86% margins, a $51B quarter, and a market that still refuses to pay up — Wednesday decides whether this cycle is structural or just the best-dressed peak in memory history.
中文版 Chinese edition →The call: Hold starter MU $1,040–1,080 into print, no new full size tonight — add washout $980–1,000 or on confirmed Q1 guide raise; target $1,200, invalidation daily close <$980, 1–2 weeks, SPECULATIVE.
Why now: Fiscal Q4 prints Wednesday Sep 30 (LSEG $31.61 / $51.07B, ~86% gross margin) — the beat is expected after 10 straight beats; guidance and 2027 pricing language is the entire trade.
The disagreement: Consensus fears a classic peak (Morningstar fair $850, Burry 2027 puts); I think 16 take-or-pay contracts plus HBM4 scarcity keep 2027 tight, but only ~25% of revenue is actually floored — this is a better cyclical, not a non-cyclical.
The level that changes everything: Q1 FY27 gross-margin guide above 80% validates the structural case and opens $1,335; below 78% with price-decline language confirms the moderation Burry is betting on.
Micron reports fiscal Q4 on Wednesday, September 30, and it is the most important single-stock print left in this AI tape. Not because the quarter is in doubt — LSEG expects $31.61 on $51.07B with ~86% adjusted gross margin, a company record — but because Micron is generating software-like margins while trading like a steel mill, and one of those two prices is wrong.
Q3 FY26 was not a beat, it was a regime break. Revenue $41.456B (+346% YoY) vs $35.82B consensus, non-GAAP EPS $25.11 vs $20.71, gross margin 84.9% and operating margin 81.2% — both company records. CFO Mark Murphy called the $17.6B sequential revenue jump the largest in company history. Shares rose 15.7% on June 24, then gave back 28% to $904 by mid-July in a sector-wide memory repricing that also took SK Hynix down 9% overnight.
Into this print the setup is crowded and contradictory: JPMorgan calls the setup "constructive" on tight DRAM/NAND and resilient pricing, Citi just raised estimates on better DRAM pricing, KeyBanc holds a Street-high $1,750 after walking the Asia supply chain — while Michael Burry is loading 2027 puts and publicly predicting when the shortage ends, and Benzinga notes Micron has beaten in each of its last 10 reports yet typically falls the next day. Both sides can cite the tape.
Micron guided Q4 to ~$50B revenue, ~86% gross margin and ~$31 EPS — and the CFO himself pre-announced the catch: a "meaningful moderation in the rate of price increases." That sentence is the trade. If Micron hits 86% on mix and yield with flat prices, it proves non-HBM DRAM can carry software margins — Chief Business Officer Sumit Sadana already disclosed non-HBM DRAM margins have at times exceeded HBM. If it needs another leg of double-digit price hikes to get there, Burry's clock starts ticking.
The bull mechanics are real: CEO Sanjay Mehrotra says Micron can fulfil only 50–67% of customer demand medium-term, has "no line of sight as to when supply catches demand," with new fabs delivering no meaningful output until fiscal 2028. HBM4 is already shipping to Nvidia's Vera Rubin platform with yields ahead of HBM3E, $1B+ shipped, 2026 HBM sold out, and HBM4E entering volume in calendar 2027. My opinion: this is the strongest near-term scarcity Micron has ever entered a print with — but scarcity without contracted price is just a spot market with good manners.
Strategic Customer Agreements are the structural bull case, and the math is both better and thinner than bulls admit. Sixteen signed, $22B in deposits and commitments ($18B cash + ~$4B letters of credit), ~$100B cumulative minimum revenue on 14 of them, with floors management says deliver margins "well above peak quarterly margins in any past cycle" — plus new auto deals with GM, Ford and major suppliers since June.
Then UBS's Timothy Arcuri asked the right question, and Mehrotra answered honestly: only ~20% of DRAM and ~30% of NAND volume is covered — about 25% of revenue projectable over the term. The destination (40% at fixed prices/ceilings, 50%+ under SCAs overall) is a target, not a signature. Deposits sit as unrestricted cash in financing flows and are returned in the back half. My judgment: SCAs raise the floor of the next trough materially, they do not remove the cycle. Three-quarters of revenue still reprices into whatever 2028 supply looks like — including ChangXin's $8.55B STAR listing, the largest-ever Chinese semis IPO, funding a fourth DRAM player at 7.7% share.
Trailing P/E of ~23x on peak earnings is the value trap of every memory peak. Finnhub shows TTM EPS $44.17, 52-week range $154.65–$1,255, beta 2.36. Morningstar raised fair value to $850 from $455 and still calls it overvalued with no moat and Very High uncertainty, expecting price pressure in 2028 and a downturn in 2029. TIKR's mid-case — itself assuming 14% revenue CAGR and 67% net margins, 4.5x the 15.1% five-year average — gets to only ~$935 by August 2030, ~3% total return. The Street mean near $1,490 (range $361–$2,200 across 42 estimates) is a 12-month momentum target, not a through-cycle value.
