| Aug 27, 2026 | -10.28% | — | Transcript briefing unavailable |
| May 27, 2026 | +3.09% | Q1 FY2027 | Read transcript briefingQuarter in one view- Q1 FY2027 revenue $2.418B, a record, +9% QoQ and +28% YoY — above the midpoint of the $2.4B ±5% guide. Non-GAAP EPS $0.80, $0.01 above the $0.74–$0.84 guide midpoint, +29% YoY. GAAP EPS came in $0.04 below guidance due to Celestial/XConn purchase accounting and earn-out obligations; management says this normalizes in Q2.
- The headline is another guidance escalation: FY2027 revenue raised >$0.5B to ~$11.5B (+~40% YoY, vs. ~$11B / +>30% last quarter); FY2028 raised ~$1.5B to ~$16.5B (+~45% YoY, vs. ~$15B / +~40%). Data center growth now ~50% FY2027 and ~55% FY2028 — accelerating off a 46% FY2026 base.
- Q2 FY2027 guided $2.7B ±5% (+12% QoQ, +35% YoY) — double-digit sequential growth vs. the "high single-digit" cadence promised last quarter. The $3B quarterly revenue milestone is pulled forward to Q3, one full quarter ahead of plan; Q3 and Q4 each guided to grow ≥10% sequentially, with YoY growth reaching ~50% by Q4.
- Interconnect FY2027 growth raised again: >70% YoY vs. >50% last quarter and ~30% in September — the third raise in three quarters. Custom FY2028 raised from "at least doubling" to "more than double."
- New strategic developments: expanded NVIDIA partnership (optics/silicon photonics collaboration, NVLink Fusion integration, AI-RAN on OCTEON) including an NVIDIA equity investment; acquisition of Polariton (plasmonic silicon photonics, >1 THz modulator bandwidth demonstrated) for 3.2T+ DCI/coherent-lite roadmaps.
- Data center Q1: $1.83B (76% of revenue), +11% QoQ, +27% YoY. Comms/other $585M, +3% QoQ, +29% YoY.
What management is focused on- Selling acceleration, not just growth: Murphy's framing is that growth rates are rising each year (46% → 50% → 55% data center) and every guidance raise is backed by "demand trends we are seeing today and programs already in execution," record bookings, and secured supply.
- The NVIDIA partnership as validation and TAM expansion: three pillars — silicon photonics collaboration for scale-up, NVLink Fusion (Marvell as the bridge between custom and NVIDIA architectures), and AI-RAN. Murphy positioned NVLink Fusion as giving hyperscalers "complete flexibility to mix and match," with Marvell "uniquely providing the bridge."
- Interconnect as "the star of the show": >70% FY2027 growth, with Murphy explicitly telling Seymore there is "upward bias for sure" to the FY2028 interconnect outlook — 1.6T content step-up, DCI ramp, AEC/retimers, and scale-up optics (~$300M FY2028, more than double the prior ~$150M Celestial-only outlook).
- Scale-up as the next land-grab: Murphy stressed scale-up switching revenue is "very little or nothing" in the $16.5B FY2028 outlook — pure upside — and that Marvell supports all three protocols (UALink, ESUN, NVLink via Fusion). Each Tier-1 scale-up engagement framed as a "multibillion-dollar lifetime revenue opportunity."
- Supply as competitive weapon: ~$1B of supplier prepayments planned this fiscal year starting Q2, 5-year forecasts shared with a small number of key suppliers — Koopmans explicitly credited "doing what we said we were going to do" through the constrained period since 2020–21.
- Custom long-range credibility: Murphy confirmed the >$10B FY2029 custom target (20% share of a ~$55B TAM) and walked Arya through the math; new Tier-1 XPU program is "about 1/3" of FY2028 custom growth, with the FY2028 double split roughly thirds across existing programs, XPU-attach, and the new program.
Key numbers and quarter mechanics- Q1: revenue $2.418B (+9% QoQ, +28% YoY). Data center $1.83B (+11% QoQ, +27% YoY, 76% of revenue). Comms/other $585M (+3% QoQ, +29% YoY).
- Margins: GAAP GM 52.1% (top of 51.4–52.4% guide); non-GAAP GM 58.9% (within 58.25–59.25% guide — the drift from 59.7% → 59.0% has stabilized for one quarter but not recovered). Non-GAAP OpEx $577M vs. ~$575M guided; non-GAAP operating margin 35%; GAAP operating margin 14% (down from 18.2% on acquisition costs).
- Cash flow/balance sheet: CFO a record $639M (vs. $374M in Q4) — the working-capital concern from last quarter reversed hard. Inventory $1.4B, "almost flat" QoQ after the $374M Q4 build. Buybacks $200M; dividends $54M. Debt $4.96B (up from $4.47B); leverage 1.44x gross / 0.32x net.
- Q2 FY2027 guide: revenue $2.7B ±5%; GAAP GM 52.1–53.1%; non-GAAP GM 58.25–59.25% (flat again); GAAP OpEx ~$960M; non-GAAP OpEx ~$600M (vs. "flat in Q2" guided last quarter — actually +$23M sequentially); non-GAAP other expense ~$35M; non-GAAP tax 11%; basic 899M / diluted 915M shares (up from 876M/883M on Celestial/XConn deal shares plus NVIDIA investment shares); GAAP EPS $0.32–$0.42; non-GAAP EPS $0.88–$0.98 (+16% QoQ at midpoint — leverage resumes).
- FY2027 non-GAAP OpEx ~$2.45B (includes Celestial/XConn). FY2028 non-GAAP OpEx guided to grow mid-to-high teens vs. 45% revenue growth — management now targets the upper end of the 38–40% operating-margin model during FY2028.
- ~$1B of supplier prepayments planned in FY2027, starting Q2, funded from balance sheet and operating cash flow — a material new working-capital commitment to monitor against CFO.
- Q2 end-market color: data center mid-to-high-teens sequential, mid-40s% YoY; comms/other down mid-single-digit sequentially, up high-single-digit YoY.
Product and launch scorecard- Interconnect (>70% FY2027 growth, third raise): 1.6T ramping post-H2 FY2026 launch with "another substantial step up" in FY2028; 800G demand still strengthening; 400G/lane demonstrated (OFC April 2025) for the next PAM4 generation. TIAs/drivers disclosed for the first time as approaching a >$1B annualized run rate "in the next few quarters."
- DCI/scale-across: first-ever revenue disclosure — ~$500M in FY2026, line of sight to $1B annualized during FY2028. Shipping to all five major U.S. hyperscalers; >1M silicon-photonics DCI modules shipped over the past decade; industry's first secure 1.6T ZR/ZR+ on 2nm coherent DSP sampling this year. Scale-across bandwidth requirements cited at >10x current front-end DCI.
- Scale-up optics: FY2028 revenue now ~$300M, "more than double" the prior ~$150M Celestial-only outlook — now includes Marvell's own NPO/CPO products. Celestial's photonic fabric "selected by a Tier 1 hyperscaler for its next generation of XPU scale-up networks" (first deployment confirmation); third-gen 6.4T light engine engaged with multiple Tier-1 customers; 15B device-hours across 4 SiPho generations cited as reliability proof.
