three anchors are 15x an assumed through-cycle EPS, cross-checked against the FY2027 consensus of $154.67, book value per share of $89.294, and the street target set.

• Bear ~$450 — 15x through-cycle EPS of ~$30 (revenue settling near $110B at a ~40% gross margin, opex ~$14B, 17% tax). Cross-checks: 2.9x FY2027E consensus; ~5.0x a book value that would have compounded to roughly $150 by then; and 50.4% below spot, consistent with a tracked independent claim of 2026-05-12 assigning a "75-80% chance of a 40% decline at some point in the next 12 months." The scenario: prices revert as the filing's own risk factor contemplates, the take-or-pay floors cover only part of the book, and the ~$27B capex lands into a softening market. −49.6%.
• Base ~$1,050 — 15x through-cycle EPS of ~$70, which is 45% of the FY2027 consensus and 9.0x the prior-cycle peak. It concedes a genuine and permanent structural step-change while refusing to capitalise an 85% gross margin. Cross-checks: 14.3x FY2026E, 6.8x FY2027E, 11.8x book. And a deliberately uncomfortable one — $1,050 is 33% below the street's $1,575.91 consensus and 4.5% below the street's LOWEST published target of $1,100. We are outside the entire range of 70 analysts on the low side and we think that is the correct place to be. +17.6%.
• Bull ~$1,700 — 15x through-cycle EPS of ~$113, i.e. 73% of the FY2027 consensus proving durable. Cross-checks: 11.0x FY2027E; 8% above the street consensus target; below the $2,200 high; 40% above the prior 52-week high of $1,213.56. The scenario: take-or-pay floors genuinely de-cyclicalise the trough, trough margins hold above prior-cycle peaks, and memory re-rates from a commodity multiple to an oligopoly multiple. +90.4%.

Base is +17.6% above spot; the tails are −49.6% / +90.4%, roughly 1.8:1 to the upside. Probability-weighted at 25/50/25 the expected return is +19.0%. That is a positive but not a compelling expected return on a beta-2.142 instrument, which is exactly why the verdict is Tactical and staged rather than Core.

7. Knowledge base — 259 raw, 179 naming Micron, 162 used

Raw hits: 259. Entity matches naming Micron (MU or Micron): 179. Used: 162. Discarded: 97.

Search terms: MU, Micron, HBM, DRAM across 51,928 distilled claims.

Discards, with reasons:

1. 77 match: text hits discarded — claims where "DRAM" or "HBM" appears in prose about Samsung, SK hynix, SanDisk, Western Digital or the memory theme generally without naming Micron as an entity. Supporting colour at best; not first-class.

2. 3 entity hits on HBM/DRAM as technology tags discarded — the entity is a technology, not a company.

3. 13 claims from the quarantine_misattributed channel discarded on attribution grounds, irrespective of content. Several are near-duplicates of genuinely attributed Visser claims; using them would double-count a single voice.

4. 4 claims from MU_mgmt excluded from the conviction pool as the company's own voice, and quoted separately below at explicitly low weight.

A note on the expected false positives that did not materialise. "MU" is a short token and the brief anticipated collisions. In practice all 143 MU entity hits are genuine — they co-occur with SK hynix, Samsung, SNDK, WDC, NVDA, MRVL and the semiconductor complex. The false-positive problem here is not the ticker; it is concentration.

The concentration problem, stated plainly, because it governs how much weight this lane can carry. Of 162 used claims, 84 (52%) come from a single voice, Jordi Visser. And that voice has reversed repeatedly:

> 2026-05-11 · neutral · conviction 60 · Jordi Visser — "Sold two-thirds of his DRAM stocks (incl. Micron, $100→$660); DRAM is in the fifth or sixth inning and exhausted, with better double opportunities in earlier-inning names."

> 2026-05-23 · bearish · conviction 70 · Jordi Visser — "Exited Micron, won't re-enter at any price; memory/DRAM names up 4-8x, retail-heavy (ETF $10B+ fast), long in the tooth — need a pause/correction."

> 2026-08-02 · bullish · conviction 70 · Jordi Visser — "Buying Micron below 900; it broke below a head-and-shoulders neckline and then back above it — constructive after being oversold."

A trading log is not a conviction lane. These claims are informative about positioning and crowding — genuinely so — but they carry almost no information about durable value, and they are the reason the conviction rating is Moderate rather than High despite 162 claims and 22 channels. The 2026-08-02 claim is nonetheless notable for timing: it names $900 as an accumulation level two days before this dive, and MU closed at $892.67.

The bear lane is smaller, independent, and has been right about the last eight weeks. It is quoted at full weight.

