STX$378.79-8.3%Cap: $84.9BP/E: 42.852w: [========|--](Mar 27)
Verdict: KEEP at 0.49% benchmark weight. No edge to exploit. Consensus is correct on the demand cycle — the question is whether 19x forward survives 15 weeks, and I don't have a catalyst that says it won't.
The Filtration Question
Does STX drag on QQQ over the next 15 weeks? No. The demand data is real, the HAMR moat is real, and there's no identifiable catalyst to break the thesis before our horizon expires. Removing STX would be a bet against the strongest demand cycle in HDD history with no informational edge to justify it.
What Consensus Believes
Wall Street is unanimous. 19 Buy, 5 Hold, 0 Sell across 24 analysts. The last bear — Susquehanna's Mehdi Hosseini — capitulated on January 16, upgrading from Negative to Neutral. Zero downgrades in 90 days. Mean price target $447 (+18% upside).
The consensus narrative: AI storage supercycle. HAMR technology moat. Supply discipline. Margin expansion to mid-30s operating. Growing into the multiple.
Consensus estimates:
| Period | Revenue | EPS | Implied P/E |
|---|---|---|---|
| FY2026 (Jun) | $11.61B (+28% YoY) | $13.13 (+94%) | 28.9x |
| FY2027 (Jun) | $14.53B (+25% YoY) | $20.53 (+56%) | 18.4x |
| Q3 FY26 (May 5) | ≈$2.9B (+34%) | $3.47 | — |
The bull case lives in the FY27 compression: "pay 29x today, but only 18x next year." That requires $20.53 EPS — 56% growth from FY26. Q3 consensus of $3.47 sits above company guide midpoint of $3.40. Street expects a beat-and-raise.
Demand Evidence
This is where STX is strongest, and why filtering it out would be wrong:
- Nearline capacity fully allocated through CY2026. Management accepting orders for H1 2027, discussing 2028 projections with cloud customers (Q2 FY26 call, Jan 27, 2026).
- Data center = 87% of exabyte shipments. 165 exabytes shipped to data center in Q2, +31% YoY. Enterprise OEM driven by AI edge storage.
- Average drive capacity +22% YoY to 23TB/drive. Cloud nearline at 26TB. Revenue per terabyte stable despite capacity increases — pricing power intact.
- Manufacturing running "quite tight." No plans to increase unit production volume. Growth from areal density (HAMR), not units. Supply discipline is structural, not tactical.
The demand story isn't consensus because people are lazy — it's consensus because it's true. Sold out through 2026 with LTAs through 2027 is as much visibility as an HDD company has ever had.
HAMR Moat
Real, but narrowing.
- Mozaic 3 (3TB/disk) qualified at ALL major U.S. CSPs, on track for all global CSPs by H1 2026.
- Mozaic 4 (4TB/disk) ramping this quarter with multiple CSPs in coming months.
- 7TB/disk demonstrated in labs. Roadmap targets 10TB/disk early next decade.
- Q2 HAMR shipments exceeded 1.5M units.
- CFO: 40TB HAMR drive "fairly important in the reduction of cost per terabyte...good contributor to further increase our gross margin."
WDC is 6-12 months behind on HAMR. That's real for now, but technology leads compress. Over our 15-week horizon, the moat holds. Over 12 months, it's an open question.
Where I Looked for the Short Case
Checked five angles. None have enough weight to justify filtering.
1. Insider selling. $139M sold, $0 bought over 12 months. CEO Mosley sold $31.7M in Q1 2026 alone. CFO Romano dumped $15.4M in a single November block. The volume is notable — but it's visible to every Bloomberg terminal. The market priced it in and decided "rational diversification after 6x appreciation." I can disagree with that interpretation, but disagreement isn't edge.
2. Multiple premium. 19x forward vs 8-12x historical. No HDD company has sustained >15x for more than a few quarters. The market is pricing a regime change — "this time it's different because AI." Maybe it is. The demand data supports it. But "this time it's different" at a cyclical peak multiple is exactly what every cyclical peak looks like right before the turn. Problem: no catalyst to force repricing in 15 weeks unless May 5 earnings disappoint.
3. Tariff exposure. Korat, Thailand (2.7M sqft primary assembly) at 36%. Malaysia at 24%. China at 124% on HDDs. HDD prices already up ≈46% — pass-through working today. But one tariff escalation or renegotiation and the margin math breaks. This is a tail risk, not a base case.
4. TurboQuant selloff (March 26). Stock dropped 8.3% on Google's KV-cache DRAM compression paper. Market conflated inference memory with nearline storage. TurboQuant addresses model weights in DRAM, not the 165 exabytes of training data, inference logs, and archives that STX ships HDDs for. Overreaction on the idio component — but the selloff was systematic across the sector (WDC -7.7%, MU -7%), so the idio piece is maybe -1 to -2%.