- Switching: FY2027 scale-out switch revenue >$600M reaffirmed (doubling YoY); now "tracking to more than $1 billion in annualized revenue in fiscal 2028" — a new disclosure. 51.2T ramping with existing and new customers; 100T platform engaged; 200T roadmap in execution. Scale-up switching (UALink/ESUN/NVLink) has multiple Tier-1 engagements but contributes ~nothing to FY2028 numbers.
- Custom: FY2027 >20% growth reaffirmed; FY2028 raised to "more than double" with the same ~1/3, 1/3, 1/3 split across existing programs, 10+ XPU-attach programs, and the new Tier-1 XPU (firm requirements for all of next fiscal year, milestones being hit). Several new custom designs won since last quarter (~2-year revenue lag; FY2029 "insurance policy," not needed for current targets). >$10B FY2029 custom target reaffirmed.
- XPU-attach: >10 programs reaching higher production volumes in FY2028, demand exceeding prior forecasts; NIC and CXL memory-attach called out, driven by inference KV-caching requirements. CXL adoption getting an additional push from the tight memory cycle (Meintjes).
- AEC/retimer: 3 Tier-1 U.S. hyperscaler AEC wins (consistent with last quarter); combined revenue more than doubling in FY2027, rapid growth continuing FY2028.
- Polariton (new acquisition): plasmonic modulators >1 THz bandwidth, up to 10x current SiPho/TFLN — targeted at 3.2T+ DCI and coherent-lite roadmaps. No revenue or cost figures disclosed.
- NVIDIA partnership: announced but unquantified — no revenue timing, sizing, or program detail given for any of the three pillars.
Sell-side read-through- Ten questions, again uniformly congratulatory in tone; zero pushback on the raised $11.5B/$16.5B framework. The street is now in "how high" mode, not "is it real" mode.
- Arya (BofA) did the math live on the >$10B FY2029 custom target — implying a $5–6B single-year custom increase from FY2028 to FY2029 — and pressed again on the new Tier-1 XPU program's exclusivity and disclosure timing. Murphy confirmed the number, gave the TAM/share derivation, but offered no exclusivity evidence and deferred detail ("as we progress through the year... we'll all gain confidence").
- Arcuri (UBS) asked the most strategically interesting question: is Marvell moving into the largest custom compute wallet, and is that in the forecast? Murphy's answer — current targets rest entirely on designs already won; newer wins are an "insurance policy" — is a meaningful conservatism claim that is checkable over time.
- Seymore (Deutsche Bank) directly challenged the interconnect deceleration assumption (70% → CapEx-indexed in FY2028) and got Murphy to volunteer "upward bias for sure" — a live upward nudge on FY2028 interconnect, similar to last quarter's "$5 is a floor" moment.
- Svanberg (Stifel) and Pajjuri (RBC) both probed scale-up; the key extraction was Murphy's statement that scale-up switching is "very little or nothing" in the $16.5B — explicit unmodeled upside.
- Leopold (Raymond James) got the FY2028 custom growth split confirmed at roughly thirds across the three drivers.
- Notably absent again: no questions on non-GAAP gross margin (58.9%, guided flat at 58.25–59.25% — now four quarters below the old ~60% norm, still unprobed), the GAAP EPS miss vs. guide, the Q2 non-GAAP OpEx increase vs. last quarter's "flat" promise, the ~$1B prepayment program's cash-flow impact, or any financial detail on the NVIDIA investment (share count impact visible only in the diluted share guide).
Management credibility- Delivery against last quarter's checkable items: revenue $2.418B vs. $2.4B ±5% (above midpoint); non-GAAP EPS $0.80 vs. $0.74–$0.84 (above midpoint); non-GAAP GM 58.9% within 58.25–59.25%; non-GAAP OpEx $577M vs. ~$575M; data center +11% QoQ vs. ~+10% guided; comms/other +3% QoQ vs. low-single-digit guided. Clean on every explicit commitment.
- Last quarter's watchlist items resolved positively: the $374M inventory build did convert (inventory flat at $1.4B, CFO rebounded to a record $639M); the sequential-growth promise kept (Q1 +9%, Q2 guided +12%).
- Credibility credits: the $3B-quarter milestone pulled in a full quarter early; new quantified disclosures (DCI ~$500M FY2026 → $1B annualized FY2028; TIAs/drivers → $1B run rate; scale-up optics ~$300M FY2028; switch >$1B annualized FY2028) give the street fresh checkable anchors; Murphy's FY2029 custom math was transparent under Arya's pressure; the "insurance policy" framing on unannounced wins is a disciplined claim.
- Credibility debits/risks: the guidance-raise cadence is now four quarters running (FY2027: $9.5B → $10B → $11B → $11.5B) — each raise has been validated so far, but the pattern means the street now expects raises, raising the bar for a merely in-line print. Q2 non-GAAP OpEx guided ~$600M vs. "flat in Q2" promised last quarter — a small but real miss on an explicit OpEx commitment. GAAP EPS missed guidance by $0.04 on acquisition accounting — explained, but the earn-out overhang was not quantified. The NVIDIA partnership was announced with zero financial specifics. Non-GAAP GM remains stuck below 59% with only "mix" as explanation, and no analyst is asking.
- Consistency note: Murphy's "1/3, 1/3, 1/3" FY2028 custom split was held consistent with last quarter even as the total went from "doubling" to "more than double" — internally coherent, and Leopold's question confirmed it deliberately.
What changed versus the prior quarter- FY2027 revenue raised >$0.5B to ~$11.5B (+~40% vs. +>30%); FY2028 raised ~$1.5B to ~$16.5B (+~45% vs. +~40%). Data center FY2028 growth raised to ~55% from ~50%. Interconnect FY2027 raised to >70% from >50%. Custom FY2028 raised to "more than double" from "at least doubling."
- The $3B quarterly revenue milestone moved from Q4 exit to Q3 — one quarter early; sequential growth cadence raised from high-single-digit to ≥10% for Q2–Q4.
- NVIDIA partnership announced (optics/SiPho, NVLink Fusion, AI-RAN) with an accompanying NVIDIA equity investment — new since last quarter; share count up ~23M diluted partly because of it.
- Polariton acquired (plasmonic photonics) — third photonics-related deal after Celestial and XConn.
- New quantified disclosures: DCI ~$500M FY2026 revenue and $1B annualized FY2028 line of sight; TIAs/drivers approaching $1B annualized; scale-up optics ~$300M FY2028 (vs. ~$150M prior); scale-out switching >$1B annualized FY2028; scale-up switching explicitly ~zero in FY2028 numbers.
- Working capital normalized: CFO record $639M, inventory flat — last quarter's bear point resolved. But ~$1B of supplier prepayments now planned for FY2027, a new cash commitment.
- Q2 non-GAAP OpEx guided ~$600M vs. "flat" promised last quarter; FY2028 OpEx growth framed at mid-to-high teens with 38–40% operating margin targeted during FY2028.
