Setup

The storage cohort confirmed the structural thesis at the same time it confirmed the entry math is broken. STX trades at $786, RSI 91, 75x trailing P/E after +734% in twelve months. WDC — identical HDD-pure-play exposure — trades at $483, RSI 90, 29x P/E after +995%. The 2.5x P/E ratio between two same-business operators is the actual edge in the cohort, and it's larger than the directional bet on whether memory-allocation tightness sustains.

What the filings say

Hyperscaler primary sources from late April converge on memory and storage as the binding constraint. Microsoft's 10-Q (April 29) showed server component receivables doubling from $8.2B (June 2025) to $17.8B (March 2026); Hood attributed ≈$25B of the ≈$190B CY26 capex guide to higher component pricing. Meta added $107B in non-cancelable contractual commitments in a single quarter, plus $24B more after quarter-end. Amazon's AWS PP&E grew $190B → $223B sequentially. Jassy on the call: "memory has skyrocketed... not enough capacity for the amount of demand... suppliers are prioritizing their very largest customers." Aggregate 2026 hyperscaler capex now $695-725B vs $640B January implicit; Google's CFO said 2027 will "significantly increase compared to 2026."

Treat this as one signal, not five — they're downstream of the same reality.

What the market thinks

The cohort has owned this trade for twelve months. STX +734%, WDC +995%, MU +708%, all at RSI 82-92. SNDK +3932% is post-spin pricing artifact. STX at 75x trailing P/E vs WDC at 29x is the anomaly inside HDD pure-plays where neither business has structurally diverged. Idio variance: STX 56%, WDC 55%, MU 52%, SNDK 85.7% — that last number is direct evidence the cohort is not uniformly over-covered. SNDK is still in a thinly-watched, news-driven regime; STX/WDC/MU are running as 2x sector-beta to SOXX.

Forward P/E softens the bear math. If FY27 EPS doubles consistent with the thesis, STX forward P/E is ≈35-40x and WDC is ≈14-15x. That's not "cycle peak" — it's "earnings step-up trading at premium." The bear case rests on the step-up unwinding, not on trailing multiples reverting in isolation.

Why the spread

Three candidate explanations for STX/WDC P/E divergence, none cleanly dominant:

1. Earnings-cycle-stage — STX margin nearer peak, WDC's still expanding into the multiple
2. Post-spin transition discount on WDC (SanDisk spin Feb 2025, integration risk priced)
3. STX's HAMR/Mosaic cloud-share leadership earns a structural premium

If (1) or (2), the spread closes. If (3), the spread is permanent and the pair won't pay.

The trade structure

The WDC/STX spread at β-neutral 1:1 (STX β 2.01, WDC β 2.16). Captures within-HDD multiple compression without a directional call on cohort top. Spread reversion to a 50% gap from current 60% is +10-15% on the pair. The pair pays in (1) and (2); only loses if (3) is right. Doesn't require timing parabolic mean-reversion — extracts the within-cohort anomaly, which is closer to alpha than beta.

The directional thesis on the basket has negative expected value at current RSI / multiples. EV math is directionally negative; magnitude is loose without options-implied calibration. The pair structure sidesteps the magnitude question.

Risks (ranked)

1. STX-specific premium is structural (HAMR leadership, cloud share) — pair never converges
2. Cohort mean-reverts harder on WDC than STX — spread moves against the thesis
3. Korean memory disruption (Hormuz/helium constraint on Samsung/SK Hynix HBM) — bullish MU specifically (US-domestic HBM); orthogonal to the HDD pair but reframes MU from avoid to hedge candidate
4. CXMT HBM3 sampling pulled forward — bearish MU; neutral on the HDD pair
5. Q3-Q4 capex air pocket — affects cohort roughly evenly, pair-neutral

The latent factors on both sides are asymmetric, not parallel: China supply (factor-5) is bearish MU, but Korean disruption (factor-7) is bullish MU. The HDD pair is robust to both.

Catalysts

• Late May 2026: NVDA Q1 FY27 print
• Late June: MU FQ3 (allocation-premium sustainability test)
• Late July: STX, WDC, SNDK FQ4 prints — gross margin expansion test (pred-wwsdjn, 72%)
• August: Q2 hyperscaler prints — does memory-allocation language repeat in a second quarter?
• Q4 2026: China HBM3 sampling milestones; Hormuz/helium status

What would change our mind

• WDC FQ4 in-line or below while STX beats-and-raises with GM acceleration — explanation (3) wins, pair structurally wrong
• STX FQ4 GM acceleration beyond consensus expansion AND FY27 guide raised — bull-extension triggered; pair still works but the spread widens before it closes
• Memory contract pricing rolls over (TrendForce monthly) — bear cohort, pair may invert
• Hormuz disruption tightens HBM materially — reframes MU as a candidate on the other side; HDD pair unchanged

Limitations

Sizing requires iev regress on each ticker for σ_idio and α_orth — not yet run. Form 4 insider activity for the last 90 days on STX/WDC/MU/SNDK not swept; counterparty identity unknown. STX options chain showed sparse OI on liquid strikes, blocking options-implied probability extraction. These gate position size, not memo direction. The pair recommendation is provisional until at least the regressions land.

Evidence