> 2026-07-08 · bearish · conviction 68 · Mike Green, independent — "Nobody who has studied memory cyclicality expects Micron's or SK Hynix's current margins to persist; the question is only how far they fall."

> 2026-07-24 · bearish · conviction 80 · Mike Green, independent — "Roughly two-thirds of the move in memory/semi names is a mechanical function of levered-ETF flows, not fundamentals; multiples should have contracted, not expanded."

> 2026-05-27 · bearish · conviction 55 · Jan van Eck, independent — "Wary of memory stocks — no durable moat, entrants will come; profits are exploding from higher prices not volume, so customers will economize. Feels 'bubblelicious' — but not yet cutting active-fund exposure."

That last claim is the most valuable in the entire lane, because it was made independently and the filing subsequently confirmed its central factual assertion to the decimal. "Profits are exploding from higher prices not volume" is precisely what the 10-Q's +low-260% price / +low-20% volume decomposition says.

> 2026-07-23 · bearish · conviction 65 · Fred Teal, independent — "Korean memory makers enjoying huge paydays while Chinese build memory factories as fast as they can, which just means a memory bust in 18 to 24 months."

> 2026-07-29 · neutral · conviction 50 · Doug O'Loughlin, independent — "Chinese memory like CXMT historically dumps at ~10% gross margins to win share and may eventually pressure prices, but near-term it's #4 in a shortage market and will make money; Apple already buying CXMT over Micron."

The most useful neutral claims are specific and numerate, and one of them lands exactly on the filing:

> 2026-06-25 · neutral · conviction 60 · Angelo Zino, independent — "Micron's ~85% gross margins are near a peak, level off in the mid-80s for a couple quarters, then normalize toward mid-70s — still above prior-cycle peaks."

The same speaker raised his price target to $1,500 on the same date, citing "long-term customer agreements improve revenue and free-cash-flow visibility." A bull whose margin forecast is a decline is the most credible kind, and his mid-70s normalisation is above our own through-cycle assumption — i.e. the informed specialist view is more optimistic than our base case, which we record as a check against our own bearishness.

> 2026-06-23 · bearish · conviction 62 · compound_and_friends — "Memory/semi stocks severely overextended — semi index 68% above its 200-day MA, most stretched in 25 years; Micron 161% above; pullback severely overdue, need to sober up."

This one is directly checkable against today's payload and it has partly resolved: Micron closed 70.1% above its 200-day average, down from the 161% cited six weeks ago. The pullback the claim called for happened.

The strongest bull claims from independent voices:

> 2026-05-15 · bullish · conviction 78 · Gavin Baker, independent — "Holding memory 'for dear life'; though 25 years of cycles say sell, this may be the first true capacity cycle like the mid-90s, so still very early — takes the over on every number."

> 2026-07-13 · bullish · conviction 85 · Andreas Steno, independent — "Memory/semis haven't peaked; spot prices still accelerating and memory is less cyclical now. Tempting to buy this dip; peak likely late 2027/early 2028." — the same speaker discloses on the same date that he holds "~9-10% of portfolio in memory stocks, over 25% in semiconductors overall — 'I have my money in the trade so I think it will go up.'" That self-disclosure is admirable and it is also a reason to discount the claim.

Low-weight and management-sourced claims, quoted for completeness and used for nothing:

> 2026-06-24 · MU_mgmt · conviction 85 — "Management guides fiscal Q4 2026 outlook even stronger than record Q3."

> 2026-06-24 · MU_mgmt · conviction 80 — "Multi-year Strategic Customer Agreements will significantly enhance durability and predictability." — both are the company's own voice at half weight; the second appears near-verbatim in the 10-Q, so it is the filing, but it remains forward-looking management framing.

Twelve claims come from Jensen Huang, who is Micron's largest-adjacent customer and a promoter of the buildout that determines Micron's price. His assertion that "this memory cycle isn't the historical boom-and-bust but a structural change" is directionally the bull thesis, and it is discounted to low weight on the same principle applied to promoter-class voices. Eight claims sit on Pompliano-branded channels, where the standing rule assigns the host low weight and treats guests as first-class; the guests here (Dan Ives, Avi Felman) are quoted on that basis.

Net stance: 122 bullish / 22 bearish / 18 neutral among used claims — positive-moderate. Stripping Visser's 84, the remainder is 52 bullish / 14 bearish / 12 neutral — still net positive, and a better representation of the panel.

8. Data integrity — what we rejected from the vendor file and why

Nine material items. Four are cases where a filing contradicted the vendor.