- Debt rose to $4.96B from $4.47B; net leverage down to 0.32x on higher EBITDA.
Bull case- Acceleration on acceleration: data center growth guided to rise from 46% (FY2026) to ~50% (FY2027) to ~55% (FY2028) — off an ever-larger base, with the $3B quarter arriving a quarter early and ~50% YoY growth by Q4.
- Interconnect keeps outrunning the model: >70% FY2027 growth (third raise), 1.6T content step-up in FY2028, DCI doubling toward $1B, TIAs/drivers adding a newly disclosed $1B run-rate business, and management openly flagging "upward bias" for FY2028.
- Custom is de-risking in layers: FY2028 more-than-double split roughly thirds across existing programs, 10+ attach programs, and the new Tier-1 XPU with firm full-year requirements — plus newer wins explicitly excluded from targets as an "insurance policy," and a confirmed >$10B FY2029 line of sight.
- Scale-up is a free option: ~$300M optics in FY2028 (doubled), switching essentially unmodeled, Celestial selected by a Tier-1 hyperscaler, and Marvell now the only player spanning UALink, ESUN, and NVLink Fusion — with NVIDIA's partnership and equity as ecosystem validation.
- Operating leverage resumes: Q2 non-GAAP EPS guided +16% QoQ at midpoint; FY2028 OpEx growing mid-to-high teens vs. 45% revenue; 38–40% operating margin targeted during FY2028.
- Supply secured with cash: ~$1B prepayments plus 5-year supplier forecasts — Koopmans's track record of delivering through constraint since 2020 supports the claim.
Bear case- The raise cadence is now the expectation: four consecutive FY2027 raises mean an in-line quarter could read as a miss; the Q2 guide (+12% QoQ) and ≥10% Q3/Q4 promises leave no room for a single soft print without breaking multiple explicit commitments.
- Gross margin remains stuck: 58.9% printed, 58.25–59.25% guided again — four quarters below the old ~60% norm, still explained only by "mix," and still unquestioned by the street. Custom dilution at a more-than-double FY2028 growth rate is the obvious structural headwind.
- FY2028's $16.5B leans harder on custom than ever: a more-than-double year where ~1/3 of growth is a new Tier-1 XPU program with no revenue history, and Arya's exclusivity question was again answered qualitatively. The FY2029 >$10B target implies a $5–6B single-year custom increase — a very steep hill Murphy himself acknowledged "looked" steep two years ago.
- OpEx discipline slipped at the margin: Q2 non-GAAP OpEx ~$600M vs. "flat" guided last quarter; GAAP EPS missed on acquisition accounting with the Celestial earn-out still unquantified.
- Cash is now committed forward: ~$1B of prepayments begins in Q2 against a record-CFO quarter — if demand pauses, Marvell holds both inventory and prepaid capacity. Debt rose ~$0.5B QoQ.
- The NVIDIA partnership is unquantified — strategically validating but with no disclosed revenue, timing, or margin structure, and it came with share issuance (diluted count up ~23M QoQ).
- Concentration and cyclicality unchanged: the growth story remains the top U.S. hyperscalers' CapEx, and Murphy is now explicitly planning for CapEx growth to "moderate" to 30%+ in FY2028 — the interconnect-outpaces-CapEx claim must carry the whole multiple.
Next-quarter watchlist- Q2 delivery vs. $2.7B ±5%, non-GAAP EPS $0.88–$0.98, non-GAAP GM 58.25–59.25% — and whether GM ever gets a real explanation; four quarters of sub-59% with no analyst pressure is now the call's biggest unasked question.
- The ≥10% sequential cadence: Q2 guided +12%; any stumble breaks the pulled-forward $3B-in-Q3 milestone and the ~50% Q4 YoY claim simultaneously.
- Interconnect >70% FY2027: 1.6T vs. 800G mix quantification (deferred again), TIAs/drivers progress toward the $1B annualized run rate "in the next few quarters," and DCI trajectory toward $1B annualized in FY2028 with 1.6T ZR/ZR+ sampling this year.
- Custom FY2028 evidence: production-plan locking for the new Tier-1 XPU (Murphy promised more confidence "as we progress through the year"), any detail on the several new socket wins, and whether the ~1/3, 1/3, 1/3 growth split holds as numbers firm.
- Scale-up optics toward ~$300M FY2028: Celestial HVM milestones at the named Tier-1 hyperscaler, Marvell-native NPO/CPO engagement conversion, and Polariton integration into the 3.2T roadmap.
- NVIDIA partnership specifics: any revenue sizing, program timing, or structure for the optics, NVLink Fusion, and AI-RAN pillars — currently zero financial detail.
- Cash mechanics: prepayment cadence within the ~$1B FY2027 plan starting Q2, CFO sustainability after the record $639M, inventory behavior, and debt trajectory.
- OpEx: Q2 non-GAAP OpEx at ~$600M (vs. the broken "flat" promise), FY2027 ~$2.45B total, and whether the FY2028 mid-to-high-teens OpEx growth frame survives another raise cycle.
- Switching: progress toward >$600M FY2027 and the newly disclosed >$1B annualized FY2028; 100T sampling status; UALink 115T sampling due H2 FY2027.
- Share count: diluted 915M guided — confirm the NVIDIA investment and deal-share issuance are fully absorbed and buybacks (~$200M/quarter) keep pace.
|
| Mar 5, 2026 | +18.35% | Q4 FY2026 | Read transcript briefingQuarter in one view- Q4 FY2026 revenue $2.219B, a record, +7% QoQ and +22% YoY — above the midpoint of the $2.2B ±5% guide, driven by data center demand. Non-GAAP EPS $0.80, $0.01 above the $0.74–$0.84 guide midpoint, +33% YoY. GAAP EPS $0.46 (guide $0.31–$0.41 — above the top end).
- FY2026 full year: revenue $8.195B (+42% YoY, ~+45% ex-auto); data center >$6B (+46%); custom doubled to $1.5B; non-GAAP EPS $2.84 (+81%, ~2x revenue growth); non-GAAP GM 59.5%; non-GAAP operating margin 35.3% (+640bps YoY).
- The headline is the guidance escalation: FY2027 revenue now "approaching $11B" (+>30% YoY) vs. ~$10B in December and ~$9.5B in September — Murphy explicitly framed this as roughly double the growth rate implied a few months ago. Q1 FY2027 guided $2.4B ±5% (+8% QoQ, +27% YoY), with revenue growing every quarter and Q4 exit >$3B.
- FY2028 outlook raised ~$2B to ~$15B (+~40% YoY): data center +~50%, custom at least doubling, interconnect outpacing CapEx, plus ~$250M combined Celestial AI + XConn revenue. Non-GAAP EPS framed as "well over $5" — later clarified by Murphy as a floor ("5-plus"), not a ceiling.
- Both acquisitions closed: Celestial AI and XConn (new — PCIe/CXL switch asset, not previously disclosed in the prior-quarter summary). Combined ~$75M added to FY2027 non-GAAP OpEx; no meaningful revenue until FY2028.
