Marvell Technology

1 · What's new since 2026-06-21 (filings, transcript, tape)

No new 10-Q (Q2 FY2027 reports in August; the on-disk 10-q-2026-q2.md remains the latest). CSV templates (financials/segments/guidance/customers) still empty — unchanged from prior pass.

---

2 · Performance / financials delta

No new quarter has printed since the prior dossier. The financial facts are unchanged from Q1 FY2027 (ended 2026-05-02); the transcript ingest now grounds them at the primary level (previously web):

• Revenue $2.418B, +9% QoQ / +28% YoY, above guide midpoint. Data center $1.8B = 76% of total, +11% QoQ / +27% YoY.
• GAAP GM 52.1% / non-GAAP GM 58.9%; GAAP OM 14% / non-GAAP OM 35%. (Reconciles to the prior dossier's 10-Q read; the $0.04 GAAP EPS remains driven by the Celestial earnout mark — carried forward.)
• Record OCF $639M; capital returned in Q1 = $200M buybacks + $54M dividends.

New primary-grounded forward granularity from the call (all management guidance given 2026-05-27) — sharper than the prior web-level read:

• FY2028 revenue guide RAISED to $16.5B — $1.5B higher than the prior-quarter outlook (~$15B), +~45% over FY2027. FY2027 ~$11.5B (+~40%).
• Data center FY2027 growth ~50% YoY.
• Interconnect now guided >70% YoY growth in FY2027 — a sharp upward revision from ~50% the prior quarter. (Direct read on the optical/interconnect thesis — momentum accelerating, not just holding.)
• DCI (data-center-interconnect) module revenue → $1B annualized in FY2028 (double FY2026).
• Scale-out switching → >$600M FY2027, >$1B annualized FY2028.
• Custom silicon >20% FY2027, "more than double" FY2028, and explicit "line of sight to over $10B annual custom-silicon revenue in FY2029". (New long-dated custom-ASIC anchor.)
• Target operating-margin model: upper end of the 38–40% range as FY2028 progresses. (Note: materially above the ~32–33% non-GAAP op-margin the prior dossier's projection lens assumed — a carry-forward-projection upside flag, not re-modeled here.)
• $1B of supply-chain prepayments planned in FY2027 to lock capacity (funded from internal cash). Record-breaking bookings cited.
• Acquisitions closed: Celestial AI (photonic fabric) + Polariton (plasmonic silicon photonics — NEW name vs prior dossier) + XConn. Optics roadmap tuck-ins continue.

---

3 · Guidance + catalysts delta (custom-silicon / interconnect momentum)

Guidance status: INTACT and reaffirmed. The last official guide touch was the June 11 reaffirmation of the Q2 FY2027 outlook ($2.7B midpoint, +~35% YoY — carried forward) [primary: SEC 8-K, acc 0001193125-26-267688]. The July de-rate is sentiment/macro, not a company guide cut — July coverage notes CEO Murphy's raised FY27/FY28 outlooks still stand.

New catalyst signals (custom-silicon + interconnect):

• KeyBanc (John Vinh) raised PT to $400, Overweight (2026-07-14) — bull case explicitly names: (a) an imminent 2H-2026 volume ramp of Amazon Trainium 3, and (b) a NEW design win for Google's "Merope" LPU, projected up to ~$12B lifetime revenue. (Two-sided vs the prior dossier: the Trainium-3-ramp claim complicates — does not cleanly confirm — the Dec-2025 "lost Trainium 3/4 to Alchip" scare; the Google "Merope" LPU is a fresh named win where the prior dossier had Google only "in talks." Treat both as analyst assertions pending company confirmation — FLAG for deeper review.)
• Custom-silicon framing in market coverage now cites 18 XPU/XPU-attach socket wins, many in volume production, against a >$75B lifetime revenue funnel — a sharper articulation of the prior "18 programs"; note the $75B is a Marvell-sourced funnel figure repeated by third parties, not independently verified here.
• NVIDIA partnership (optics + NVLink Fusion + AI-RAN) reconfirmed on the Q1 call — carried forward from prior dossier ($2.0B convertible preferred).

De-rate catalyst chain (the moving picture on the tape):

• ~$324 (6/18) → ~$196 (7/24), ≈ −40% in ~5 weeks. Drivers: hyperscaler-capex-slowdown fears (revised capexCapital expenditureMoney spent on long-lived things — buildings, machines, servers — rather than on running costs. forecasts), a Korean-semi rout (SK Hynix worst single day in 18 years), and AI-monetization doubts; a specific ~8% down day 7/15–16 ("AI CapexCapital expenditureMoney spent on long-lived things — buildings, machines, servers — rather than on running costs. Slowdown Fears Weigh on Chips; Broadcom, AMD, Intel slide").
• Mechanism (unchanged thesis, now realized): 76% DC revenue + β≈2.2 + a ~64–65x P/E make MRVL the market's purest hyperscaler-capex beta — "every hyperscaler capex headline is a Marvell headline". This is precisely the "hyperscaler-wobble re-rate" the prior dossier flagged; the stock is now at the top of that dossier's stated "own the dips ~$160–200" zone and below the then-average analyst PT.
• S&P 500 inclusion (effective 6/22) forced-buying appears to have marked the local top; the subsequent unwind + macro rotation is the backdrop.