1. est.ebitdaAvg and est.ebitAvg are FABRICATED PLACEHOLDERS, not estimates — REJECTED ENTIRELY. This is not the usual corruption; it is arithmetic. For every fiscal year FY2024 through FY2030, ebitdaAvg equals exactly 50.000% of revAvg and ebitAvg equals exactly 33.000% of revAvg. FY2027: $249.491B × 0.50 = $124.746B ✓ and × 0.33 = $82.332B ✓. Seven consecutive years at two constants to five significant figures is not a forecast. The rows are also internally impossible: FY2026 ebitdaAvg of $64.714B sits below niAvg of $84.093B — EBITDA cannot be less than net income. Confirmed against actuals: FY2025 actual EBIT was $9.870B against the placeholder's $12.274B, a 24% error. All EBITDA and EBIT estimate rows rejected; forward valuation runs on epsAvg.

2. ratios_ttm.ebitdaMarginTTM carries the same 50%-of-revenue placeholder — REJECTED, and all EV/EBITDA multiples with it. ebitdaMarginTTM of 0.49806 × revenue $90.274B = $44.96B of implied trailing EBITDA — below the trailing operating income of $59.366B, which is impossible by construction. evToEBITDATTM (22.009), enterpriseValueMultipleTTM (22.009) and netDebtToEBITDATTM (−0.414) are all built on it and are rejected. Real trailing EBITDA is operating income $59.366B plus depreciation and amortisation (FY2025: $8.352B), so approximately $68B, giving a corrected EV/EBITDA of ~14.5x rather than 22.0x. Note that ebitMarginTTM (0.65868 → $59.46B) is correct and ties to actual operating income — only the EBITDA line is synthetic.

3. FILING CONTRADICTS VENDOR — the balance sheet is nine months stale and understates cash by ~$20B. The vendor's most recent balance sheet is FY2025 (2025-08-28), showing $10.307B of cash plus short-term investments and $15.278B of total debt. The 10-Q states cash and marketable investments of $30.13B at 2026-05-28 and discloses $9.38B of debt repayments in 9M FY2026 (corroborated by the 8-K of 2026-04-01 announcing cash tender offers for six series of senior notes). We use the filing. Corrected position: ~$23.7B net cash, versus the $5.636B of net debt the vendor's balance sheet implies. Even at the identical FY2025 date the two disagree — the filing says $11.94B, the vendor says $10.307B, a $1.63B gap of longer-dated marketable securities.

4. km_ttm.enterpriseValueTTM ($989.554B) omits marketable investments — corrected to ~$984.4B, but the error is small here. The vendor's EV implies net cash of $18.619B against our filing-derived $23.7B, so it understates net cash by ~$5.1B (27.6%) and overstates enterprise value by ~$5.1B. In direction this is the textbook defect. In magnitude it is 0.52% of enterprise value and changes no conclusion. Recorded for completeness, not inflated.

5. FILING CONTRADICTS VENDOR — free cash flow is understated by ~$3B because the vendor's capex is gross of government incentives. The vendor's derived trailing capex of $25.261B ties to the filing to within $4M ($5.657B implied FQ4'25 + $19.60B disclosed 9M FY2026 = $25.257B) — which proves it is the gross PP&E line and excludes the separately disclosed "proceeds from government incentives to offset capital expenditures" of $2.99B in 9M FY2026. Corrected trailing free cash flow is ~$29.2B, not $26.171B, and the free-cash-flow yield is ~2.9%, not the vendor's 2.596%. evToFreeCashFlowTTM (37.810) and priceToFreeCashFlowRatioTTM (38.521) are correspondingly overstated.

6. FILING CONTRADICTS VENDOR — the segment series is off-basis and discontinuous; REJECTED as a series. Note 17 of the 10-Q reports four business units: CMBU, CDBU, MCBU, AEBU. The vendor's seg_prod contains none of them, giving a DRAM/NAND technology split for FY2025 and the superseded CNBU/EBU/MBU/SBU names for FY2024 and earlier. The FY2025 technology split also fails to tie: $28.578B + $8.503B = $37.081B against reported revenue of $37.378B, a $297M (0.8%) shortfall. We use the filing's four-unit structure and its disclosed growth rates, and the FY2025 DRAM/NAND split for mix only. The dollar segment tables did not survive text extraction from the 10-Q, which is stated as a gap rather than estimated around.

