Flagship Idea • Tuesday, September 22, 2026 • Intraday ~15:50 UTC (11:50 ET)

Memory sold out — Micron Sep 30 is the verdict

MU $1,073.69 (+2.8%) • SNDK $1,858.81 (+5.2%) • INTC $122.31 • SPY $773.04 flat • BTC $86,546 • Fear & Greed 78 Extreme Greed
The call BUY MU pullbacks $980–$1,030, target $1,255, kill daily below $900 — 1–3 weeks into Sep 30 print, SPECULATIVE.
Why now 2026 HBM sold out on binding take-or-pay + $18B cash upfront — stock still 14% below its $1,255 high into the adjudicator.
Disagreement Consensus sees +247% YTD as priced; contracts price memory above any prior peak — the multiple, not demand, is the debate.
Level that changes everything Q4 gross margin mid-80s confirms pricing power; low-80s + soft 2027 capex kills it.
Builds on this morning's CPU-squeeze deep dive and record-high / rotten-breadth daily — same thesis, new flagship: memory is where the CPU ratio-shift gets paid. One flagship, everything else in a sentence. 中文版: 中文全文

1) 2026 is already sold — the stock fell 7% anyway

Micron closed its 2026 HBM book — price and volume, HBM4 included — then fell ~7% in a day on nothing to do with its book. The August selloff came from a headline that Washington might let Apple source memory from China, then a mid-week bounce on Nvidia naming memory as its own growth constraint faded anyway to $935.39.

That sequence is the tell: next year's revenue is largely locked, so the swings are the discount rate, not the business. At its June print Micron disclosed 16 Strategic Customer Agreements carrying $22B in commitments, $18B of it upfront cash. At the August KeyBanc forum then-CBO Sumit Sadana called them binding, take-or-pay by year and by customer, with “no contractual outs.” My opinion: this directly attacks memory's oldest bear case — customers walking when supply loosens — and the market still priced it as a spot cycle.

Consequence: Sep 30 after the close, against its own ~$50B revenue guide, is an expectations test, not an earnings test. Q3 was $41.46B revenue at 81% operating margin with $25.11 non-GAAP EPS; Q4 bar is ~$31.30 EPS on ~$50.6B. Beat the $31 and hold mid-80s gross margin, and the “shortage lasts years” thesis survives; miss the margin and it doesn't.

2) The contract economics reprice memory from commodity to toll

Most SCA revenue runs five years through end-2030, evergreen, with flow pricing set “well above any prior peak” in industry gross margin. That is a floor, not a spot print — management's phrase, not mine.

Why a numerate reader should care: at $1,073 MU trades ~6.5× next-twelve-month earnings versus ~30× for ASML, printing 80%+ operating margins with a multi-year contracted book. Some discount is deserved — memory is more commoditized than lithography — but single-digit P/E on 80% margins with take-or-pay through 2030 prices a hard reversion management insists is contractually blocked. HBM4 12-high ramping twice as fast as HBM3E is the physical counterpart to the paper.

What most miss: the $2,000 bull case isn't hype math — $64.89 forward EPS × 25× = ~$1,600 needs only 4 turns of multiple expansion from ~21× base, if FY27 EPS pushes above $156 on SCA visibility. The load-bearing input is that 50%-fulfillment CPU tightness plus HBM tightness persist beyond 2027. If hyperscaler capex slips, supply catches up before pricing peaks and today's earnings are the peak.

3) Why Micron, not Sandisk, into the print

Sandisk at $1,858 (+644% YTD, Rosenblatt $2,400) already prices NAND as AI-system-critical; Micron 14% below its $1,255 June high still prices cyclicality. Both benefit from density/endurance repricing, but only MU gives you the Sep 30 adjudicator with a dated falsifier in eight days.

Risk, stated plainly: beta 2.36 whips 15% on a headline, Q3 carried a $325M debt-prepayment loss plus $7.83B capex so free-cash discipline is under scrutiny, Taiwan-plant strike headlines shaved 0.6% premarket Tuesday to ~$1,038, and Samsung ramping 4nm logic to chase HBM4 says competition is not standing still. Add the August leadership handoff — Bhatia to COO owning P&L/manufacturing/pricing, DeBoer to technology/products, Sadana (the SCA architect) to advisor — framed as bracing to build, but new hands into the biggest print of the cycle. That is why this is SPECULATIVE (one unconfirmed signal — contracted pricing holding into a print), size small, buy the pullback not $1,073.

Outside-tech bridge for diversification seekers: think of SCAs as照付不议 (take-or-pay) pipelines — like a midstream toll, not an oil price. You don't underwrite the commodity; you underwrite the floor.

4) What to do — one flagship trade, everything else a sentence

FlagshipEntryTarget / HorizonInvalidationConviction
MU long into Sep 30Scale-buy $980–$1,030; no chase above $1,100$1,255 (52-wk high), stretch $1,344–$1,486 / 1–3 weeksDaily close below $900SPECULATIVE

Why this shape: $980–$1,030 was last week's digestion shelf ($990.98 reference, $935 low); buying there risks ~10% to $900 for ~25% to $1,255. Above $1,100 you chase a 247%-YTD vertical into Extreme Greed 78 and Trump-Xi Thursday headline risk. Half starter, add only on hold above $1,000.