- Data center Q4: $1.65B (74% of revenue), +9% QoQ vs. high-single-digit guided, +21% YoY — sequential growth across optics, custom, switching, and storage. Comms/other $567M, +2% QoQ, +26% YoY.
What management is focused on- Selling the raised multi-year arc as validation, not hope: Murphy anchored the $11B/$15B path to "demand we are seeing now and designs already in execution," record bookings/backlog, and supply-chain alignment — and explicitly tied it back to the April 2024 calendar-2028 targets ("everyone thought we were nuts... we're very much on track").
- Interconnect re-rating: FY2027 interconnect growth raised from ~30% to >50% — Murphy conceded the prior CapEx-indexed modeling was "very conservative" and credited analysts who "dinged" the framework; optics now tracks accelerator/CapEx growth (~50%+), with momentum into FY2028.
- Custom credibility, round two: FY2027 custom now ">20%, biased higher" (vs. ≥20% prior); POs cover the full-year lead-customer next-gen program, now ramping production; H2-weighted profile reaffirmed; exit run-rate "intact with upward bias." FY2028 doubling rests on three named drivers: existing program growth, XPU-attach at high volume, and the new Tier-1 XPU program with "firm volume requirements for all of next year."
- Scale-up platform assembly: Celestial (CPO/photonic fabric) + XConn (PCIe/CXL switching) + UALink 115T (sampling H2 FY2027, production FY2028) + in-house Ethernet scale-up = an end-to-end optical scale-up stack. First CPO scale-up deployment at "one large customer" next year; $500M run rate Q4 FY2028, $1B Q4 FY2029 reaffirmed.
- Defending the custom business philosophically: Lipacis's "why not just do connectivity" question drew the most impassioned answer of the call — custom as "tip of the spear" for nodes/packaging/IP, NRE-underwritten by customers, and a flat denial of last year's socket-loss reporting ("it was all noise... analysts retracting notes... Do you see me blinking? You don't").
Key numbers and quarter mechanics- Q4: revenue $2.219B (+7% QoQ, +22% YoY). Data center $1.65B (+9% QoQ, +21% YoY, 74% of revenue). Comms/other $567M (+2% QoQ, +26% YoY).
- Margins: GAAP GM 51.7% (guide 51.1–52.1%); non-GAAP GM 59.0% (guide 58.5–59.5% — landed mid-range, confirming last quarter's step-down from 59.7% was real, not conservative). Non-GAAP OpEx $517M (guide ~$515M); non-GAAP operating margin 35.7%; GAAP operating margin 18.2%.
- Cash flow/balance sheet: CFO $374M — down sharply from Q3's record $582M. Inventory jumped to $1.39B, +$374M QoQ (from $1.01B), framed as working capital supporting growth — a notable reversal after two consecutive declines. Buybacks $200M (vs. $1.3B in Q3); dividends $51M. Debt $4.47B; leverage 1.38x gross / 0.57x net.
- Q1 FY2027 guide: revenue $2.4B ±5% (+8% QoQ, +27% YoY); GAAP GM 51.4–52.4%; non-GAAP GM 58.25–59.25% (flat-to-down again); GAAP OpEx ~$872M; non-GAAP OpEx ~$575M (up ~$58M on payroll-tax seasonality, merit increases, and Celestial/XConn — ~$75M annual non-GAAP OpEx from the deals); non-GAAP other expense ~$48M; non-GAAP tax 11%; basic 876M / diluted 883M shares; GAAP EPS $0.26–$0.36; non-GAAP EPS $0.74–$0.84 (flat QoQ at midpoint despite +8% revenue — OpEx step-up absorbs the leverage this quarter).
- FY2027 OpEx shape: flat in Q2, low-to-mid-single-digit sequential growth in Q3/Q4 — "well below" revenue growth; Murphy said the company approaches its target operating-margin model exiting the year (mid-30s now, floating up).
- Q1 end-market color: data center ~+10% QoQ including a seasonal decline in on-premise; comms/other low-single-digit sequential, ~+30% YoY.
Product and launch scorecard- Custom XPU: FY2026 doubled to $1.5B. Lead-customer next-gen program ramping with full-year POs; follow-on generation already in deep engagement. New Tier-1 XPU program: "firm volume requirements for all of next year," HVM planning underway, and Murphy disclosed the internal plug is "significantly lower" than reserved wafer/production capacity — an explicit upside tell. Arya's exclusivity question (first program was dual-sourced) got "two hands on the steering wheel," multi-generational engagement — confident but not a contractual exclusivity claim.
- XPU-attach: sized for the first time — "couple of hundred million" in FY2026, doubling in FY2027, ~$1B business by FY2028, against a $15B CY2028 TAM with a ~20% share goal (~$3B). NIC + CXL line of sight >$2B by FY2029 reiterated. CXL demand "accelerating, partly driven by tight memory supply" (DRAM reuse pitch); a hyperscaler white paper cited Marvell's Structura — named third-party validation.
- Optics: 1.6T in production since H2 FY2026 with "very strong bookings from multiple Tier 1 customers"; rapid FY2027 ramp with more growth in FY2028; 800G remains the volume majority through at least next year ("stronger for longer" validated). 400G/lane demonstrated for the 3.2T transition. Coherent Lite 1.6T Gen-1 shipping; Gen-2 with MACsec introduced.
- Scale-across/DCI (new emphasis): expects to supply DCI modules to all five major U.S. hyperscalers this year; announced industry's first secure 1.6T ZR/ZR+ on a new 2nm coherent DSP plus a 2nm 800G DSP; sampling later this year; DCI pluggable TAM cited at >5x growth by CY2030.
- Switching: FY2027 switch revenue raised again — >$600M vs. >$500M last quarter (which was itself a raise); 51.2T ramping, 100T sampling H1 FY2027. UALink 115T sampling H2 FY2027, volume FY2028 — on schedule.
- Celestial AI: closed; HVM engagement underway; $500M Q4 FY2028 / $1B Q4 FY2029 run rates reaffirmed; first CPO scale-up deployment at one large customer next year (XPU + switch side); "strong interest from a broad range of customers" post-announcement — still pre-revenue.
- XConn (new): PCIe Gen6/CXL 3.1 monolithic switch, up to 256 lanes; >20 customer engagements pre-acquisition; adds PCIe/CXL switch TAM; combined with Celestial, ~$250M FY2028 revenue.
- AEC/retimer: base finally disclosed — ~$200M range, more than doubling in FY2027; AEC wins now at 3 Tier-1 US hyperscalers (up from 2) plus model builders/OEMs; Golden Cable ecosystem program.
Sell-side read-through- Eleven questions, uniformly congratulatory; zero pushback on the raised $11B/$15B framework. The prior quarter's antagonists are fully converted — Sur opened with "Congratulations" and asked only about ramp shape.
- The most probing exchange was Arya's: he forced the first-ever XPU-attach sizing (couple hundred million → ~$1B by FY2028) and directly raised the dual-sourcing history at the lead XPU program as the template risk for the new Tier-1 program. Murphy's answer was confident but qualitative — no exclusivity evidence offered.