My fair-value framing: $1,200 on a 12-month base case (Q1 guide holds 80%+ margins, 2027 supply still tight, SCAs expand toward 35% coverage) is ~28x my $43 FY27 EPS — full but earnable if HBM4 mix holds. Below $850 is where Morningstar's through-cycle math lives and where I would get aggressive long. Above $1,335 (the 61.8% Fibonacci extension from the March $311 bottom) is euphoria pricing with no margin of safety into 2028 capacity.
Do not chase $1,069 into the print with full size — this name falls on beats. This updates open call EQ-MU-1 (BUY washout $1,040–1,080, target $1,200, kill daily <$980): we are inside the buy zone, up ~1.4% today with chips leading while SPY faded. Keep the starter, keep the kill.
| Flagship: MU into Q4 print | Level |
|---|---|
| Action / entry | Hold starter $1,040–1,080; add only on washout $980–1,000 or on post-print Q1 GM guide >80% |
| Target | $1,200 base (Q3 $1,255 high retest stretch) |
| Invalidation | Daily close below $980 — breaks the washout shelf and confirms margin-peak repricing |
| Timeframe / conviction | 1–2 weeks / SPECULATIVE (one signal: scarcity; unconfirmed until guide) |
| Audience / size | Active traders only; half normal size into binary; no LEAPS until SCA coverage disclosed above 30% |
Options are pricing ~10–13% move. A beat without a Q1 raise is a fade — history says so. A beat plus Q1 gross-margin guide at/above 80% and a new SCA count above 16 is a trend-confirm that justifies adding into strength toward $1,200. Pair with existing EQ-SNDK-1 and EQ-SOXX-1 longs; do not triple the memory beta — if you hold SNDK $1,712 and SOXX, MU starter is enough.
| Scenario | Trigger | Price path | Weight |
|---|---|---|---|
| Bull: structural print | Q4 GM ≥86% on mix + Q1 GM ≥82% + SCAs >18 | $1,200 → $1,335 Fib ext, then KeyBanc $1,750 debate | 30% |
| Base: beat, moderating guide | Q4 in-line, Q1 GM 78–80%, "normalizing" price language | Chop $980–1,150, fade rips, buy washouts | 50% |
| Bear: peak confirmed | Q4 GM miss or Q1 GM <78% + NAND price cuts | $980 breaks → $904 July low → $850 Morningstar fair | 20% |
I am wrong if Q1 FY27 gross margin guides below 78% with explicit DRAM price declines, or if SCA coverage stalls at 25% while CXMT/CapEx guides accelerate into 2028. Specifically: daily close below $980 invalidates the washout thesis; two consecutive quarters of NAND ASP declines over 5% confirms commodity reversion; a disclosed floor at or below prior-cycle 60s gross margins collapses the structural premium. Any of those flips me from buy-washouts to sell-rallies and I will say so.
Tonight: do nothing new. Hold starter, set $980 alerts, define your add size for $980–1,000 in advance. Tomorrow after close: read the press release for Q1 revenue/GM guide first, listen for the words "moderation" vs "tight beyond 2027," count SCAs, and check HBM4 2027 allocation language. Downstream: a strong MU guide lifts SNDK, STX ($907), WDC ($452), NVDA ($230) and SOXX — a weak guide hits all four harder than MU itself given beta 2.36. No new memory longs into a sub-78% guide, no matter how far it washes.
MU $1,068.96 (+1.42%, O $1,073.35 H $1,082.66 L $1,057.70, prev $1,053.98); SNDK $1,712.51 (-0.02%); STX $907.04 (-1.57%); WDC $452.41 (-0.18%); NVDA $230.46 (+0.70%); INTC $116.75 (+0.62%); SPY $763.27 (-0.31%); QQQ $737.29 (+0.10%); BTC $82,980 (-0.46%); crypto mcap $2.85T (-2.75% 24h), BTC dom 58.22%, fear & greed 73 Greed. Fundamentals (Finnhub): MU TTM P/E 23.25, EPS TTM $44.17, gross margin 72.57%, ROE 70.55%, ROA 49.92%, beta 2.36, 52wk $154.65–$1,255 (high Jun 25 2026), div 0.60%.
Base $1,200 = ~28x $43 FY27 EPS assuming Q1 GM ≥80%, DRAM/NAND pricing flat-to-up low-single digits through H1 2027, SCA coverage rising to ~35%, HBM4 yields on track. Load-bearing input: gross margin persistence. If through-cycle GM reverts to 45–50% (prior peaks low-60s), EPS power halves and fair value collapses toward Morningstar $850 / TIKR $935 — the model breaks to the downside, not the upside. Where this could be wrong: coordinated Samsung/SK Hynix/CXMT capacity adds in 2028 arrive faster than fabs suggest; then $980 fails and $850 is generous.