Micron at 5.5x Forward P/E: Cycle Trap or Generational Buy?
美光 Forward P/E 仅 5.5 倍，是周期陷阱还是历史性买点？
TL;DR
- 1MU posted record Q2 FY2026 ($23.86B, +28% beat) then dropped 32% in 12 days — fundamentals accelerating while stock collapsing.
- 2At $321, forward P/E is 5.5x — nearly identical to the 2018 cycle peak (5.6x) which preceded a 56% decline and 88% EPS collapse.
- 3The central question is not whether 5.5x is cheap, but whether the E ($58.50 FY2026E) is sustainable. Watch TrendForce Q2 contract prices (April) and MU Q3 earnings (June 25) for verification.
Key Terms
Chapter 1
The Paradox
矛盾的现象
Record earnings, record decline
On March 18, 2026, Micron Technology reported the strongest quarter in its 46-year history. Revenue hit $23.86B — beating guidance by 28%. Gross margin reached 75%. Non-GAAP EPS came in at $12.20, crushing the $8.42 guide by 45%.
The Q3 guidance was even more explosive: $33.5B revenue, 81% gross margin, $19.15 EPS. A single quarter's EPS guidance exceeded the entire FY2025 full-year EPS of $8.29.
That same day, the stock hit an all-time high of $471.34.
Twelve trading days later, it closed at $321.78. Down 32%.
The fundamentals were accelerating. The stock was collapsing. One of them must be wrong.
Chapter 2
Reality Check: Fear vs Data
恐慌拆解：催化剂的真相
Three catalysts drove the selloff. How many are real?
NVIDIA Rubin Ultra Downgrade
NVIDIA Rubin Ultra 减配传闻
Claim
Rubin Ultra reduced from 4 GPU dies to 2, halving HBM demand per rack.
Reality
No credible source found across Tom's Hardware, VideoCardz, WCCFtech, or SemiAccurate. Published specs unchanged from CES to GTC 2026: VR200 = 2 die, VR300 (Ultra) = 4 die.
Impact on E
None unless confirmed. CY2026 HBM contracts already locked.
No source found
DRAM Spot Price Decline
DRAM 现货价下跌
Claim
DRAM spot prices falling signals the memory cycle has peaked.
Reality
DDR4 spot -0.29%/week, DDR5 retail kits -20% from peak. BUT contract prices (which drive earnings) surged +90-95% QoQ in Q1 CY2026 — the largest increase in memory history. Spot leads contract as a leading indicator, but has not yet hit the bear threshold.
Impact on E
Leading indicator. Not yet confirmed. Watch for 4+ consecutive weeks of >15% decline.
TrendForce weekly spot reports
Google TurboQuant
Google TurboQuant 量化压缩
Claim
6x KV-cache compression eliminates the need for HBM capacity growth.
Reality
Google Research claims 6x theoretical compression (32-bit to 3-bit). However, 70-80% of AI inference already uses 8-bit precision, yielding only ~2.6x practical effect. More critically, TurboQuant only affects inference — HBM demand is primarily driven by training workloads, which require full precision.
Impact on E
Minimal. Does not affect training HBM demand. May even expand total market via Jevons Paradox.
Google Research blog, Seoul Economic Daily
What the Hard Data Actually Says
| Metric | Value | Significance |
|---|---|---|
| Q2 Revenue | $23.86B (guide $18.7B, +28% beat) | Not a meet — a massive upside surprise |
| Q2 Gross Margin | 75% (guide 68%, +700bp beat) | Highest in memory industry history |
| Q3 Guidance | $33.5B / 81% GM / $19.15 EPS | Single quarter EPS > entire FY2025 |
| HBM Supply | CY2026 fully committed, price & volume locked | Not 'might sell' — already contracted |
| MU Ending Inventory | $8.3B, 123 days (DRAM <120 days) | Tight vs history, no buildup signal |
| FY2026E Reconstruction | ~$107-111B revenue, ~$57-60 EPS | Forward P/E at $321 = 5.5x |
Chapter 3
The Cyclical Valuation Paradox
周期股的估值悖论
Why 5x P/E might be the most dangerous signal
Imagine buying an ice cream stand for $500K when it makes $100K/month in summer. P/E = 5x. Sounds cheap. But in winter it makes $20K/month — your real P/E is 25x. You didn't buy 'cheap.' You bought 'summer.'
