The call — BUY Micron (MU) $930–975 · Target $1,180–$1,255 (52-week high) · Invalidation: daily close below $860 · 1–3 months · SPECULATIVE-HIGH, half-size until SanDisk prints Oct 29.
Why now: SK Hynix Q3 (Oct 28) — revenue ₩104.6T, operating profit ₩83.3T at 80% margin — plus >50% of 2027–28 capacity locked in long-term agreements. Micron's own balance sheet shows $12.9B of customer prepayments, up from $0.6B last quarter.
The disagreement: consensus sees +275%-plus YTD memory names as late-cycle. The primary sources say sold-out 2026, 75% sold-out 2027, and price floors — a utility order book, not a spot cycle.
The level that changes everything: $860. Below it, the LTA/pricing story is failing. Above $1,255, the re-rating to a toll multiple has begun.
The takeaway: SK Hynix did not just beat — it reclassified memory from commodity to constraint.
On October 28, SK Hynix reported Q3 2026 revenue of ₩104.6 trillion (+32% QoQ, +328% YoY) and operating profit of ₩83.3 trillion — an 80% operating margin. Net profit was ₩39.8T. Cash rose to ₩114.5T; net cash ₩98.7T. The company guided that HBM4 overtakes HBM3E in both volume and revenue in Q4, and that the demand-exceeds-supply structure persists through 2030 given leading-edge difficulty and fab lead times.
Two details matter more than the headline. First, 14 long-term agreements with leading cloud providers, covering >50% of 2027–2028 production capacity, with a price-band mechanism that reflects market conditions — stable profitability with upside, not fixed-price giveaway. Second, ≈₩100T of 2026 shareholder returns (≥₩30T dividends including a ₩10,000 special from Kioxia gains, ≥₩60T buybacks). A commodity peak does not prepay customers and buy back 60 trillion won.
Market reaction was instant: Kospi +6.6%, SK Hynix +11%, Samsung +9%, Micron +4.5% to $970.67 intraday and SanDisk +5.6% to $1,378.25 ahead of its October 29 print. Bitcoin bounced with risk to ~$81,400. Tonight's Fed hold (∼92% priced) and MSFT/GOOGL prints are noise around this signal.
The takeaway: Micron's Q4 is the American confirmation of the Korean print — and the balance sheet proves customers are financing the build.
Micron's fiscal Q4 2026 (ended Sept 3, reported Sept 30): revenue $54.23B vs $41.46B prior quarter and $11.3B a year ago; GAAP gross margin 86.8%; non-GAAP EPS $33.42; operating cash flow $44.0B; adjusted free cash flow $33.2B. Full FY2026: revenue $133.2B, GAAP EPS $74.33. Fiscal Q1 2027 guide: $61.5B ± $1.5B revenue, ~86% gross margin, $37.84 ± $1.00 GAAP EPS. That guide, annualized, is a $150+ EPS run-rate on stock at $971 — a ~6.5x forward run-rate multiple if you believe one quarter. You shouldn't annualize a peak blindly — which is exactly why the contract structure matters.
CEO Sanjay Mehrotra on the Sept 30 call: memory and storage supply is 75% sold out for 2027. Micron's entire 2026 HBM was already under price-and-volume agreements. And the tell the income statement hides: noncurrent customer contract liabilities $12.895B vs $568M the prior quarter — customers wiring deposits to reserve supply. That is vendor financing in reverse: hyperscalers funding Micron's capex because they fear shortage more than overpaying. SK's 50%-of-capacity LTAs are the same phenomenon from the supplier side.
Opinion: this is the load-bearing shift. A spot DRAM seller deserves 6–8x peak EPS. A toll with 2–3 years of contracted volume and floor pricing deserves 12–15x mid-cycle earnings. At 12.4x trailing (handler data) on depressed-then-explosive earnings, Micron is priced as if the LTAs don't exist.
The takeaway: attention is on MSFT/GOOGL and Warsh — leaving the memory read-through mispriced for 24 hours.
CNBC's Oct 28 analysis shows Mag-7 60-day pairwise correlation at a decade low — the group no longer trades as a block (MAGS +7% YTD vs S&P +12%). Tonight's read is Azure vs Google Cloud growth and capex durability — the same capex that is memory demand. Micron reports into that capex number with a one-month lag already banked, while SanDisk (Oct 29 AMC, EPS est $47.15, revenue est ~$10.85B) is the near-term validator or spoiler.