7. FILING CONTRADICTS VENDOR — the fiscal calendar is wrong in the estimates block, and FQ4'26 is a 14-week quarter. Every row of est is dated -08-28 (2023-08-28 through 2030-08-28), but the actual fiscal year-ends in inc_a are 2023-08-31, 2024-08-29 and 2025-08-28 — the est dates are a rolled-forward placeholder. More materially, the 10-Q states that "fiscal 2026 contains 53 weeks" and "the fourth quarter of 2026 contains 14 weeks" against 13 weeks in FQ3. FY2026 therefore ends ~2026-09-03, not 2026-08-28, and the FQ4 consensus of +21.7% sequential revenue growth is +13.0% on a per-week basis. No vendor field carries this and it materially changes how the September print should be read.

8. The debt components do not tie, and capitalLeaseObligations partially overlaps longTermDebt. FY2025: short-term debt $634M + long-term debt $14,017M + capital lease obligations $701M = $15,352M against a reported totalDebt of $15,278M — a $74M (0.5%) failure to tie, indicating partial double-counting of finance leases inside long-term debt. Immaterial at this scale and flagged only because the same defect is material elsewhere. Note also the 10-K discloses $1.16B of executed-but-not-yet-commenced finance lease obligations at FY2025 year-end that appear on no balance sheet.

9. Internally inconsistent receivables metrics, and a 52-week range disagreement. daysOfSalesOutstandingTTM reads 125.4 days (from receivablesTurnoverTTM 2.9097, implying ending receivables of ~$31.0B), while averageReceivablesTTM of $24.169B against trailing revenue of $90.274B gives 97.7 days. The two cannot both be right; we report the range and rely on the direction, which the 10-Q independently confirms ("a significant increase in receivables due to higher revenue"). Separately, the tech block reports a 52-week high of $1,213.56 and low of $107.77, while quote reports $1,255 and $106.75. We use the tech figures throughout because they are internally consistent with pct_from_hi (−26.442%) and pct_from_lo (+728.31%); on the quote high the drawdown would be −28.9%. Likewise tech gives SMA50 $969.41 / SMA200 $524.82 against quote's $964.85 / $518.10; tech is used.

Also flagged, not rejected. priceToEarningsGrowthRatioTTM of 0.0284 is an arithmetic artifact of peak-cycle growth and is meaningless — rejected. grahamNumberTTM of $299.83 (price is 3.0x it) and grahamNetNetTTM of $17.90 are not applicable to a capital-intensive manufacturer and are not used. capexToDepreciationTTM of 2.803 is correctly reported and is itself a striking figure — Micron is spending 2.8x its depreciation, which is what a genuine capacity cycle looks like.

Not defects, correctly reported and independently confirmed: trailing revenue of $90.274B and net income of $50.473B tie to the quarterly series exactly; netProfitMarginTTM (0.5591) and priceToEarningsDilutedRatioTTM (20.205) both reproduce; bookValuePerShareTTM ($89.294) and priceToBookRatioTTM (9.997) are mutually consistent at spot; incomeQualityTTM of 1.019 (operating cash flow $51.432B ÷ net income $50.473B) is correct and is a genuine quality signal; and the FY2025 geographic split sums to reported revenue exactly.

Non-equity tripwire — check performed and passed emphatically. MU is common stock, not a preferred, baby bond or depositary unit. Price $892.67 is nowhere near a $25 or $1,000 par; beta is 2.142, the opposite of bond-like; the dividend is discretionary and was raised from $0.115 to $0.15 in FQ3'26 rather than fixed; volume is 36.0M shares; the 52-week range spans $107.77 to $1,213.56, an eleven-fold band and the least narrow in this batch; isEtf, isFund and isAdr are all false; and the Form 4 filings name the security as "Common Stock."

9. Technicals — a bounce inside a correction

• Price $892.67, +7.62% ($63.17) on the day. Opened $865.39 from a prior close of $829.50, ranged $858.50–$900.59, and closed near the high of the day on 36.0M shares.
• −26.44% from the 52-week high of $1,213.56, which is also the maximum drawdown from peak over the trailing year — the stock is at its worst point of the year relative to its own high.
• 7.9% BELOW the 50-day moving average of $969.41. Today's 7.62% gap did not reclaim it; the close is $76.74 short.
• +70.1% above the 200-day moving average of $524.82.
• RSI14 49.2 — precisely neutral, neither oversold nor overbought. MACD −34.10 — negative.
• Relative performance: 3-month +54.9% vs SPY +7.6%; 6-month +103.9% vs SPY +11.1%; 12-month +751.1% vs SPY +24.3% and QQQ +30.8%.

Today's move and what it does to the entry

This is the question, and the answer for Micron is different from the rest of the batch.