- Schneider (Goldman) caught the EPS math gap: $15B/~$5 implies revenue ~15% above peak consensus but EPS only ~half that gap. Murphy's response — "$5 is a floor," margins reach target model exiting FY2027, number "floats above $5" — was a live upward nudge on EPS, delivered casually.
- Reitzes extracted the cleanest account of the $2B FY2028 raise: interconnect was modeled too conservatively (CapEx-indexed vs. accelerator-indexed), plus firmer requirements and bookings.
- Curtis pressed for the FY2027 custom number (is it 30%?) and got only "north of 20%, biased higher, not significant enough to give a new number" — a deliberate refusal to re-anchor. His second-customer timing question got the capacity-reservation disclosure, the most concrete new custom evidence of the call.
- Notably absent again: no questions on the non-GAAP GM trajectory (59.7% → 59.0% → guided 58.25–59.25% — three quarters of drift, still unprobed), the $374M inventory build, the CFO drop to $374M, flat Q1 EPS despite +8% revenue, or the consumer/on-premise seasonal declines. Moore's supply-chain question was the only operational-risk query; Koopmans answered it cleanly (multi-year visibility, supply secured).
Management credibility- Delivery against last quarter's checkable items: revenue $2.219B vs. $2.2B ±5% (above midpoint); non-GAAP EPS $0.80 vs. $0.74–$0.84 (above midpoint); non-GAAP GM 59.0% within 58.5–59.5%; non-GAAP OpEx $517M vs. ~$515M; data center +9% QoQ vs. high-single-digit guide with the promised custom rebound; comms/other +2% QoQ vs. low-single-digit guide; Celestial closed on the Q1 FY2027 timeline; UALink sampling timing held. Clean against every explicit commitment.
- The "substantially stronger Q4" and "no air pockets" claims from last quarter are now validated by results — the digestion narrative survived its final test.
- Credibility credits: the multi-year framework has now been raised twice and both times the prior quarter's delivery backed it; switch revenue raised twice in two quarters ($500M → $600M) with FY2026's >$300M delivered; new quantified disclosures (XPU-attach sizing, AEC/retimer ~$200M base, capacity-vs-plug gap on the new XPU program) give the street checkable anchors; Murphy's willingness to concede the interconnect modeling error ("we had called it very conservatively, to be frank") is unusual candor.
- Credibility debits/risks: the guidance cadence itself is now a pattern — $9.5B → $10B → $11B in six months cuts both ways (visibility improving vs. numbers being made up as demand arrives); non-GAAP GM has drifted 59.7% → 59.0% → guided 58.25–59.25% with only "mix" as explanation and no analyst pressure; the FY2028 custom double leans on a new Tier-1 program where Murphy's own framing ("a lot of folks aren't really believing it's going to do anything") acknowledges street skepticism; the "it was all noise" dismissal of last year's socket reporting is fighting the last war — satisfying rhetorically, but the PO disclosure, not the denial, is what actually resolved it.
- Consistency note: Murphy now claims the April 2024 CY2028 targets (~$15B data center, later ~$18B on higher TAM) are "on track" — a checkable long-range claim that will be tested for three more years.
What changed versus the prior quarter- FY2027 revenue outlook raised ~$1B to ~$11B (+>30%), with the entire increase from data center (+40% vs. +>25% prior); interconnect raised from ~30% to >50%; custom from ≥20% to ">20%, biased higher"; comms/other held at +10%. FY2028 raised ~$2B to ~$15B (+~40%), data center +~50% vs. ~+40% prior.
- XConn appeared as a second, previously undisclosed acquisition (PCIe/CXL switching) — closed alongside Celestial; together ~$250M FY2028 revenue and ~$75M FY2027 OpEx.
- XPU-attach quantified for the first time (couple hundred million FY2026 → ~$1B FY2028); AEC/retimer base disclosed (~$200M); AEC Tier-1 hyperscaler count went 2 → 3.
- Switch FY2027 target raised again to >$600M; 100T sampling pulled into H1 FY2027.
- Gross margin: last quarter's unexplained Q4 step-down printed at 59.0% and Q1 guides flat-to-lower (58.25–59.25%) — the step-down is now a trend, still unexplained beyond "mix."
- Working capital flipped: inventory +$374M to $1.39B and CFO down to $374M after two quarters of inventory declines and record CFO — the cost of the ramp.
- Capital return normalized: $200M buyback vs. $1.3B in Q3 (the ASR was a one-time deployment).
- DCI/scale-across elevated to a headline growth vector (all five US hyperscalers, 2nm DSP launches) — new emphasis versus prior quarters.
Bull case- The guidance ladder keeps climbing and delivery keeps confirming: $9.5B → $10B → $11B FY2027 with Q4's beat-and-raise in between; Q1 guided +8% QoQ with acceleration every quarter and a >$3B exit — the YoY growth rate reaccelerates from 22% to 27% and rises all year.
- Interconnect is the engine and it was under-modeled: >50% FY2027 growth, 1.6T ramping at multiple Tier-1s with ASP tailwind, 800G stronger-for-longer, DCI to all five hyperscalers, and management now indexing to accelerator growth rather than CapEx — the segment is ~half of data center and compounding at ~50%.
- Custom de-risked and re-accelerating: full-year POs on the lead program, H2 weighting reaffirmed, FY2028 double underpinned by a new Tier-1 program where reserved capacity exceeds the internal plug — Murphy essentially told the street the FY2028 custom number is conservative.
- Scale-up optionality is now a portfolio: Celestial CPO (first deployment next year, $500M/$1B run rates), XConn PCIe/CXL (>20 customers), UALink on schedule, AEC/retimer doubling off a disclosed $200M base — multiple shots on the >$10B scale-up TAM.
- Operating leverage intact: FY2026 EPS +81% on +42% revenue; OpEx guided to grow well below revenue through FY2027; target margin model reached exiting the year; "$5-plus" FY2028 EPS framed as a floor.
- Supply secured: Koopmans flatly claimed supply for "this year, next year and beyond" — the binding constraint on AI hardware growth is addressed head-on.
Bear case- The company has now guided ~$11B FY2027 and ~$15B FY2028 on the record — a third of the FY2028 number depends on a custom double whose biggest single driver is a new Tier-1 XPU program with zero revenue history, plus attach ramps that are currently a "couple hundred million." Arya's dual-sourcing precedent question was answered with confidence, not contracts.
- Gross margin is quietly eroding: 59.7% → 59.0% → 58.25–59.25% guided, with only "mix" offered and no analyst pressing. Custom dilution is the obvious suspect; the ~60% norm is gone and management's "target model" language suggests margins, not GM, are the priority.
- Q1 EPS is flat at the midpoint ($0.79) despite +8% sequential revenue — OpEx seasonality plus $75M of acquisition OpEx absorbs a full quarter of leverage; the "EPS grows 2x revenue" story pauses here.
- Cash conversion weakened exactly as the ramp steepens: CFO $374M (from $582M), inventory +$374M to $1.39B. If the H2 custom weighting slips, that inventory becomes a liability, not an asset.
- The growth is increasingly second-half-loaded and exit-rate-dependent: "grow every quarter," ">$3B Q4 exit," "accelerating YoY each quarter" — any single-quarter wobble now breaks multiple explicit promises simultaneously.