想象你在夏天花 50 万买了一个冰淇淋摊，月赚 10 万，P/E = 5x。看起来很便宜。但冬天月赚 2 万 — 你的实际 P/E 变成 25x。你买的不是“便宜”，你买的是“夏天”。
Conceptual: EPS vs P/E Through a Memory Cycle
| Growth Stock | Cyclical Stock | |
|---|---|---|
| Low P/E means | Cheap — buy opportunity | E at peak — most dangerous |
| High P/E means | Expensive — caution | E at trough — safest to buy |
| Best time to buy | When P/E is low | When P/E is highest (E at bottom) |
When E (earnings) is at a cyclical peak, the denominator in P/E is abnormally large, making the ratio look small. The market knows peak E won't last, so it refuses to pay a high multiple for temporary earnings.
Once E starts declining, the denominator shrinks — even if the stock price doesn't move, P/E expands rapidly. In practice, the stock falls faster than E because the market prices forward.
When you see a cyclical stock at 5x forward P/E, the market is not saying 'it's cheap.' It's saying 'I don't believe this E will last.'
Beyond P/E: Other Valuation Lenses
Chapter 4
Historical Evidence: Three Cycle Peaks
历史验证：三轮周期见顶实录
Actual P/E data from 2014, 2018, and 2022
Theory is theory. Let's test it against actual data. We reconstructed the P/E ratios at every Micron cycle peak using quarterly GAAP EPS from SEC filings and daily closing prices from Yahoo Finance. No estimates, no consensus — just what actually happened.
Three complete cycles. Three peaks. Three crashes. And a striking pattern.
| 2014 Peak | 2018 Peak | 2022 Peak | 2026 ATH | 2026 Now | |
|---|---|---|---|---|---|
| Peak Price | $35.70 | $63.09 | $96.29 | $471 | $321.78 |
| Peak Date | Dec 2014 | May 2018 | Jan 2022 | Mar 18 | Mar 30 |
| TTM EPS | $2.55 | $8.60 | $7.12 | ~$16.98* | ~$16.98* |
| Trailing P/E | 14.0x | 7.3x | 13.5x | 27.7x | 18.9x |
| Forward 12M EPS (actual) | $2.69 | $11.28 | $5.76 | $58.50E | $58.50E |
| Forward P/E | 13.3x | 5.6x | 16.7x | 8.1x | 5.5x |
| Subsequent EPS Change | +6% | -47% | -164% | ? | ? |
| Trough Price | $9.12 | $27.70 | $47.63 | — | — |
| Trough Date | May 2016 | Dec 2018 | Dec 2022 | — | — |
| Peak-to-Trough Duration | 17 months | 7 months | 11 months | — | — |
| Peak-to-Trough Decline | -74% | -56% | -51% | — | -32%* |
| Trailing P/E at Trough | 8.3x | 2.2x | 6.0x | — | — |
| Peak FY EPS | $2.54 | $11.51 | $7.75 | $58.50E | $58.50E |
| Next FY EPS | $2.69 | $6.10 | -$4.96 | ? | ? |
2014 Cycle: The Gentle Landing
Peak
$35.70
December 8, 2014
Trough
$9.12
May 12, 2016
Decline
-74%
Duration
17 months
The 2014 cycle was driven by PC and smartphone DRAM demand. MU stock peaked at $35.70 on December 8, 2014, with FY2014 full-year EPS of $2.54.
What made this cycle different: EPS actually continued rising slightly into FY2015 ($2.69, +6%). The stock peaked before earnings did — but the subsequent decline was the steepest of all three cycles at -74%.
