Phase Space AI

Trade Construction

Marvell Technology [MRVL]

Marvell Technology, Inc. [MRVL] — Trade Construction

Spot $174.36 (2026-07-28) · 12-month target $156 (−10.5%) · Valuation Criteria FAIL

The memo issues no position verdict. The Valuation Criteria is BINDING on the long-only absolute-return strategy and FAILS here.

Liquidity Criteria — chain pulled live

Pulled from Alpaca 2026-07-28/29.

15-Jan-2027 calls (171 days), IV 92–95%

Strike Open interest Bid Ask Bid size Ask size Delta
175 1,262 34.43 37.44 164 265 0.582
190 4,509 29.87 31.52 1 120 0.527
200 2,029 26.69 28.60 37 149 0.493
220 813 21.75 24.90 47 360 0.435
250 1,475 17.03 19.09 27 294 0.359
280 1,839 12.86 15.34 29 404 0.296

16-Oct-2026 calls (80 days), IV 96–103%

Strike Open interest Bid Ask Delta
175 20 22.70 25.78 0.521
180 133 21.46 23.52 0.496
200 1,079 15.76 18.07 0.406
250 2,071 6.98 8.37 0.226
280 6,765 4.33 5.77 0.160

Liquidity Criteria: PASS, with two warnings that must be stated rather than buried.

  1. Implied volatility of 92–95% on the January-2027 chain, and 96–103% in October. That is nearly double AMD's ~74% and almost twice AVGO's ~54%. The January-2027 at-the-money straddle prices roughly a ±47% move. Long premium here is prohibitively expensive: the $175 call asks $37.44, 21.5% of spot, for 171 days.
  2. October-2026 near-strike open interest is genuinely thin — 20 contracts at $175, 27 at $165 and $170, 133 at $180. That is HCA territory (18 contracts across a whole chain) at the exact strikes a near-the-money position would use. The October chain is not investable at the money at size, even though the offer sizes look adequate. Bid/ask on the $175 October call is $3.08 wide on a $24 mid — a 12.7% round-trip.

The January-2027 chain is genuinely liquid (four-figure open interest, 200–400 contracts on the offer). The equity is a $159.5bn market cap and presents no constraint.

Why no long is constructed

Four independent reasons:

  1. The implied path requires a 54.2% five-year CAGR — 41.6pp above the corrected trailing record and 26.6pp above the latest quarter, implying $76.1bn of FY2031 revenue from $8.7bn (8.7x). No cell in the exit-multiple sensitivity produces a PASS, including at twice the anchor.
  2. The trailing record moves the wrong way. Data-centre year-over-year growth fell +98% → +21% across five quarters, in the exact end market that must compound at 54%. Over the same period Broadcom's AI line went +46% → +143%. Same market, overlapping customers, opposite directions.
  3. Concentration is the worst in the cluster with nothing to offset it. Distributor A at 45% of revenue (from 24% two years ago), two customers at 61%, three at 75% of gross receivables — and no committed order book at all ($63.7m of deferred revenue, no RPO disclosure). This is the CRDO configuration one tier larger, and CRDO's precedent is a −40.9% sequential quarter.
  4. The 12-month target is 10.5% below spot on the same method that gives AVGO +58.5%.

Within the cluster, the same capital buys AVGO — where the mechanism is a $164.6bn firmly committed non-cancellable order book, the trailing growth accelerates rather than decelerates, receivable concentration is 44% rather than 75%, volatility is 47% rather than 77%, and the 12-month target is above spot. On slot competition MRVL loses to AVGO on every axis the framework scores.

The Short Mechanism Criteria — the only partial score in the cluster

criteria.md requires both conditions: "decelerating growth and exhausted margin runway (operating margin already high and already expanded, incremental leverage visibly spent)."