- Celestial/XConn contribute nothing until FY2028 (~$250M combined), the $2B earnout hangs over the photonic-fabric targets, and CPO scale-up deployment rests on one large customer with copper persisting alongside.
- Concentration hasn't actually decreased: Murphy's answer to Seymore was that custom "isn't that big a percentage" and Marvell is diversified within the top four hyperscalers — which concedes the top four hyperscalers are the revenue base.
Next-quarter watchlist- Q1 delivery vs. $2.4B ±5%, non-GAAP EPS $0.74–$0.84, and non-GAAP GM 58.25–59.25% — whether GM stabilizes or the drift continues; someone should force the mix/custom-dilution question that has now gone unasked for two straight calls.
- The sequential-growth-every-quarter promise: Q1 is the first of four required prints; data center ~+10% QoQ guided despite the on-premise seasonal decline — check the on-premise drag's size.
- Custom H2 weighting evidence: lead-customer next-gen ramp milestones, any quantification of the ">20%, biased higher" FY2027 custom number (Curtis's ~30% trial balloon was neither confirmed nor denied), and new-Tier-1 program NPI milestones against the reserved-capacity plan.
- Inventory and CFO: whether the $374M inventory build converts to revenue in H1 or keeps climbing; CFO recovery toward prior levels.
- Interconnect >50% FY2027: 1.6T ramp quantification vs. 800G mix (Svanberg's question was deferred — "mix will have a better view as we progress"), DCI module wins at all five hyperscalers, and 2nm DSP sampling later this year.
- Switch trajectory toward >$600M FY2027; 100T sampling in H1; UALink 115T sampling H2 FY2027 on schedule.
- Celestial/XConn integration: first CPO scale-up deployment milestones at the named "one large customer," progress toward the $500M Q4 FY2028 run rate, and any second-customer engagement; XConn revenue evidence toward the combined ~$250M FY2028.
- XPU-attach: progress from "couple hundred million" toward the ~$1B FY2028 framing; CXL demand durability if memory supply loosens (the tight-DRAM tailwind was explicitly cited).
- OpEx discipline: Q2 flat sequential OpEx as guided, and whether the $75M acquisition OpEx stays contained; tax rate at 11%.
- Capital allocation: buyback cadence at the normalized ~$200M/quarter level now that both deals have closed.
|
| Dec 2, 2025 | +7.87% | Q3 FY2026 | Read transcript briefingQuarter in one view- Q3 FY2026 revenue $2.075B, a record, +3% QoQ and +37% YoY — above the midpoint of the $2.06B ±5% guide, driven by stronger-than-forecast data center demand. Ex-auto Ethernet, implied growth ~+6% QoQ and +41% YoY. Non-GAAP EPS $0.76, $0.02 above the $0.69–$0.79 guide midpoint, +77% YoY and +13% QoQ. GAAP EPS $2.20, dominated by the auto divestiture gain as pre-flagged.
- The guided Q3 custom decline happened as promised — but data center still beat: $1.52B (+2% QoQ vs. flat guided, +38% YoY), with optics, storage, and switching all growing double digits sequentially, more than offsetting the custom drop. Q4 data center guided high-single-digits sequential growth (~+20% YoY) on a custom rebound — the "substantially stronger Q4" claim is now testable.
- The quarter's defining event is strategic, not financial: the Celestial AI acquisition (photonic fabric for scale-up optical interconnect), announced same-day, with an AWS warrant extension (1M shares for photonic fabric products) effectively identifying the lead hyperscaler. Revenue targets: $500M annualized run rate by Q4 FY2028, $1B by Q4 FY2029; ~$50M annual OpEx; $2B earnout through FY2029; funded with stock plus cash, no new debt; close expected Q1 FY2027.
- Management broke its quarter-at-a-time convention and gave a full FY2027 framework: data center +>25% YoY (interconnect above cloud CapEx, custom +≥20%, storage/switch/other +≥15%), comms/other +10%, implying ~$10B total revenue (Murphy confirmed Seymore's math). FY2028 sketched at ~40% data center growth with custom doubling off FY2027.
- Q4 FY2026 guide: $2.2B ±5% (+6% QoQ, +21% YoY; ~+24% ex-auto), non-GAAP EPS $0.74–$0.84, non-GAAP GM 58.5–59.5% (down from Q3's 59.7% — notable), non-GAAP OpEx ~$515M (up on bonus accruals).
What management is focused on- The Celestial AI thesis: copper scale-up interconnects are at fundamental reach/bandwidth limits; Celestial's PF chiplet delivers 16Tbps per chiplet (10x a 1.6T port), >2x copper power efficiency, nanosecond-class latency, and thermal stability enabling vertical 3D co-packaging with XPUs — freeing die-edge "beachfront" for more HBM. Murphy framed it as "Inphi-like": internal silicon photonics strength plus a lead Tier-1 hyperscaler pulling the first wave. TAM framing: merchant scale-up switch ~$6B by 2030, optics content of similar magnitude, "over $10B" combined.
- Rebuilding custom credibility with backlog language: FY2027 custom +≥20%, "no air pockets," second-half-weighted, and — the key new disclosure — purchase orders in hand for the entirety of next fiscal year's forecast for the lead customer's next-generation XPU program, with the program forecast "consistent with prior expectations." This directly answers last quarter's socket-loss noise without naming the socket.
- Multi-year guidance as deliberate strategy: Murphy explicitly acknowledged the departure from quarter-at-a-time guidance, citing multiyear AI build cycles, customer capacity planning 6–8 quarters out, and investor demand for a bridge to the FY2029 targets. He labeled the FY2027/FY2028 numbers "base case, not dream the dream."
- XPU-attach as the next quantified leg: 15+ attach wins; NIC and CXL use cases alone give "line of sight to revenue exceeding $2B by fiscal 2029." CXL: 5 sockets at 2 Tier-1 US hyperscalers, engaged with a third; first socket shipping since Q1, second in production a year out, rest in CY2027.
- Capital deployment: $1B ASR executed plus $300M regular buyback in Q3 ($1.35B total capital return); acquisition funded without debt; buybacks/dividends to continue in parallel.
Key numbers and quarter mechanics- Revenue $2.075B (+37% YoY, +3% QoQ). Data center $1.52B (73% of revenue, +2% QoQ vs. flat guide, +38% YoY). Communications and other $557M (+8% QoQ, +34% YoY as reported; ~+20% QoQ and ~+50% YoY ex-auto Ethernet) — enterprise and carrier recovery on inventory normalization plus refreshed products.
- Non-GAAP: GM 59.7% (+30bps QoQ, within 59.5–60% guide); OpEx $485M (in line with ~$485M guide); operating margin 36.3% (+150bps QoQ); EPS $0.76. GAAP: GM 51.6% (guide 51.5–52%); OpEx $712M (vs. ~$719M guide); operating margin 17.2%; EPS $2.20 including the divestiture gain.