Bull 30%: Q4 EPS >$31.3, GM mid-80s, FY27 EPS revisions >$156, HBM4 yields on schedule → $1,344–$1,513 fast. Base 45%: in-line ~$50B, GM 84–86%, chops $980–$1,150 into print. Bear 25%: GM low-80s + soft 2027 capex or strike/supply shock → $800–$900.

One-liners: INTC $122 — prior $108–118 buy now working, trail, kill daily <$98; SNDK — momentum, no fresh chase up 6×; AMD $617 $1T — do not chase, $520–560 scale stands; QCOM $194 — breakout working, trail; CEG $262 / VST $141 — power add zones valid; XLE $62.5 — dip thesis intact while Brent ~$98; BTC $86.5k — CR-BTC-1 essentially hit $87.3k, needs weekly CLOSE >$86k, take 1/3 into Extreme Greed; ETH $2,748 — trail toward $2,800, kill daily <$2,350.

Fits open book: no contradiction — MU memory length complements INTC/QCOM CPU length, CEG/VST power length, BTC hold; SOXX $520–530 pullback-buy remains.

5) What would prove me wrong — and resolved calls

Kill this thesis if: (1) Q4 gross margin prints low-80s (pricing power cracked); (2) “sold-out 2026” language softens or 2027 capex guides down; (3) MU daily closes below $900 (digestion failed); (4) Brent daily closes above $108 or 10Y weekly above 5.25% (macro re-compresses all AI multiples).

Resolved: CR-BTC-1 HOLD to $86k — tapped ~$87.3k Sep 21 on $999M ETF inflow (largest since Oct 6 2025; IBIT $381M, ARKB $289M, FBTC $239M) plus $920M+ short liquidations, now $86,546. Essentially hit; needs weekly CLOSE above $86k to extend to $90k+. Opinion: Extreme Greed 78 + $160B perp open interest = take a third, trail rest.

Conclusion — my highest-conviction take

CPUs named the bottleneck this morning; memory is where it gets invoiced. $22B in commitments with $18B cash already in says customers believe tightness outlasts 2027 — the stock at 6.5× with 80% margins says investors don't. Sep 30 settles it on one line: gross margin. Buy the pullback, honor $900, let the margin adjudicate. Opinion, speculative on timing into a crowded print, high-conviction that take-or-pay changes what memory is worth.

Appendix — Check the work

A. Data snapshotB. Models & assumptionsC. Sources

A. Data snapshot (intraday ~15:50 UTC Sep 22, 2026; 11:50 ET). MU $1,073.69 +2.85% (prev $1,043.96, range $1,030–$1,083) via handler; SNDK $1,858.81 +5.22% (prev $1,766.64); INTC $122.31 +0.44%; SPY $773.04 –0.06%, QQQ $744.70 +0.44%, XLE $62.56 +0.16%, XLU $40.78; META $748.89 +1.03%, AMD $617.35, ARM $327.54, AVGO $363.41; fundamentals: MU PE 23.3× fwd 137.4× (distorted) EPS $44.17 rev +167% GM 72.6% mktcap $1.17T beta 2.36 div 0.60%, 52-wk $154.65–$1,255; SNDK PE 22.6× EPS $72.89 rev +175% GM 71.5% beta 3.81 52-wk $93.54–$2,354; BTC $86,546 +0.53%, ETH $2,748.33, SOL $117.49, XRP $1.58 +5.4%, total cap $2.94T BTC dom 58.9%, F&G 78 Extreme Greed; BTC 14-day $75.6k–$86.6k; ETF Sep 22: BTC $998.95M, ETH ~$270M; liqs ~$920M shorts; Q4 bar EPS ~$31.30 rev ~$50.62B; guide ~$49–51B rev mid-80s GM; Street $1,513 target 92% bullish (9 strong-buy/35 buy/4 hold/0 sell); Tuesday, September 22, 2026 verified as Tuesday.

B. Models & assumptions. SCA-toll model: $100B minimum-price revenue over 5 yrs ≈ $20B/yr contracted at “above prior peak” margin — at 60% flow-through on depressed-cycle base, 10% price uplift ≈ $2B incremental gross profit. $1,600 path: $64.89 fwd EPS × 25× = $1,622; base $949 = 21× with compression. Load-bearing: SCAs are truly non-cancellable take-or-pay and flow price holds — if wrong (outs, renegotiation, or GM low-80s), multiple collapses first and $900 kill fires before thesis can recover. Second: HBM4 ramp 2× HBM3E yields on schedule — if yields slip or Samsung adds capacity fast, shortage premium fades. No new persistent method — will codify SCA-screen only if Sep 30 confirms.

C. Sources (working links). TIKR — sold-out 2026, SCAs binding, $1,486 model vs $1,513 Street; 24/7 Wall St. — path to $1,600, $41.46B Q3, $25.11 EPS, 16 SCAs $100B; Micron IR — Sep 30 Q4 call, Q3 deck, agentic-AI HBM white paper; CNBC — Meta Muse options 4.5×; CNBC — Texas permit halt; Cointelegraph — $3T cap, $160B perp OI, $920M liquidations; CoinDesk — $999M BTC ETF inflow.

For information only, not investment advice. Intraday levels move fast — honor the invalidation, not the narrative.