Today was a violent, sector-wide semiconductor advance: INTC +10.92%, MRVL +12.81%, LRCX +7.85%, MU +7.62%, AMD +7.00%, KLAC +6.95%, AMAT +5.48%, ASML +4.22%. This was beta, not news. Micron reported six weeks ago and does not report again for 49 days; nothing company-specific happened today.

But Micron is not gapping to a new high, and that distinction is the whole entry case. LRCX, KLAC and AMAT gapped inside uptrends. Micron gapped inside a 26.4% drawdown, from below its 50-day average, with a negative MACD. The stock peaked at $1,213.56, fell as much as 26.4%, and today bounced 7.62% without reclaiming resistance. You are not being asked to buy a breakout. You are being asked to buy the third or fourth bounce in a correction, at a price 8% below the 50-day average.

Is that acceptable? Partly, and with conditions. Three things argue that the correction has done real work: the excess over the 200-day average has more than halved from the 161% a tracked claim measured on 2026-06-23 to 70.1% today; RSI at 49.2 is neutral rather than stretched; and the drawdown has already delivered roughly half of the 40% decline that a tracked bearish claim of 2026-05-12 assigned a 75-80% probability. Two things argue against buying today specifically: the close is below the 50-day average, which means the downtrend has not been invalidated, and a 7.62% single-day move on no company news is exactly the kind of day on which a position bought in full is bought 8% worse than it needed to be.

The honest answer: do not buy the gap. Stage in over six to ten weeks in three or four tranches. A reasonable structure, stated as arithmetic rather than advice: an initial tranche at or below current levels (the most-tracked knowledge-base voice was on record buying below $900 two days ago, and the close of $892.67 is inside that zone); a second on a reclaim and hold of the 50-day average at $969.41, which would invalidate the downtrend; a third only after the 2026-09-22 print, which removes the largest single source of variance. Keep meaningful capital in reserve for the bear case, because a −49.6% move to $450 is a real scenario in a name with a beta of 2.142 and realized volatility above 100.

A cross-reference within the batch. The wafer-fab-equipment names (LRCX, KLAC, AMAT, ASML) are one trade with four expressions and they are levered to Micron's capital expenditure — the approximately $27 billion net that this dive treats as a cash-flow burden is their revenue. An investor holding both MU and the equipment names is doubled up on the same variable in a way that is not obvious: if memory pricing rolls, Micron's earnings fall and its capex is cut, hitting both positions from the same cause. Size accordingly.

10. Insiders — the only activity is the chief executive selling

Every one of the eight insider transactions in this file is a sale by chief executive Sanjay Mehrotra, all executed on 2026-07-24 and filed 2026-07-28. There is no insider buying of any kind.

The reading, stated carefully. Eight tranches on a single day at incrementally rising prices from $942.87 to $954.19 — including a one-share execution — is the unmistakable signature of a laddered limit order under a pre-arranged trading plan, not a discretionary decision to exit. The payload contains no field identifying a Rule 10b5-1 plan, so we cannot confirm it and say so.

Proportion matters more than the headline. Roughly 5,469 shares against a holding of approximately 312,168 is about 1.8%, worth roughly $5.2M, leaving the chief executive with 307,749 shares worth approximately $275M at today's close. This is not abandonment; it is routine diversification by an executive with a quarter of a billion dollars of exposure to one stock.

Two observations that do carry weight. First, the sales were executed at $943–$954, roughly 6% above today's close — the insider sold higher than where you would be buying. Second, and more important: in the quarter in which Micron produced the largest print in the history of the memory industry, the only insider transaction of any kind was a sale, and the company itself repurchased zero shares. Neither is damning. Together they are a modest negative and they are recorded as such.

11. Verdict, kill-criteria and flip conditions

Buy — Tactical.

What you are buying. A company at the centre of the binding physical constraint on artificial-intelligence compute, one of three credible suppliers of leading-edge DRAM and HBM worldwide, that has just reported $41.456B of revenue at an 85% gross margin — verified against four independent disclosures in the 10-Q — with roughly $23.7B of net cash, inventory flat in dollars while revenue quadrupled, five consecutive earnings beats, a supply response that physically cannot arrive before 2028, and a genuinely unprecedented structural development: multi-year take-or-pay agreements under which customers are committing approximately $18 billion of cash deposits and $22 billion of total financial commitments. It sits 26.4% below its high after a correction that has already halved its extension over the 200-day average.

