MU $1,133.99 (+8.7% on the day, at a fresh all-time high, +766% 1Y); record Q2 FY26 74.4% gross margin and 33.5% TTM ROIC confirm super-cycle pricing power, but ~34x EV/EBITDA and ~22x P/S on cycle-peak earnings leave no margin of safety.
Researched today
Earnings Jun 24· After close· in 5 days
MU $1,133.99 (+8.7% on the day, at a fresh all-time high, +766% 1Y); record Q2 FY26 74.4% gross margin and 33.5% TTM ROIC confirm super-cycle pricing power, but ~34x EV/EBITDA and ~22x P/S on cycle-peak earnings leave no margin of safety.
Price is above the trim-above target — rich versus fair value.
Recommendation
Conviction
47/100
watch
Upside
23/100
bull 30% · ~30% odds · +10% expected
Risk-adjusted upside
0/100
+0% after downside pressure
Thesis quality
9.6/10
Opportunity
6.0/10
Risk pressure
8.1/10
Valuation
ExpensiveAI fair value
$340.00
Fundamentals check
$963.89
12-24mo fair-value range
$200.00 / $340.00 / $700.00
width 147%
Estimates disagree — look closer.
Buy below
—
Trim above
$840.00
Implied expectations
achievable47.6% implied revenue CAGR
Today's P/S implies ~47.6% revenue CAGR for five years versus 47.9% realized growth.
Agree the HBM 3:1 wafer consumption, 2027-first fab supply, and oligopoly discipline make this shortage tighter and longer than a normal cycle. Push back that this de-risks the equity: the Substack thesis is about supply tightness, not valuation, and ~34x EV/EBITDA on a 74.4% peak gross margin already prices a multi-year super-cycle. Structural-but-finite tightness still ends in 2028-2029 by the author's own timeline.
Agree the Vera Rubin SOCAMM cut (192GB to 96GB) is not automatically a demand collapse, since bit demand is module capacity x modules x systems and lower content can be offset by more systems. But the risk is real and asymmetric at this price — it is exactly the kind of per-system content cut that punctures a peak-margin narrative, supporting caution rather than initiation.
MicroLED optical interconnect is genuine long-dated optionality (post-HBM5) that all three memory makers are watching, but the author frames it as insurance, not a near-term product — it does not change the FY26-27 valuation or cyclicality picture.
Edge-AI LPDDR6/24GB demand is a plausible incremental bit-demand driver, but the post carries no structured evidence in the packet; treat as supportive color, not a verdict driver.
Micron is the most direct US-listed HBM/DRAM bottleneck play. Q2 FY26 records (74.4% GM, 33.5% TTM ROIC, 191% YoY per-share revenue) confirm AI-capex-driven pricing power, and TIER_A DD argues HBM's 3:1 wafer consumption makes the shortage structural into 2028-2029. Defensibility is capped by SK Hynix's HBM lead, Samsung re-qualification risk, and the FY2023 -9.1% gross margin from the same company evidencing an ~83-point peak-to-trough swing. Structural cyclicality, not thesis fit, gates the verdict.
Moat
Three-player DRAM oligopoly with high barriers but commoditized product and SK Hynix HBM lead — narrow, not wide.
Bottleneck fit
Most direct US-listed HBM/DRAM bottleneck play; 191% YoY per-share revenue confirms AI-capex exposure.
Valuation
~34x EV/EBITDA and ~22x P/S on cycle-peak earnings at an all-time high; no margin of safety.
Catalyst
FQ3 FY26 earnings late June with HBM4 share and ASP readout is a specific, dated catalyst.
Why not higher
Cycle-peak valuation at an all-time high with ~70% downside to normalized value and unresolved HBM4 share risk prevent moving above WATCH.
Description
Micron is one of three global DRAM/NAND memory makers, now a primary US-listed supplier of HBM (high-bandwidth memory) into AI accelerator systems.
Value Chain
Component supplier (memory) to AI accelerator and server platforms; upstream of NVDA/AMD compute systems.
Moat
narrowDRAM is a three-player oligopoly (Micron, Samsung, SK Hynix) with high capital and process barriers, but the product is largely commoditized and Micron trails SK Hynix in HBM leadership.
Pricing Power
highRecord 74.4% Q2 FY26 gross margin and TrendForce-reported 90-95% QoQ DRAM contract price jumps reflect acute shortage-driven pricing power — but this is cycle-peak, not structural, pricing.