At the peak, trailing P/E was 14.0x — the highest of any cycle peak. The market was still pricing in growth, not yet fully skeptical. This made the subsequent crash especially brutal for latecomers.
The trough of $9.12 came 17 months later in May 2016. At that point, TTM EPS had fallen to $1.10, and P/E at trough was 8.3x — still looking 'reasonable' on trailing numbers while earnings were at rock bottom.
| Quarter | Period End | GAAP EPS | Note |
|---|---|---|---|
| Q1 FY14 | Nov 2013 | $0.30 | Depressed by $233M Rambus settlement |
| Q2 FY14 | Feb 2014 | $0.61 | |
| Q3 FY14 | May 2014 | $0.68 | |
| Q4 FY14 | Aug 2014 | $0.96 | |
| Q1 FY15 | Nov 2014 | $0.97 | Stock peaks here, EPS still rising |
| Q2 FY15 | Feb 2015 | $0.81 | |
| Q3 FY15 | May 2015 | $0.54 | |
| Q4 FY15 | Aug 2015 | $0.37 | |
| Q1 FY16 | Nov 2015 | $0.24 | |
| Q2 FY16 | Feb 2016 | -$0.05 | First loss quarter |
2018 Cycle: The Classic Value Trap
Peak
$63.09
May 30, 2018
Trough
$27.70
December 26, 2018
Decline
-56%
Duration
7 months
The 2018 cycle is THE most instructive case for 2026. It was driven by data center expansion and cryptocurrency mining demand. MU peaked at $63.09 on May 30, 2018.
At the peak, trailing P/E was 7.3x and forward P/E was 5.6x. MU appeared to be the cheapest large-cap stock in America. Wall Street was overwhelmingly bullish — 27+ analysts rated it Buy.
Here's the critical deception: quarterly EPS was still rising when the stock peaked. Q3 FY18 ($3.12) and Q4 FY18 ($3.56) were both higher than the quarters before the stock peak. EPS peaked in August 2018 — three months AFTER the stock peaked in May.
If you bought at the peak because 'forward P/E is only 5.6x,' you were right about the next few quarters of earnings. But the stock still fell 56% because the market was pricing in what came next: EPS collapsed from $3.56/quarter to $0.42/quarter — an 88% decline in per-quarter earnings over just 5 quarters.
The cruelest part: at the absolute stock price low of $27.70 on December 26, 2018, trailing P/E was just 2.2x. It looked like the deal of the century. But forward earnings were about to fall off a cliff — FY2019 EPS came in at $6.10 vs FY2018's $11.51, a 47% decline.
| Quarter | Period End | GAAP EPS | Note |
|---|---|---|---|
| Q4 FY17 | Aug 2017 | $1.99 | |
| Q1 FY18 | Nov 2017 | $2.41 | |
| Q2 FY18 | Feb 2018 | $2.80 | |
| Q3 FY18 | May 2018 | $3.12 | Stock peaks May 30 |
| Q4 FY18 | Aug 2018 | $3.56 | EPS peaks HERE, 3 months after stock |
| Q1 FY19 | Nov 2018 | $2.93 | Decline begins |
| Q2 FY19 | Feb 2019 | $1.67 | |
| Q3 FY19 | May 2019 | $1.00 | |
| Q4 FY19 | Aug 2019 | $0.50 | |
| Q1 FY20 | Nov 2019 | $0.42 | EPS trough: -88% from peak |
2022 Cycle: The Earnings Wipeout
Peak
$96.29
January 5, 2022
Trough
$47.63
December 22, 2022
Decline
-51%
Duration
11 months
The 2022 cycle was driven by pandemic-era demand for PCs, gaming, and data center buildout. MU peaked at $96.29 on January 5, 2022.
Unlike 2018, this peak had a higher trailing P/E of 13.5x — the market was more cautious but still constructive. Forward P/E was 16.7x (using actual FY2022 forward earnings of $5.76), which appeared 'expensive' relative to 2018.