What you are accepting. That roughly 80% of the earnings step-change is price, not volume — the company's own filing decomposes it that way. That the 85% gross margin will fall, on which literally every participant agrees including the analyst who raised his target to $1,500. That the 5.8x forward multiple is not cheapness but the market's estimate of the transitory fraction, and that at a normal multiple today's price already capitalises roughly $59.51 of permanent earnings against a six-year average of $3.03 and a prior peak of $7.74. That the free-cash-flow yield at the top of the greatest cycle in the industry's history is 2.6-2.9%, because capex takes 49% of operating cash flow and is guided to approximately $27 billion net. That the buyback was zero last quarter with $2.16B of authorisation left. That the disclosed contractual backlog is ~$5 billion, and the largest agreements cap the price at roughly today's level. And that a majority of DRAM output comes from Taiwan, that Chinese state-backed capacity is named in the filing as an oversupply threat, and that live litigation seeks injunctions against products accounting for "substantially all" of revenue.

Why Tactical and not Core, and not Watch. Not Core, because the base case is only +17.6% with a −49.6% bear on a beta-2.142 instrument, because the conviction lane is 52% one reversible voice, and because we sit below the entire published street target range. Not Watch, because the structural evidence is contractual rather than narrative — $18 billion of customer cash is not an opinion — because the correction has already delivered a real discount, because the balance sheet genuinely underwrites the downside, and because a Watch on a name that could compound another 90% on the resolution of a question the filing says cannot be answered before 2028 is a false economy. A tactical, staged, modestly sized position is the position the evidence supports.

Pre-registered KILL criteria — what would break the thesis:

• DRAM or NAND average selling prices declining sequentially in any reported quarter, with bit shipments flat or down. This is the thesis in one variable. The filing's own risk factor records annual declines of up to a high-40% range.
• Inventory building materially above ~$10B while revenue is flat or falling. Flat inventory at $8.4B is currently the cleanest evidence for the shortage; a build reverses it.
• Gross margin falling below 55% on a quarterly basis, which would put through-cycle EPS below the ~$50-$59 the current price already capitalises and make the stock expensive on its own terms.
• A leading-edge DRAM or HBM capacity announcement from CXMT, YMTC or any new entrant at a scale that changes the 2028 supply picture. The filing names both companies and links them to the oversupply threat.
• An injunction granted in the YMTC proceedings in Germany, England or Texas against products representing a material share of revenue.
• FY2027 capital expenditure guided materially above ~$30 billion net without a corresponding step in contracted volumes — that would be the industry building the next glut with shareholders' money.
• Customer deposits failing to convert, i.e. the $22B of expected commitments not appearing as contract liabilities and remaining performance obligations in subsequent filings.
• A decisive break below ~$650 on company-specific news rather than sector beta.

Pre-registered FLIP TO CORE (upsize) conditions:

• Bit shipment growth accelerating above ~30% year on year while pricing holds. This is the single most important positive falsifier: it would convert a price story into a volume story and justify a structural multiple.
• Disclosed remaining performance obligations rising from ~$5B into the tens of billions, which would give the contractual floor real substance.
• A reclaim and hold of the 50-day moving average ($969.41), invalidating the current downtrend.
• Resumption of the buyback in scale, or an expansion of the $10B authorisation — which after a quarter of zero repurchases would signal both that management sees value and that the CHIPS-agreement restriction is not binding.
• A FQ4'26 print on 2026-09-22 beating on a per-week basis, i.e. above roughly $3.60B of revenue per week after adjusting for the 14-week quarter.
• Any insider buying, of which there is currently none.

Where MU fits in the Synthos Framework Portfolio. The AI infrastructure / semiconductor cyclicals sleeve, as a 1.0-1.5% tactical position, built in three or four tranches over six to ten weeks. The size is deliberately below the batch norm for a name of this quality because the volatility is extreme (beta 2.142) and the bear case is a halving. Do not hold MU alongside a full weight in the wafer-fab-equipment names (LRCX, KLAC, AMAT, ASML) without recognising that they are the same trade — Micron's ~$27B of capital expenditure is their order book, and a memory downturn cuts Micron's earnings and their revenue from the identical cause. If the memory thesis is expressed once, this is the highest-quality expression of it, because it holds the net cash and the customer prepayments. Logged as a tracked Synthos call (Buy — Tactical) as of 2026-08-04 at $892.67.

Single biggest risk: the price of DRAM. Roughly 80% of the revenue increase is average selling price. Micron does not set that price, cannot hedge it, and has now contractually capped its participation above roughly the Q2 CY2026 level on its largest agreements while retaining full exposure below the floors on everything not under contract. The company's own risk factors record that within the last five fiscal years DRAM prices have fallen by as much as a high-40% range in a single year, and that "in some prior periods, average selling prices for our products have been below our manufacturing costs." Two years ago, in FY2023, this company reported a −9.1% gross margin and lost $5.34 per share. Nothing in the current structure of the industry makes that impossible again; it makes it less likely, and later.