Meanwhile insider sales — Deboer ~28.5k shares ~$900 on Oct 16, Ray ~2.8k shares $930–$947 on Oct 14 — are routine tax/plan selling into strength, not a supply signal; sizes are trivial against 1.15B diluted shares. The real supply signal is the opposite: SK cancelling 40T won of buyback execution last week and promising quarterly repurchases.
| Field | Level |
|---|---|
| Entry | $930–$975 |
| Target | $1,180–$1,255 (52-week high $1,255 of June 25); stretch $1,430 on LTA repricing |
| Invalidation | Daily close below $860 (≈11% below entry; breaks Oct breakout + insider-sale cluster) |
| Timeframe | 1–3 months |
| Conviction | SPECULATIVE-HIGH (SK 80% margin + MU $12.9B prepayments agree; SNDK confirmation pending) |
| Audience / sizing | Tactical growth; half-size (≤2% book) until Oct 29; book rule: halve if META <$600 or 10Y >5.10% |
Math: FY26 non-GAAP EPS $75.52 at $971 = 12.9x. Q1 FY27 guide midpoint non-GAAP $38.15; two quarters at that run-rate cover half of FY26's full-year profit. If HBM floor pricing holds and DRAM mix stays server-heavy, a $110–$120 mid-cycle EPS at 12x = $1,320–$1,440 — hence the stretch. If pricing rolls, $60 EPS at 8x = $480 — hence the hard $860 stop, not hope.
Contradiction check: no open MU call. Complementary: SPEC-AVGO-1 ($320–335 → now $346) and MACRO-LITE-1 ride the same capex durability; SPEC-BTC-1027 ($78.5–80k entry, BTC now $80,440) needs the same Fed hold. If META breaks $600, halve this too.
The takeaway: SNDK tomorrow and hyperscaler capex this week decide which path.
| Scenario | Weight | Trigger | MU implication |
|---|---|---|---|
| Bull: shortage repriced | 30% | SNDK beats + MSFT/GOOGL raise capex; HBM4 pricing up | $1,255 breaks → $1,430 stretch |
| Base: grind with volatility | 50% | In-line SNDK, steady capex, Fed hold | $950–$1,180 chop; add dips |
| Bear: cycle fear returns | 20% | SNDK margin miss, capex cut, or DRAM spot -10% | $860 fails → $780–$820; exit |
The takeaway: three falsifiers, all measurable within two weeks.
The takeaway: buy the toll, not the lottery ticket — and let tomorrow's print set size.
Contents: A. Data snapshot · B. Model & load-bearing input · C. Sources
| Item | Level |
|---|---|
| MU | $970.67 +4.5%; 52wk $192.59–$1,255; P/E TTM 12.36; beta 2.40 |
| Micron FQ4 FY26 | Rev $54.23B; GM 86.8%; non-GAAP EPS $33.42; OCF $44.0B; customer deposits $12.895B |
| Micron FQ1 FY27 guide | Rev $61.5B ±1.5B; GM ~86%; GAAP EPS $37.84 ±1.00 |
| SK Hynix Q3 | Rev ₩104.6T; OP ₩83.3T (80%); net ₩39.8T; >50% 2027–28 LTA; ~₩100T 2026 returns |
| SNDK / WDC / STX | SNDK $1,378.25 +5.6% (earnings Oct 29 AMC); WDC $392.87 +3.1%; STX $676.80 +1.3% |
| NVDA / AMD / AVGO | NVDA $217.30 -0.3%; AMD $567.24 +1.2%; AVGO $346.33 +1.2% |
| Macro / crypto | SPY $771.35; QQQ $722.28; Fed decision 18:00 UTC (~92% hold); BTC $80,440; F&G 51 neutral |
Base: FY27 H1 run-rate EPS ~$38 × 2 = $76 H1; assume H2 moderation to $50 → FY27 ~$126? Haircut 30% for cyclicality → $88 mid-cycle × 13x toll multiple ≈ $1,144 ≈ target low-end. Bull: $110 × 13x = $1,430. Load-bearing input: LTA floor pricing holds. If wrong — DRAM spot -20% with volume flat — EPS falls to ~$55 and multiple compresses to 8x = $440. That is why $860 is the stop, not a suggestion.