- Cash flow/balance sheet: record CFO $582M (+$121M QoQ, second consecutive strong quarter). Inventory $1.01B, -$37M QoQ (second consecutive decline). Cash $2.7B (+$1.5B QoQ on $2.5B divestiture proceeds less $1.35B capital return). Debt $4.5B; leverage 1.47x gross / 0.58x net (from 1.63x/1.19x). Dividends $51M.
- Q4 FY2026 guide: revenue $2.2B ±5%; GAAP GM 51.1–52.1%; non-GAAP GM 58.5–59.5% (a sequential step-down worth probing — mix, custom ramp, or conservatism not explained); GAAP OpEx ~$741M; non-GAAP OpEx ~$515M (up ~$30M on investment and bonus payouts); other expense ~$30M; non-GAAP tax 10%; basic 850M / diluted 857M shares; GAAP EPS $0.31–$0.41; non-GAAP EPS $0.74–$0.84.
- Q4 end-market color: data center high-single-digit sequential growth (~+20% YoY) on custom rebound plus interconnect/switch/storage growth; comms/other low-single-digit sequential growth (~+25% YoY reported, ~+40% ex-auto) — carrier strong, enterprise growing, consumer steep seasonal decline. Enterprise networking to hit ~$1B annualized run rate in Q4 = "complete normalization" of inventory; thereafter grows with enterprise IT spend. Carrier nearly doubles YoY in Q4, still recovering toward carrier-CapEx-linked growth.
- FY2027 framework (organic only, excludes Celestial): data center +>25%; interconnect (~half of data center) growing faster than cloud CapEx (now expected +30%+); custom (~quarter of data center) +≥20%, H2-weighted, no air pockets; storage/switch/other +≥15% (up from +10% prior); comms/other +10%. Non-GAAP OpEx to grow ~half the revenue rate; non-GAAP tax rate moves to ~12%. Sequential revenue growth every quarter of FY2027 with a "really compelling exit rate."
- FY2028 sketch (Seymore Q&A): custom doubling off FY2027 (XPU attach + a new meaningful XPU socket ramping); interconnect above assumed 20% CapEx growth; storage/switch/other +10%; data center ~+40%; total Marvell ~+30%.
Product and launch scorecard- Celestial AI (new): PF chiplet, 16Tbps, first-gen product; one major design win at a Tier-1 hyperscaler (AWS strongly implied via the 8-K warrant extension — Murphy confirmed the warrant adds a "photonic fabric" swim lane to the year-old AWS agreement and noted AWS's supportive quote in the deal release). Co-packaged into both custom XPUs and scale-up switches; claimed as the industry's first large-scale commercial optical scale-up deployment. Revenue: meaningful contribution H2 FY2028, $500M run rate Q4 FY2028, $1B Q4 FY2029; accretive to non-GAAP EPS from H2 FY2028. Future applications: pooled memory appliance, optical die-to-die. Execution risk is real: pre-revenue at Marvell, regulatory close pending, $2B earnout sets a high bar.
- Custom XPU: lead-customer next-gen transition fully baked into FY2027 with POs for the entire year — the strongest reassurance since the no-comment posture began. New wins since June: additional sockets representing >10% of the $75B funnel, including XPU-attach at an emerging hyperscaler and an electrical I/O chiplet inside an XPU (new multi-die packaging entry point). FY2028 custom doubling predicated on attach ramps plus "a new meaningful XPU socket" — the second big XPU customer contributes little in FY2027 (Arya Q&A).
- XPU attach: 15+ wins. NICs: multiple hyperscaler wins, attach expanding to broader AI server fleets (1M+ units/year at large hyperscalers). CXL: 5 sockets at 2 Tier-1 hyperscalers, first shipping since Q1 and entering volume production now. NIC + CXL line of sight >$2B by FY2029.
- Optics: double-digit sequential growth delivered as guided. 1.6T shipping since H2 this fiscal year with "exceptionally strong demand heading into next year"; 3nm 1.6T sampled Feb 2025; 3.2T production deployments expected CY2028 on 2nm. Coherent Lite 1.6T ships next year, 3.2T the year after. TIAs claimed to have significant 1.6T performance lead. LPO: several sockets across multiple hyperscalers, still small.
- AEC/retimers: AEC design wins with significant share at 2 Tier-1 US hyperscalers plus emerging hyperscalers; PCIe Gen6 retimers engaged with 30+ customers, 10+ sockets designed in, production H2 next calendar year, full revenue FY2028. Combined AEC+retimer revenue to more than double YoY next year.
- Switching: FY2026 switch revenue >$300M; 51.2T now shipping with strong ramp next year; FY2027 switch revenue to surpass $500M — "faster than what I had indicated last quarter" (an explicit upward revision). 100T scale-out products introduced next year.
- Scale-up switching: UALink 115T and 57T to sample H2 FY2027, volume production FY2028 — firmer than last quarter's "next 2 years." Parallel ESUN collaboration. Up to 576 ports, reticle-size dies.
- Storage: double-digit sequential growth in Q3; part of the +≥15% FY2027 bucket.
Sell-side read-through- Only eight questions on a heavily pre-announced call; tone was congratulatory, and the prior quarter's pressure points largely dissolved. Sur — last quarter's antagonist on the lead-customer 3nm socket — opened with "Congrats" and got a constructive answer: POs for the full FY2027 next-gen program, 2nm programs ramping in FY2028, "design funnel keeps increasing." He closed with "Great execution." The socket-loss controversy was neutralized via backlog disclosure rather than program detail.
- Seymore's math (~$10B FY2027) was confirmed on the record — the street now has a management-endorsed FY2027 revenue anchor and a FY2028 ~30% total-company growth sketch. This is a material change in modeling visibility.
- Kumar pressed the right question: is custom +20% a floor ("in the bag") and what is the comfort level on multi-year forecasts? Murphy: model 20%, it's a "good safe base case," upside would be updated later; multi-year disclosure justified by customer planning horizons. Notably he referenced "history on this custom business where people got ahead of themselves" — an implicit nod to past expectation resets.
- Arya probed two soft spots: why index optics to CapEx rather than accelerator units (Murphy conceded optics is AI-driven and grows above CapEx — CapEx was just a proxy), and whether the custom 20% depends on the unproven second customer (answer: no — FY2027 is mostly the lead-customer transition plus attach; the second XPU customer is a FY2028 story). Both answers were clean.
- Curtis extracted the AWS identification via the 8-K warrant (1M shares, photonic fabric swim lane) — Murphy confirmed the structure and the $2B earnout through FY2029 without explicitly naming AWS as the Celestial lead customer, but left little doubt.
- Rolland's rack-level question (competitor moving to systems/racks) got a clear strategic answer: rack-scale vision, one-stop interconnect shop, but no system-level revenue in any forecast — Marvell stays a silicon/IP supplier.
- Notably absent: no questions on the Q4 non-GAAP GM step-down (58.5–59.5%), the custom Q3 decline itself (accepted as done), GAAP EPS optics, SBC, or the consumer seasonal decline. The AI-mix "majority of company" question from last quarter was not re-asked.