Most fragile assumption in the price: that the 85% gross margin normalises toward 55-75% rather than reverting toward the 22-45% band of the prior cycle. Every valuation anchor here depends on it — the through-cycle EPS of $30 / $70 / $113 that generates the $450 / $1,050 / $1,700 range is, at root, a gross-margin assumption wearing an earnings costume. The most credible independent forecast we have models a mid-70s normalisation, which is above our base case. If instead the answer is the prior cycle's 45% peak, through-cycle EPS is closer to $30 and the bear case is the right one. The filing will answer this question one quarter at a time, and the first instalment arrives on 2026-09-22.

Provenance & disclosures

• Traceability: 259 raw knowledge-base hits across the terms MU, Micron, HBM, DRAM over 51,928 distilled claims. 179 are entity matches naming Micron; 162 are used (breadth 22 channels, 28 named speakers, 2023-01-27 to 2026-08-02, net conviction positive-moderate). 97 discarded: 77 match: text hits about the memory theme without naming Micron as an entity; 3 entity hits where the entity is a technology tag (HBM/DRAM) rather than a company; 13 from the quarantine_misattributed channel discarded on attribution grounds regardless of content, several of which duplicate genuinely attributed claims; and 4 MU_mgmt claims excluded from the conviction pool as the company's own voice and quoted separately at low weight. Contrary to expectation, the short token "MU" produced no false positives — all 143 such hits co-occur with memory-sector entities. The material limitation is concentration, not collision: 84 of 162 used claims (52%) come from one voice who reversed position at least six times between 2026-05-11 and 2026-08-02. Twelve claims from Jensen Huang are discounted to low weight as an adjacent-customer and promoter-class voice; eight Pompliano-channel claims follow the standing rule of low weight to the host and first-class treatment of guests. All quotations are verbatim from stored claim text with date, stance, conviction and source stated.
• Data as-of: income statement through FQ3'26, quarter ended 2026-05-28, per the 10-Q filed 2026-06-25 · balance sheet and cash flow in the vendor payload are FY2025 (2025-08-28) and nine months stale — rebuilt from the 10-Q, Section 8 · estimates 2026-08-04 · prices 2026-08-04 (quote timestamp 1785873601 = 2026-08-04T20:00:01Z, the 16:00 ET close; $892.67, +7.62%; SMA50 $969.41, SMA200 $524.82, RSI14 49.2, MACD −34.10) · knowledge-base claims 2026-08-04. Fiscal/calendar mismatch: Micron's fiscal year ends on the Thursday closest to 31 August. FY2026 contains 53 weeks and ends ~2026-09-03; FQ4'26 contains 14 weeks. All "FY" references are fiscal unless stated.
• Filing-versus-vendor contradictions, all resolved in favour of the filing: (i) balance sheet — vendor's latest is 2025-08-28 showing $10.307B cash and $15.278B debt; 10-Q states $30.13B cash and marketable investments at 2026-05-28 and $9.38B of 9M debt repayments, giving ~$23.7B net cash against the vendor's implied $18.6B and its balance sheet's $5.636B of net debt; (ii) free cash flow — vendor capex is gross of the $2.99B of government incentives the 10-Q discloses separately, so trailing FCF is ~$29.2B not $26.171B and the yield ~2.9% not 2.596%; (iii) segments — 10-Q Note 17 reports CMBU / CDBU / MCBU / AEBU, none of which appear in the vendor's seg_prod, which carries a DRAM/NAND split for FY2025 and superseded unit names for FY2024 and earlier, and fails to tie by $297M; (iv) fiscal calendar — 10-Q states FY2026 has 53 weeks and FQ4'26 has 14 weeks, so the +21.7% sequential revenue consensus for the September quarter is +13.0% per week, a fact no vendor field carries. Additionally the 10-Q gives cash and marketable investments of $11.94B at 2025-08-28 against the vendor's $10.307B at the same date.