Customer Concentration
AI/hyperscaler demand flows largely through NVIDIA's accelerator platform; specific top-customer % not disclosed in packet.
| Metric | Value |
|---|---|
| Revenue growth YoY | 191.40% |
| Gross margin | 74.41% ↑ |
| Operating margin | 67.62% |
| FCF margin | 23.12% |
| Cash position | 14585000000 |
| Net debt / EBITDA | -0.10 |
| Share count change YoY | 1% |
| ROIC | 33.46% |
Forward P/E
10.75
Trailing P/E
53.54
PEG
0.07
EV/EBITDA
34.29
P/S
21.73
| Peer | Metric | Value |
|---|---|---|
| NVDA | relationship | Primary HBM demand driver, not a clean multiple comp; MU is demand-linked to NVDA accelerator volumes. |
| SK Hynix | HBM leadership | Holds HBM share/qualification lead at NVIDIA; competitor for HBM4 sockets. |
| self | P/B | P/B 6.45x (Feb-26 quarter) vs MU's own 1.3-2.5x historical range — far above normal. |
Bull fair value
$1450.00
Probability
30%
Horizon
Medium term
YTD
—
1Y
766.87%
Vs sector
—
From 52w high
0%
| Risk | Severity | Explanation |
|---|---|---|
| Cycle-peak margins mean-reverting | high | 74.4% Q2 FY26 gross margin vs -9.1% in FY2023; memory margins historically swing violently and 2027-2028 capacity is the documented easing path. |
| Valuation compression on peak earnings | high | ~34x EV/EBITDA and ~22x P/S on cycle-peak earnings at an all-time-high price leave no margin of safety; forward P/E of 10.75x is misleadingly cheap if peak EPS reverts. |
| HBM4 share loss at NVIDIA | high | TIER_A X commentary cites a market narrative that MU holds zero share of NVIDIA HBM4; SK Hynix lead and Samsung re-qualification could displace Micron in the key socket. |
| AI capex / demand bend | medium | NVIDIA's reported Vera Rubin SOCAMM cut from 192GB to 96GB modules halves CPU-side DRAM per rack; a slowdown in system count would weaken memory bit demand. |
| Insider selling at peak | medium | CEO Mehrotra sold at ~$972-979 on 2026-05-29 with no offsetting buys, consistent with multi-month net selling. |
0%–0%
WATCH — 0% weighting. The structural-shortage DD is credible, but at an all-time high with ~34x EV/EBITDA and ~22x P/S on cycle-peak earnings, ~70% implied downside to normalized value, fresh CEO selling, and HBM4 share uncertainty, the entry price offers no margin of safety. Re-evaluate on a meaningful pullback or disclosure of locked multi-year HBM pricing/share giving FY2027-2028 visibility.
This is a textbook memory-cycle top dressed as a structural story. Every headline metric is a peak: 74.4% Q2 GM vs -9.1% in FY2023 (an ~83-point peak-to-trough swing in three years), 33.5% TTM ROIC, 191% YoY per-share revenue — all printed at a fresh all-time high after +766% in one year. The market caps EV/EBITDA ~34x and P/S ~22x on these peak margins, implying durability that the same company's own history flatly contradicts. The 'cheap' 10.75x forward P/E is a trap: it embeds a ~$105 cycle-peak EPS as the base. DDR4 priced above DDR5 (cited as bullish) is a classic late-cycle scarcity inversion, and the bull's own timeline puts supply easing at 2H2027. A bear presses simple mean reversion: 2027-2028 fab capacity arrives, ASPs roll, and the earnings base the multiple rests on halves.
| Load-bearing assumption | Why it might be wrong | Severity |
|---|---|---|
| Structural-shortage thesis extends tight DRAM/HBM pricing into 2028-2029, supporting the lifted ~$340 fair value. | This rests entirely on a single Substack author (damnang) and one X handle (aleabitoreddit) — no primary disclosure, no company guidance, no 10-Q figure in the packet confirms multi-year locked pricing. Even that author bounds easing to 2028-2029 and lists 'AI investment bends' as the risk. A supply-tightness narrative is not a valuation de-risk. | high |
| Micron holds/will win meaningful HBM4 share at NVIDIA Vera Rubin. | The only direct HBM4 datapoint in the packet is a TIER_A market narrative that MU holds ZERO share of NVIDIA HBM4. The bull rebuttal is one tweet asserting the selloff was 'dumb' — not evidence of qualification. The report under-weights a thesis-level demand risk while crediting unverified rebuttal. | high |
| Forward EPS (~$105 implied) is a usable earnings base, making the multiple defensible. | Consensus forward EPS is itself anchored to the peak; PEG 0.07 and forward P/E 10.75 are consensus artifacts, not independent fair value. TTM EV/FCF is 122x and annual EV/FCF 757x — cash generation does not remotely support the price, so the 'cheap on earnings' frame ignores that owner earnings are thin. | medium |
| Vera Rubin SOCAMM cut (192GB to 96GB) is offset by higher system count. | This is asymmetric at peak valuation: a confirmed ~50% per-system DRAM content cut needs system unit growth to fully compensate, which is unproven. The bit-demand identity cuts both ways and the report's own analyst_take rates this 'real and asymmetric.' | medium |
The $340 base (normalized EPS $25-30 at ~12x) is methodologically the most defensible anchor in the report and points to ~70% downside — it is conservative, not aggressive. The weakness is on the upside: the $700 high and $1,450 bull-case embed structural-shortage durability that even the cited author caps at 2028-2029, and they lean on the same unverified single-source DD. Given EV/FCF >100x and FCF margins historically swinging 4-14%, a normalized FCF-based cross-check would likely sit below $340, not above. The peer/forward multiples (PEG 0.07, fwd P/E 10.75) are consensus prices embedding peak EPS and should not be read as independent valuation support.