But the earnings collapse was far more severe than 2018. FY2022 full-year EPS was $7.75. FY2023 came in at -$4.96 — a NEGATIVE number. This was a -164% change, the most extreme of any cycle. Micron went from generating $7.75/share in profit to losing $4.96/share.
The quarterly EPS trajectory was devastating: $2.50 (Q3 FY22 peak) to $1.36, then -$0.15, -$2.03, -$1.57, -$1.21. Four consecutive quarters of losses.
At the trough of $47.63 on December 22, 2022, trailing P/E was 6.0x based on peak-year FY2022 earnings. Forward P/E was meaningless because forward earnings were negative. This is the ultimate expression of the cyclical paradox: the stock 'looked cheap' on trailing numbers, but forward earnings had literally evaporated.
| Quarter | Period End | GAAP EPS | Note |
|---|---|---|---|
| Q4 FY21 | Sep 2021 | $2.37 | |
| Q1 FY22 | Dec 2021 | $2.04 | Stock peaks Jan 5, 2022 |
| Q2 FY22 | Mar 2022 | $2.05 | |
| Q3 FY22 | Jun 2022 | $2.50 | EPS peaks, 6 months after stock |
| Q4 FY22 | Sep 2022 | $1.36 | Decline begins |
| Q1 FY23 | Dec 2022 | -$0.15 | First loss — stock near trough |
| Q2 FY23 | Mar 2023 | -$2.03 | Deepest loss quarter |
| Q3 FY23 | Jun 2023 | -$1.57 | |
| Q4 FY23 | Aug 2023 | -$1.21 |
The Pattern
- 1In every cycle, the stock peaked BEFORE EPS peaked. The lag ranged from 0 to 6 months. If you wait for an earnings miss to sell, you're already late.
- 2In every cycle, the stock looked 'cheap' on trailing P/E at both the peak and the trough. Low P/E was never a reliable buy signal for cyclical Micron.
- 3The severity of E collapse has increased with each cycle (+6% to -47% to -164%), driven by rising capital intensity. MU's FY2026 capex of $25B+ is the highest ever — the operating leverage cuts both ways.
- 4The fastest crash was 2018 (7 months peak to trough). The deepest was 2014 (-74%). The most painful for EPS was 2022 (full-year loss).
Structural Comparison Across All Four Cycles
| Dimension | 2014 Peak | 2018 Peak | 2022 Peak | 2026 Now |
|---|---|---|---|---|
| Peak Date | Dec 8, 2014 | May 30, 2018 | Jan 5, 2022 | Mar 18, 2026 |
| Demand Driver | PC/smartphone refresh | Data center + crypto mining | Pandemic WFH + DC buildout | AI training + HBM structural |
| DRAM Price at Peak | Flat to declining | +20-30% QoQ | +10-15% QoQ | +90-95% QoQ (record) |
| Inventory at Peak | Rising (4-6 weeks) | Elevated (6-8 weeks) | Moderate (4-5 weeks) | Tight (MU 123 days, DRAM <120) |
| Customer Lock-in | Spot/short-term contracts | Standard procurement | Standard procurement | HBM annual price/volume locked |
| Data Center Mix | <15% of revenue | <30% of revenue | ~35% of revenue | >57% of revenue |
| Peak FY EPS | $2.54 (FY14) | $11.51 (FY18) | $7.75 (FY22) | $58.50E (FY26E) |
| Capex / Revenue | ~30% ($4.4B) | ~22% ($8.2B) | ~34% ($12.1B) | ~23% ($25B+) |
| Forward P/E at Peak | 13.3x | 5.6x | 16.7x | 8.1x (ATH) / 5.5x (now) |
| Trailing P/E at Trough | 8.3x | 2.2x | 6.0x | ? |
| Peak-to-Trough Decline | -74% (17 months) | -56% (7 months) | -51% (11 months) | -32% so far (12 days) |
| Next FY EPS Change | +6% | -47% | -164% (loss) | ? |
EPS Trajectories (Peak = 100)
Each cycle's peak-quarter EPS is set to 100 so you can compare the shape of the rise and collapse, not the absolute dollar amount. Below 0 means losses. Notice how steeply the 2018 and 2022 cycles declined after their peaks.