• Data rejections (detailed in Section 8): est.ebitdaAvg and est.ebitAvg rejected as fabricated placeholders — exactly 50.000% and 33.000% of revAvg for every year FY2024–FY2030, and internally impossible (EBITDA below net income in all forward years); ratios_ttm.ebitdaMarginTTM (0.498) rejected on the same 50%-of-revenue construction, implying trailing EBITDA of $44.96B below trailing operating income of $59.366B; evToEBITDATTM, enterpriseValueMultipleTTM (both 22.009) and netDebtToEBITDATTM (−0.414) rejected as derived from it, with a corrected EV/EBITDA of ~14.5x; enterpriseValueTTM corrected from $989.554B to ~$984.4B (a 0.52% error, recorded but not inflated); evToFreeCashFlowTTM (37.810) and priceToFreeCashFlowRatioTTM (38.521) overstated by the gross-capex treatment; seg_prod rejected as a series; est date column rejected as a rolled-forward placeholder; debt components fail to tie by $74M (STD + LTD + capital leases = $15,352M against totalDebt $15,278M), indicating partial finance-lease double-counting, immaterial at this scale; daysOfSalesOutstandingTTM (125.4) internally inconsistent with averageReceivablesTTM (97.7 days), both reported; priceToEarningsGrowthRatioTTM (0.0284) rejected as a peak-cycle artifact; grahamNumberTTM and grahamNetNetTTM not applicable; tech 52-week high/low ($1,213.56 / $107.77) and moving averages used in preference to quote ($1,255 / $106.75), the discrepancy stated.
• Basis note: the analyst estimate series is non-GAAP; reported GAAP diluted EPS for FQ3'26 is $24.67 ($28.243B ÷ 1.145B diluted shares, which ties exactly) against the $25.11 on which earn_cal and the estimates are struck, a ~$504M difference. All forward multiples are non-GAAP-to-non-GAAP; the trailing 20.2x is GAAP. The FQ3'26 earnings release itself (Exhibit 99.1 to the 8-K of 2026-06-24) is not present in the extracted filing text — the 8-K is a two-page cover document — and the 10-Q's financial statement tables did not survive text extraction. The $41.456B quarter and the 85% gross margin are nonetheless confirmed by four independent percentage disclosures in the 10-Q MD&A (+346% year on year, +74% sequentially, +203% nine-month, and 85%/74%/38% margins), each of which reproduces the vendor series to within rounding. The $25.11 EPS could not be directly verified and is flagged as such.
• Estimate coverage: 22 analysts on FY2026 EPS, 23 on FY2027, 12 on FY2028, 5 on FY2029, 7 on FY2030. Dispersion widens sharply: FY2027 spans $129.31–$216.83 and FY2028 spans $82.29–$200.85, a low estimate 49% of the mean. No conclusion rests on any row beyond FY2027, and FY2027 is used with its dispersion visible.
• Street note: 57 buy, 11 hold, 2 sell, zero strong buy, zero strong sell — 81% buy across 70 ratings. Target consensus $1,575.91, median $1,512.50, high $2,200, low $1,100 — 23.2% above spot. Not one published target sits below the current price. That target set implies 10.2x FY2027 consensus EPS at the mean and 7.1x at the low, so even the street is not paying a normal multiple for FY2027 earnings; the entire gap to spot is a multiple argument, not an earnings argument. The set also appears anchored to higher prices — a tracked target increase to $1,500 was recorded on 2026-06-25, since when the stock has fallen materially — and a consensus that has not moved while the stock fell 26.4% from its high is drifting rather than converging. Treated as a sentiment reading, not a valuation input.
• Peer note: the vendor peer set (AMAT, ARM, CRM, CSCO, IBM, KLAC, LRCX, QCOM, SAP, TXN) contains no memory peer whatsoever — Samsung, SK hynix, Kioxia and SanDisk, the actual competitors named in the 10-K, are all absent, and half the set (CRM, IBM, SAP, CSCO) is enterprise software and networking. No peer-multiple comparison is drawn. Batch cross-references to LRCX, KLAC, AMAT and ASML use only the prices supplied in the batch table and note that those names are levered to Micron's capital expenditure rather than its pricing.
• Fair-value caveat: the $450 / $1,050 / $1,700 anchors are 15x an assumed through-cycle EPS of ~$30 / ~$70 / ~$113, cross-checked against FY2027 consensus EPS of $154.67, book value per share of $89.294 and the street target set. These are scenario arithmetic on a normalised-earnings assumption, not a discounted cash flow, and the through-cycle EPS figures are our estimates, not consensus. The base of $1,050 sits below the lowest published street target of $1,100, deliberately. The dominant uncertainty is the gross-margin normalisation path, which the FY2020–FY2025 record shows ranging from −9.1% to +45.2% against 85% today.
• Timing: FQ3'26 was reported 2026-06-24, forty-one days before this dive, so the most recent quarter is fully reflected. The next print is 2026-09-22, forty-nine days away — outside any banner threshold, but this dive states explicitly that a memory cycle can turn inside a 49-day window, and the staged entry in Section 9 is constructed around that fact.
• Not investment advice. Independent research, educational and informational only, never personalised. No recommendation to buy, sell or hold any security is made to any person.
• Version: 2026-08-04-full.