Marvell Technology

Murphy on the Q4 FY26 call (Mar 5 2026, verbatim): custom xPU + NIC revenue scaled "from 0 revenue to $1.5 billion in fiscal 2026"; FY27 guide "more than 20% year-over-year"; FY28 "we expect it to double again" (i.e., ~$3.6B+). 50+ active designs / 10+ customers / 18 multi-generational sockets disclosed across FY26 calls. Trainium 2/3 fully subscribed at AWS; Maia 100 in production with 200 ramping at Microsoft; Google TPU 8i + companion MPU + Meta Arke design wins are the public-domain candidate set. The thesis is that MRVL is the toll-collector for hyperscaler ASIC tape-outs without committing to any one customer's capex cycle.

Apr 22 2026 PR (#1020) announced acquisition of ETH-Zurich-spinout Polariton (POH plasmonic-organic-hybrid modulators). Primary-source verification (Horst et al., Optica 12, 325, 2025, DOI 10.1364/OPTICA.544016): Polariton's commercial-grade modulators use Lightwave Logic's Perkinamine chromophore as the active EO medium — explicitly acknowledged in the paper. Marvell now owns ETH-derived plasmonic device IP (US 10,571,724; US 11,764,873) but every commercial-grade unit still consumes LWLG-patented material. Substituting LWLG (e.g., NLM Selerion-HTX, US 12,187,827) requires a 2–3 year requalification cycle. No 8-K filed by 2026-04-29 — terms below MRVL's voluntary materiality threshold (XConn precedent implies sub-$1B headline).

Celestial AI closed Feb 2 2026: $3.25B headline ($1.0B cash + 27.2M MRVL shares = $2.25B equity per Form D 0001835632-26-000001) with earnout up to $5.5B (1/3 trigger at $500M cumulative revenue by FY29; full at $2.0B). Murphy Q3 FY26 framework: $0 FY26/FY27 revenue → ~$200M H2 FY28 → $500M ARR exit Q4 FY28 → $1.0B+ ARR exit Q4 FY29. Meintjes Q4 FY26: $50M annual OpEx absorption begins Q1 FY27 against zero revenue — a margin headwind. The DCF base case credits ~$1.5B Celestial revenue by FY30; bull is $1.5B+ FY30; bear is delayed ramp / deal goodwill.

Mar 31 2026: NVIDIA disclosed a $2B strategic equity investment in MRVL tied to NVLink Fusion partner validation. The same NVIDIA owns the in-house silicon roadmap (Spectrum-X, ConnectX, custom InfiniBand DSP) that competes directly with MRVL's Inphi-derived PAM4/coherent franchise. The DCF stress-test (§6) prices a "NVIDIA in-sources networking" scenario at -$700M FY28 revenue / -$0.65 EPS / -$8 DCF $/sh / -$10 SOTP $/sh. Asymmetry: NVIDIA's quarterly architectural decisions (Blackwell Ultra, Rubin) move MRVL's optical TAM more than any single hyperscaler customer. The Apr 9 2026 Barclays upgrade to Overweight ($156 PT) explicitly cites Marvell as the optical winner of the NVIDIA + AVGO arms race; the bear inversion is that NVIDIA in-sourcing flips MRVL from beneficiary to displaced.

10–15% of MRVL's optical interconnect revenue ships through China-based optical-module customers (Innolight, Eoptolink) using Marvell PAM4 / coherent DSPs. The Oct 2022 / Oct 2023 / Dec 2024 BIS expansions targeted AI accelerators + advanced HBM + lithography but optical modulators remain EAR99 today. The DCF stress test sizes the China-tightening scenario at -$575M FY28 revenue / -$0.55 EPS / -$7-8 $/sh — meaningful but absorbable. Two trigger events to watch: (a) BIS expanding the controlled list to include >800Gbps DSPs as a category; (b) Innolight/Eoptolink getting Entity-Listed (precedent: ZTE 2018, Huawei 2019).

Q4 FY26 explicit guide (Murphy verbatim, Mar 5 2026): "We expect Marvell's overall revenue in fiscal 2028 to grow close to 40% year-over-year, reaching approximately $15 billion." "Driving our non-GAAP EPS to well over $5." The Q1 FY27 call (~May 29 2026 target; T-30 days as of 2026-04-29) is the first opportunity for management to either (a) reaffirm verbatim with the same noun phrases, (b) annotate upward ("$15B+") to credit Polariton + Celestial + NVIDIA NVLink Fusion contributions, or (c) hedge ("in the range of $14-15B"). Any softening would crystallize the Stifel customer-cliff thesis (Mar 10 2026 outlier bear note) and trigger a second downgrade wave. Even base-case DCF $98 / SOTP $114 sit ~30% below spot $154.83 — the market is pricing a bull-case fundamentals + bull-case multiples combination.