Chapter 5
Is This Time Different?
这次不一样？
The bull case, the bear case, and what would falsify each
Bull Case
- +HBM is a new product category with structural demand, not a commodity DRAM cycle repeat
- +MU ending inventory tight at 123 days; no channel buildup
- +CY2026 HBM fully sold out with price and volume agreements locked
- +Management states shortage extends 'beyond 2026'
- +Data center mix >57% vs <30% in 2018 — structurally higher margin floor
Bear Case
- –'This time is different' is the most expensive phrase in investing history
- –Each cycle's EPS collapse has been larger: +6% (2014) → -47% (2018) → -164% (2022)
- –Capex at $25B+ (all-time high) — operating leverage amplifies E decline in downturn
- –Samsung HBM4 shipping + 50% capacity surge — three-vendor competition will compress ASPs
- –DRAM spot prices already softening — the leading indicator is turning
What Would Falsify Each View?
What Would Kill the Bull Case
- ×DRAM contract prices decline QoQ in Q2 CY2026
- ×MU Q3 GM comes in below 75% (vs 81% guide)
- ×Two or more hyperscalers cut AI capex guidance >15%
What Would Kill the Bear Case
- ✓DRAM contract prices sustain +10% QoQ in Q2 CY2026
- ✓MU Q4 FY2026 guidance maintains or exceeds Q3 run rate
- ✓HBM mix lifts GM floor to 60%+ even as commodity DRAM normalizes
Chapter 6
The Verification Framework
行动框架
What to track, what to ignore, and when to act
TrendForce Q2 CY2026 DRAM Contract Price Forecast
TrendForce Q2 CY2026 DRAM 合约价预测
Source: TrendForce
Bull Pass
QoQ growth > 0%
Bear Fail
QoQ flat or negative
Samsung + SK Hynix Q1 CY2026 Earnings
三星 + SK 海力士 Q1 CY2026 财报
Source: Earnings calls
Bull Pass
No HBM pricing aggression from Samsung
Bear Fail
Samsung undercuts HBM pricing or signals capacity dump
Hyperscaler Q1 Earnings (MSFT/GOOG/META/AMZN)
超大规模客户 Q1 财报（MSFT/GOOG/META/AMZN）
Source: Earnings calls
Bull Pass
AI capex guidance maintained or raised
Bear Fail
2+ hyperscalers cut AI capex >15%
MU Q3 FY2026 Earnings — The Ultimate Test
MU Q3 FY2026 财报 — 终极验证
Source: Micron IR
Bull Pass
81% GM achieved + Q4 guidance maintains run rate
Bear Fail
GM < 75% or Q4 guidance declines
Noise Filter — Ignore These
| Signal | Why Ignore |
|---|---|
| Rubin Ultra downgrade rumor | No traceable source; HBM contracts locked |
| TurboQuant 6x compression | Practical ~2.6x; inference only, not training |
| DDR4 spot weakness | Legacy product; decoupled from HBM/DDR5 pricing |
| Single-day -10% drops | Leverage + options gamma effect in a drawdown |
What Should You Do Right Now?
- 1DO NOT catch a falling knife. MU has dropped 32% in 12 days — the decline may not be over. Historical cycles show 51-74% peak-to-trough declines.
- 2Wait for the first hard data: TrendForce Q2 CY2026 DRAM contract price forecast (early April). If contract prices are still growing, the E-collapse thesis weakens significantly.
- 3Watch Samsung and SK Hynix earnings (~April 24-25). If Samsung does NOT start a price war on HBM, the structural demand thesis holds.
- 4The ultimate test is MU Q3 FY2026 earnings on June 25. If 81% GM materializes and Q4 guidance holds, the 2018 repeat thesis is falsified.
- 5If all four gates pass: $321 may indeed be a generational entry point with +40% upside to base case $450. If any gate fails: stay away and wait for the E revision cycle to complete.