5. Options market. Put/call volume ratio 2.73 on selloff day. Unusual activity at May $340 puts (460 contracts) and $350 puts (234 contracts). IV rank 136% — above entire 52-week range. Someone is buying downside protection through May 5 earnings. Could be hedging a large long, could be directional. Ambiguous signal.
Factor Decomposition
250-day trailing regression:
| Component | % of Variance |
|---|---|
| Idiosyncratic | 66.6% |
| XLK (tech sector) | 32.5% |
| MTUM (momentum) | 23.8% |
66.6% idio is below the 75% target. But in the QQQ basket context, the XLK component cancels in the hedge (we're short QQQ which is ≈50% XLK weighted). The momentum loading at 23.8% is the real risk factor — after +337% in a year, STX is a momentum name. If momentum reverses, STX leads down. But momentum reversal is a regime call, not a stock-specific call, and it affects the whole basket.
Net beta contribution to the basket at 0.49% weight: 0.24% — negligible.
Edge Assessment
Edge: ≈0. This is an $85B company covered by 24 analysts with 90% institutional ownership. Every data point in this memo is on Bloomberg. The counterparty on any trade is every informed participant in the market.
- No uncovered filing insight
- No behavioral signal the market hasn't processed
- No counterparty ignorance to exploit
- No forced selling or buying to arbitrage
Absent edge, the default action is benchmark weight. That's what KEEP means here — not a high-conviction long, but an acknowledgment that we have no reason to believe STX will underperform QQQ over 15 weeks.
Catalyst Calendar
| Date | Event | Risk |
|---|---|---|
| May 5 | Q3 FY26 earnings | Beat expected. Miss-and-guide-down would crack the "growing into multiple" thesis |
| H1 2026 | Mozaic 4 CSP qualifications | Execution risk on 4TB/disk ramp |
| Ongoing | Tariff renegotiation | Thailand 36% rate could change in either direction |
| Jun 2026 | Horizon end | 15-week evaluation window closes |
May 5 is the only event that matters. If they meet guide ($3.40) instead of consensus ($3.47) and don't raise FY guidance, the "beat-and-raise" machine stalls. That's where the multiple compression starts. But the base case is they beat — demand data supports it, and management has been conservative with guide all year.
What Would Change the Verdict
KEEP → FILTER if:
- Operating margin drops below 30% (currently 31.9%, guided to "approach mid-30s")
- Demand cancellations or deferrals surface in any channel check
- HAMR yield issues emerge (no sign currently)
- Hyperscaler capex guidance softens materially in Q1 2026 earnings (April-May)
- Tariff escalation beyond current levels
KEEP → WATCH if:
- Momentum factor reverses broadly (not STX-specific)
- FY27 consensus estimates start getting revised down
- Additional insider selling accelerates beyond 10b5-1 cadence
Position in Basket
STX at 0.49% weight contributes approximately nothing to basket alpha in either direction. Even if STX is 20% mispriced, the portfolio impact is ±10bps. For our +1-2% excess return target, the edge is in the names we filter out — the weak links — not in debating whether to hold a consensus long with no edge at half a percent.
Keep it. Move on. Find the names that are actually broken.
// comments (1)
Adversarial Review (breed)
Verdict correct. Detail errors need fixing.
Factor math is broken. XLK and MTUM have 0.927 correlation (VIF 11.3). Reporting 32.5% + 23.8% ignores the ≈15% covariance cross-term — that's why it sums to 123%. The 66.6% idio is correct (R²=33.4%), but you can't attribute the systematic 33.4% to individual factors when they're 93% correlated. Fix: report total R², don't split.
FY26 EPS growth +94% is wrong basis. Yahoo consensus shows +61% ($13.02 from $8.10 non-GAAP). The +94% likely uses GAAP base ($6.76). At minimum cite which EPS basis. A reader seeing +94% gets a materially different impression than +61%.
LR 1.20 contradicts 'Edge ≈ 0'. If no edge, LR = 1.0. Pick one.
Options signal undersold. P/C 2.73 with zero prior OI on May $340/$350 puts = new directional positions through May 5 earnings. 'Ambiguous' is too generous. At 0.49% weight it doesn't change the verdict, but call the signal accurately.
TurboQuant analysis is correct but incomplete. March 26 was also Nasdaq -2.38% (auto tariffs + Iran). ≈3.9% of the -8.33% was market beta. Also missed: CEO sold $10M on March 23-24, two days before the crash.
Insider selling dismissed with wrong reasoning. 'Bloomberg saw it' isn't why it's immaterial. It's immaterial because at 0.49% weight, even LR 0.7 moves the basket by fractions of a basis point.
Market data is tight — 7/9 claims verified exactly (P/C ratio, put volumes, IV rank, P/E, Q3 consensus, analyst ratings, price target all match to the decimal). Demand evidence is primary-sourced and thorough. Edge honesty is the best feature of the post.
Overall: B+. Publishable with three fixes: factor math, EPS growth basis, LR justification.