Pre-raise leverage 0.72×; post-raise pro-forma 0.92× — both well under MRVL's stated 1.5× ceiling. The Apr 6/8 2026 issuance: $1.0B / 5.300% / 10-year due Apr 15 2036 (YTM 5.315%; +97bps over UST 4.125% Feb-2036; ratings Baa2/BBB/BBB+). Use of proceeds: refinances $500M 1.650% 2026 notes + general corporate (M&A optionality). The new-issue cleared in line with existing 2035/2036 paper — no liquidity concession despite $1.0B size. Cumulative M&A consideration last 12 months: Celestial $3.25B + XConn ~$540M ($280M cash + 2.1M shares per FY26 10-K Note 16) + Polariton TBD (sub-$1B implied). FY26 capital return $2.245B (buybacks + dividends) sustained alongside the acquisitions — capital discipline thesis holds at investment-grade level.

Aug 14 2025: MRVL sold the Aquantia-derived Auto-Ethernet PHY business to Infineon for $2.5B all-cash. Eliminates ~$225-250M FY26 projected revenue but redeploys $2.5B at a moment when the data-center pivot needed funding (Celestial close Feb 2 2026 was already in preparation). The $1.8B pre-tax gain inflated GAAP FY26 net income to $2.67B — a structural reason to read non-GAAP through this period. Reframes the M&A history: MRVL is now a pure-play data-infrastructure semiconductor incumbent (data-center share trending 75% → 80%+ of revenue per FY28 mix-shift framing). Cumulative M&A since 2017 is ~$28B+, all funneled into the data-center thesis.

Storage revenue trajectory (DCF segment build): FY26 $550M → FY27 $500M → FY30 $390M (-9% / -7% / -6% per year). MRVL's HDD/SSD controller business is the residual of the pre-Murphy era and competes against Broadcom, Western Digital captive, and Marvell's own legacy installed base maturity. Segment GM ~55-58% (estimate; comp-derived). The bear case here is not "storage collapses" — it is "the comms-and-other line is structurally the legacy pile and shouldn't be valued at growth multiples." SOTP applies 2x EV/Sales to storage vs. 14x to CXL/Celestial. Structural sunset is already in the model; the question is how cleanly carrier + enterprise networking + storage stabilize at $2.2B run-rate without dragging GM further.

XConn closed Feb 10 2026 at ~$540M total ($280M cash + ~2.1M MRVL shares per FY26 10-K Note 16; Form D acc. 0001835632-26-000002 confirmed $200M equity portion to 40 accredited investors). Founder Gerry Fan (ex-Broadcom/Marvell) continues. Apollo + Apollo 2 CXL+PCIe switch IP folds into the data-center connectivity portfolio. Disclosure pattern (S-8 → 8-K Item 8.01 → 424B7 → Form D → 10-K Note 16; no Item 1.01/2.01 acquisition 8-K) sets the template for Polariton's expected disclosure path. Modeled FY27 contribution ~$50-100M ramp; first contribution Q3 FY27. The competitive frame: XConn vs. Astera Labs (ALAB) Aries / Leo / Scorpio in CXL switching — ALAB trades 18-22x EV/Sales as a pure-play; SOTP applies 14x to MRVL's CXL/Celestial/Tanzanite line as a haircut to the pure-play multiple.

MRVL's historic IAD cadence has been April (the Apr 2024 AI-Day introduced the original $1.5B AI-revenue framing). As of 2026-04-29, no 2026 IAD is publicly scheduled — the April cadence is broken. The Q1 FY27 call (~May 29 2026 target) is the single most likely announcement window because (a) MRVL needs an IAD to deliver Polariton + Celestial + XConn + NVIDIA-NVLink-Fusion in one integrated narrative; (b) the FY28 ~$15B framing requires a deeper-than-call quantitative scaffolding; (c) Murphy's COMPUTEX 2026 keynote (Jun 2 2026) sits awkwardly between a missing IAD and a needed reframing. Questions only IAD can answer: disaggregated FY28 revenue bridge, customer-by-customer custom-silicon contribution, multi-year R&D allocation, capex (TSMC SoIC + CoWoS), Polariton purchase-price disclosure (Inphi precedent: voluntarily disclosed at IAD).

SEC v. Marvell + Weili Dai (Case CV 08-2367-HRL N.D. Cal., settled 2008-05-08 per SEC Press Release 2008-82 and Litigation Release LR-20559): Marvell paid $10M civil penalty + permanent injunction; Dai paid $500K + 5-year officer/director bar; no admit/deny. Underlying conduct: $362M income overstatement across FY2000–FY2006 from stock-option backdating. Sutardja was NOT named defendant. Separately, the 2016 "pull-in" revenue-recognition audit-committee investigation (PwC → Deloitte change) triggered Sutardja and Dai's Apr 2016 departures and Murphy's Jul 2016 CEO appointment from Maxim Integrated. Both episodes pre-date current management; relevant to leadership-history audit completeness but not as a current governance flag. Murphy's tenure (2016-present) is the relevant accountability window.