Test MRVL Verdict
Decelerating growth Data-centre YoY +98% → +78% → +76% → +69% → +38% → +21%, then +27%. Total revenue +63% → +22%. Satisfied on the trailing record
Exhausted margin runway TTM operating margin 16.0% against a 51.5% gross margin. R&D is 27% of revenue. Turned operating-profitable only in FY2026. Q2 FY27 guide implies 17.0%. Not satisfied — substantial runway remains

Short Mechanism Criteria: NOT ESTABLISHED. Both conditions are required and only one is met. Marvell is the only name in the cluster to score even partially, and it still does not qualify — a 16% operating margin on a 51% gross margin with 27%-of-revenue R&D has visible operating leverage ahead, and management guides to re-acceleration. Logged for the relative-value fork; acted on by nothing on the long-only fork.

The discipline point: MRVL is simultaneously too expensive to own and structurally wrong to short. That is a legitimate and common state, and the criterion exists to stop a FAIL on Valuation being read as a short.

If the book wants theme exposure

Documented, not recommended.

A. Relative value — long AVGO, short MRVL. The cleanest pair in the cluster: identical end market (Marvell's 10-K names Broadcom as a direct competitor), opposite growth trajectories (+143% versus +21% in the AI/data-centre line), opposite contracted positions ($164.6bn committed RPO versus none), and MRVL more stretched against its own history (90.5th versus 80.3rd percentile). Requires the RV fork and borrow work this long-only memo does not cover — borrow, short interest and days-to-cover are explicitly not part of the Liquidity Criteria here.

B. Defined-risk long, if overruled. Jan-2027 $190/$250 call spread: long $190 at $31.52, short $250 at $17.03 → net debit $14.49, maximum value $60, maximum payoff 4.1x, breakeven $204.49 (+17.3%). Open interest 4,509 and 1,475; both fillable with 120–294 contracts on the offer. The cap at $250 sits above the +33.7% multiple-holds bull case of $233, so it is a bet on the multiple exceeding its own 90.5th percentile.

At 94% IV, selling the $250 leg recovers 54% of the long leg's cost — which is the only reason a spread is tolerable here. Outright long premium at 21.5% of spot is not.

Sizing, if overruled

Inverse-volatility, per the framework's interim protection on the Downside Criteria.

AVGO AMD MRVL
252-day realised vol 47.2% 70.0% 76.9%
Jan-2027 ATM implied vol ~54% ~74% ~94%
Inverse-vol relative weight (realised) 1.00 0.67 0.61
Inverse-vol relative weight (implied) 1.00 0.73 0.57

MRVL is the highest-volatility name in the cluster on both measures and the weakest on the analysis. Inverse-vol sizing and the fundamental ranking point the same way, which removes the usual tension between the two.

Cluster constraint, and it is the binding one. AVGO, AMD and MRVL are one exposure. Marvell's own 10-K names the other two as direct competitors; all three sell custom accelerators, networking silicon and optics into the same handful of hyperscaler capex budgets. They must share a single concentration limit in portfolio_book.json. Adding MRVL alongside AVGO buys correlation with worse terms.

Monitoring triggers — what would reopen the name

Trigger Effect
Distributor A's share falls while revenue grows Genuine de-concentration. Disclosed quarterly in the significant-customer table. The single most important line.
Distributor A exceeds 50% of revenue Concentration dominates everything else; the name becomes uninvestable on this book regardless of valuation.
AR concentration loosens from three customers at 75% The credit exposure is the sharpest measure and it tightened this year (five at 72% → three at 75%).
Two more quarters of data-centre sequential growth above +9% Establishes that Q1 FY27's +11% is a re-acceleration rather than a bounce, and validates the guide.
Any committed-RPO disclosure emerges from the NVIDIA partnership Changes the valuation input materially, as $164.6bn did for Broadcom. Marvell currently discloses none.
GAAP-to-non-GAAP gap narrows from 28% of revenue Q1 FY27 was GAAP $0.04 versus non-GAAP $0.80. Should narrow as acquired intangibles amortise off; if it does not, the "records" stay non-GAAP indefinitely.
EV/Sales below the own-history median (10.48x), roughly $135 The implied-path arithmetic becomes worth rerunning.
NVIDIA converts the preferred, or increases its stake Governance and dilution event; the 10-K flags NVIDIA's "influence over matters submitted to our stockholders."

Ledger entry

Recorded so the call is falsifiable in both directions. No position is recommended.