---

4 · Forensic / earnings-quality delta

Carry forward the prior dossier's Lens 10 in full (clean-but-noisy accounting: recurring Celestial earnout mark through "Other expense," SBC ~$590M/yr flattering non-GAAP, $13.9B goodwill = impairment risk, no SEC LR/AAER). New this pass:

• Governance/comp signal (NEW). Say-on-pay passed only ~56% (327.6M FOR / 258.6M AGAINST) and shareholders rejected the board's position by supporting scrutiny of the combined Chair/CEO only to the extent of a failed independent-chair proposal (215.9M FOR / 369.2M AGAINST). A DEFA14A (additional proxy solicitation) was filed 6/11 — consistent with the company having to campaign against proxy-advisor pushback. Meaningful discontent with a comp structure the prior dossier already flagged as complex (mega-grants + customer-warrant dilutionDilutionIssuing new shares, so each existing share owns a smaller slice of the same company. + the new Durn package). Not an accounting red flag — a governance one.
• CFO change quality. Outgoing CFO Meintjes's exit is documented as no disagreement with a long advisory tail (thru 4/17/27) [primary: SEC 8-K, acc 0001193125-26-267688] — a soft-landing transition, not a resignation-under-fire pattern. Durn's package is standard (base $850K, 120% target, $1M sign-on, 25,877 TSR/EPS PSUs) — no unusual terms.
• Dilution watch (unchanged, small addition). The 424B7 adds only 146,504 XConn resale shares (immaterial) [primary: SEC 424B7]. The structural dilution stack (NVIDIA preferred up to ~21.8M sh, Celestial earnout up to ~22–24M sh, customer warrants) is carried forward — still the item to police.
• Insider activity. A 7/15–16 Form 4 cluster plus routine 144 notices — flagged for a dedicated read (see open items); no conclusion drawn here (could be 10b5-1Rule 10b5-1 planA schedule an insider sets in advance saying when their shares will be sold, so the sales cannot be timed on private news. A sale made OUTSIDE such a plan is a live decision, which is why it carries more information. vesting into the de-rate or discretionary sales).

No new SEC Litigation Release or AAER surfaced this pass; the prior "no material regulatory/legal findings" conclusion stands (export-control exposure to China remains the live policy risk, not an enforcement matter).

---

5 · Valuation snapshot (as of ~2026-07-24, all unless noted)

**Forward-multiple cross-check ** (using the prior dossier's carried-forward non-GAAP EPS anchors — FY27e ~$3.60, FY28e ~$5.60; not re-modeled): at ~$196, MRVL is ~54x FY27e and ~35x FY28e, down from ~86x / ~55x at $310. The valuation has moved from "25% above the sell-side" to "~29% below the average PT" — the single biggest change since the prior pass. (Management's own 38–40% target op-margin, if hit, would lift those forward EPS anchors and lower the effective forward multiple further — carried-forward projection lens, flagged not rebuilt.)

---

6 · Carried forward unchanged (read the 2026-06-21 dossier)

Lenses 1 (business), 2 (supply chain), 3 (moats), 4 (segments), 9 (management track record), 11 (projection scaffolding), 12 (bull/bear), 13 (devil's advocate), 14 (management questions), 15 (position framing) are unchanged in substance and carried forward from the previous dossier. Notably still-valid: fabless ASIC + optical-DSP + interconnect pure-play; TSMC/CoWoS single-source chokepoint; Marvell + Broadcom ~95% custom-ASIC duopoly (MRVL ~20–25%); severe concentration (Distributor A ~37%, Customer A ~14%, ~76% DC, China ~44% of shipments); Matt Murphy's >10x-EV record; the 15 management questions.

---

7 · Open items / flags for deeper review

1. Trainium 3 status is now contradictory. Dec-2025 scare said Marvell lost Trainium 3/4 to Alchip; KeyBanc's Jul-14 bull case says a 2H-2026 Trainium 3 volume ramp is imminent. These cannot both be fully true — resolve Marvell's actual Trainium-3 content (attach vs full XPU) before any thesis promotion. Primary-source gate applies.
2. Google "Merope" LPU (~$12B lifetime) is an analyst-sourced named win — not yet company-confirmed. Verify against Marvell disclosure / the next call.
3. 8-K filename collisions. The ingest script's calendar-quarter naming collapsed the June 11 (CFO/reaffirm) and April 15 8-Ks into the 8-k-2026-q2 slot (only June 25 written to disk); May 27 is the earnings 8-K already reflected. June 11 was retrieved via curl and cited above; April 15 predates the prior dossier. Consider a per-accession naming scheme so future 8-K refreshes don't silently skip same-quarter filings.
4. Empty CSV templates persist (financials/segments/guidance/customers) — populate from the on-disk 10-K/10-Qs to make the next refresh delta-clean (all FY2024–26 + Q1 FY2027 figures live in the prior dossier and this file).
5. Insider Form 4/144 cluster (Jul 15–16) — pull the actual transactions to distinguish 10b5-1 vesting from discretionary selling into the de-rate.
6. Say-on-pay ~56% + failed independent-chair vote — worth watching whether the board responds (comp redesign) before FY2028 grants; a governance-quality thread, not yet a red flag.
7. Next hard catalyst: Q2 FY2027 print (~late Aug 2026) vs the reaffirmed $2.7B (+35%) guide — the first test of whether the July capex fear is showing in Marvell's own numbers or is purely sentiment.

---

Refresh completed 2026-07-25. Grounding: 2 new 8-Ks (June 25 ingested; June 10/11 + July 9 via SEC primary), 1 resale 424B7 (SEC primary), 1 newly-ingested Q1 FY2027 transcript (fool.com → transcripts/2027-q1.md). Live price/consensus/analyst = web-labeled. No new 10-Q (Q2 pending). No verdict, no call — data + mechanics only.