Independent red-team pass · claude-opus-4-8 · 2026-06-19
Price assumes peak DRAM/HBM margins persist multi-year; I expect mean reversion from the record 74.4% gross margin toward mid-cycle as 2027-2028 fab capacity arrives, compressing the earnings base the multiple rests on.
| Claim | Source | URL | Retrieved |
|---|---|---|---|
| Price $1,133.99, +8.70% on the day, market cap $1.28T | finnhub:quote | — | retrieved 2026-06-19T01:45:11.168Z |
| 52-week high $1,110.40 (2026-06-16), 52-week low $103.38, 1Y return +766.87% | finnhub:basic-financials | — | retrieved 2026-06-19T01:45:10.275Z |
| Q2 FY26 (2026-02-26) gross margin 74.41%, operating margin 67.62%, net margin 57.77%, FCF margin 23.12% | finnhub:basic-financials | — | retrieved 2026-06-19T01:45:10.275Z |
| TTM ROIC 33.46%, ROE 40.84% (2026-02-26) | finnhub:basic-financials | — | retrieved 2026-06-19T01:45:10.275Z |
| Forward P/E 10.75, forward PEG 0.067, EV/EBITDA TTM 34.29, EV/Revenue TTM 21.73, epsTTM 21.18 | finnhub:basic-financials | — | retrieved 2026-06-19T01:45:10.275Z |
| FY2023 gross margin -9.1% vs FY2025 39.8% — historical cyclicality | finnhub:basic-financials | — | retrieved 2026-06-19T01:45:10.275Z |
| Net cash position (netDebtToTotalEquity -0.052, 2026-02-26); cash/share $12.93 | finnhub:basic-financials | — | retrieved 2026-06-19T01:45:10.275Z |
| Analyst ratings Jun 2026: 18 strong buy, 33 buy, 3 hold, 1 sell | finnhub:recommendations | — | as of 2026-06-01 · retrieved 2026-06-19T01:45:10.263Z |
| CEO Mehrotra sold shares at ~$972-979 on 2026-05-29 (filed 2026-06-02) | finnhub:insider-transactions | — | retrieved 2026-06-19T01:45:10.305Z |
| 10-Q filed 2026-03-19; recent 8-Ks 2026-06-09, 2026-04-01 | sec:recent-filings | Link | as of 2026-06-09 · retrieved 2026-06-19T01:45:11.076Z |
| Macro backdrop: 10Y 4.49%, Fed funds 3.63% (2026-06-17) | fred:macro-snapshot | — | as of 2026-06-17 · retrieved 2026-06-19T01:45:10.877Z |
| DRAM shortage argued structural: HBM 3:1 wafer consumption vs DDR5; Q1 contract prices +90-95% QoQ (TrendForce); easing 2H2027 earliest | substack:damnang | Link | retrieved 2026-06-19T01:45:10.090Z |
| NVIDIA reportedly halving Vera Rubin SOCAMM modules (192GB to 96GB), cutting CPU-side DRAM per rack ~54TB to ~28TB | substack:damnang | Link | retrieved 2026-06-19T01:45:10.090Z |
| Market narrative claims MU holds zero share of NVIDIA HBM4 (disputed by author) | x:@aleabitoreddit | Link | retrieved 2026-06-19T01:45:10.089Z |
| NVIDIA CEO warned memory shortage expected to persist for years on AI scaling demand | x:@aleabitoreddit | Link | retrieved 2026-06-19T01:45:10.089Z |
The author argues today's DRAM shortage is structural, not cyclical: HBM eats far more wafer capacity than standard DRAM, new fabs don't arrive until 2027 and get allocated to HBM first, and the three makers are deliberately withholding aggressive expansion. This locks in tight supply and pricing power for memory makers like Micron through 2028-2029.
NVIDIA is reportedly lowering the SOCAMM standard build from 192GB to 96GB modules, cutting CPU-side DRAM per rack from ~54TB to ~28TB, which dragged Micron down. The author argues this is not necessarily a demand collapse: SOCAMM bit demand is the product of module capacity, modules per system, and systems shipped, so a lower per-system content can be offset if loosened supply lets more systems install. HBM4 capacity holds while bandwidth nearly triples, and Micron is preparing the HBM4 hot lane and SOCAMM2 for Vera Rubin.
Micron, alongside Samsung and SK hynix, has invested in and is seriously evaluating MicroLED optical interconnect (Avicena's LightBundle) because its 'wide and slow' parallelization philosophy mirrors HBM's, and optics could free memory from package geometry constraints. The investment is framed as insurance on what comes after HBM rather than a near-term product plan.