Management credibility- Delivery against last quarter's checkable items: revenue $2.075B vs. $2.06B ±5% (above midpoint); non-GAAP EPS $0.76 vs. $0.69–$0.79 (above midpoint); non-GAAP GM 59.7% within 59.5–60%; non-GAAP OpEx $485M in line; optics double-digit sequential growth delivered as guided; custom declined as guided; data center beat the flat guide (+2%); comms/other strong as guided. This is a clean quarter against every explicit commitment.
- The Q3 "digestion" framing from last quarter is validated so far: custom declined exactly as pre-announced, other lines grew, and Q4 is guided to the promised rebound. The lumpiness narrative survived its first test.
- Credibility credits: the 51.2T switch ramp commitment from last quarter was not just met but raised ($500M+ FY2027, "faster than indicated"); enterprise run-rate milestones (~$1B annualized in Q4) are specific and checkable; the AWS warrant/8-K provides third-party-documented evidence for the Celestial lead-customer claim rather than pure assertion.
- Credibility debits/risks: the FY2027–FY2028 framework is a large new exposure — management has now put ~$10B FY2027, custom doubling in FY2028, and ~40% FY2028 data center growth on the record, all explicitly "base case." If FY2027 custom lands at exactly 20% with H2 weighting, fine; any slip reopens the expectation-management history Murphy himself referenced. The Q4 non-GAAP GM guide (58.5–59.5%) steps down from 59.7% with no explanation offered and no analyst asking. The lead-customer 3nm/2nm discussion remains partially no-comment ("can't go into too much"), though the PO disclosure substitutes for program detail. Celestial revenue targets ($500M/$1B run rates) are management forecasts for a pre-revenue asset — treat as aspiration with an earnout attached, not evidence.
- Consistency note: Murphy's custom growth history ("quadrupled CY23→24, doubled CY24→25, +20% FY2027, doubling FY2028") is a new, checkable framing of the ramp trajectory.
What changed versus the prior quarter- The custom overhang resolved into a framework: last quarter's refusal to discuss sockets and unquantified Q4 recovery is now a full FY2027 custom guide (+≥20%, H2-weighted, no air pockets) backed by disclosed purchase orders for the lead customer's next-gen program. Disclosure shifted from portfolio statistics back toward program-level reassurance — via backlog, not socket commentary.
- Scale-up went from narrative to balance sheet: last quarter's "products in 2 years" talking point is now the Celestial AI acquisition with a named-technology platform, a warrant-linked hyperscaler, revenue run-rate targets, and a $2B earnout. UALink switch timing also firmed to H2 FY2027 sampling / FY2028 production.
- Guidance horizon extended dramatically: from quarter-at-a-time (and explicit deferral of annual views "later in the year") to a full FY2027 framework plus FY2028 sketch. This is the biggest process change in Marvell's communication in the period covered.
- Segment reporting transition executed: Q3 reported in both formats; comms/other is now the go-forward bucket, with enterprise (~$1B Q4 run rate) and carrier (near-doubling YoY) still described but no longer separately reported going forward.
- Capital return stepped up sharply: $1.3B buybacks in Q3 ($1B ASR + $300M) vs. $200M in Q2 — the post-divestiture deployment question from last quarter was answered with the ASR, even as Celestial will consume stock plus cash.
- Switching outlook revised up: FY2027 switch revenue >$500M, explicitly faster than last quarter's indication.
- Gross margin trajectory wobbled: Q3 printed 59.7% (up 30bps) but Q4 guides down to 58.5–59.5% — a new, unexplained wrinkle after two quarters of margin-at-low-end concerns.
- Leverage and cash transformed: net leverage 0.58x (from 1.19x), cash $2.7B — the balance sheet is now a strategic weapon, and management used it.
Bull case- The custom bear case lost its legs: POs for the entire FY2027 next-gen program at the lead customer, +≥20% custom growth with no air pockets, new sockets >10% of the $75B funnel since June, and a FY2028 doubling path via attach plus a second XPU customer. The dual-sourcing/socket-loss narrative now has to fight disclosed backlog.
- Celestial AI is a credible scale-up entry: differentiated technology (16Tbps chiplet, thermal stability enabling 3D co-packaging, HBM beachfront argument), a warrant-committed Tier-1 hyperscaler, Marvell's Inphi-proven silicon photonics bench, and a $10B+ combined scale-up switch/optics TAM. If the $500M→$1B run-rate path holds, this is the next Inphi.
- Growth is broad-based and accelerating: optics double-digit growth with 1.6T demand "exceptionally strong," switching revised up to >$500M FY2027, AEC+retimer doubling, storage recovering, enterprise at a normalized $1B run rate, carrier still recovering. FY2027 ~$10B and FY2028 ~+30% total growth are now management-anchored.
- Operating leverage compounding: non-GAAP EPS +77% YoY on +37% revenue; FY2027 OpEx guided to grow at half the revenue rate; record $582M CFO; 0.58x net leverage with continued buybacks alongside M&A.
- XPU-attach is a second custom engine: NIC + CXL alone >$2B by FY2029, with attach rates exceeding initial expectations and the electrical I/O chiplet opening sockets previously unavailable.
Bear case- The FY2027/FY2028 framework is a large, self-imposed hurdle. Custom +20% is H2-weighted and depends on a product transition at one customer; FY2028's doubling depends on a second XPU customer with no history of ramping big ASIC programs (Arya's point, unanswered on the merits) plus attach ramps that are early. Murphy's own caveat about custom "history" cuts both ways.
- Q4 non-GAAP GM guided down to 58.5–59.5% from 59.7% — unexplained, unasked, and consistent with the structural custom-dilution concern; the ~60% norm is receding.
- Celestial is pre-revenue with regulatory close risk, a $2B earnout, and revenue not meaningful until H2 FY2028 — three years of execution between here and the $1B run rate, in a scale-up market where the incumbent GPU player's proprietary fabric dominates and standards (UALink vs. ESUN vs. proprietary) are unsettled.
- Custom concentration remains: FY2027 growth is "mostly our current business today" — the lead customer transition. The second XPU customer contributes little until FY2028. One customer's build timing still swings the largest product line.
- Q4 data center YoY decelerates to ~20% (from 38%) even with the custom rebound — the easy comps are over, and the FY2027 +25% requires the H2 custom weighting to materialize on schedule.
- Visibility is structurally declining: comms/other detail disappears into one segment just as enterprise/carrier recover; AI-mix quantification was never refreshed; and the consumer business goes "steep seasonal decline" in Q4 with no updated annual frame.
- Tax rate rises to ~12% in FY2027 and OpEx steps up ~$30M in Q4 — modest, but both trim EPS leverage at the margin.
Next-quarter watchlist- Q4 delivery vs. $2.2B ±5%, non-GAAP EPS $0.74–$0.84, and especially non-GAAP GM 58.5–59.5% — whether the margin step-down is mix (custom ramp), conservatism, or something structural; management should be pressed since no analyst asked.
- The custom rebound: Q4 data center guided high-single-digits sequential with custom rebounding — the final proof point of the H2-over-H1 and "no air pockets" claims. Any wobble invalidates the digestion framing retroactively.
- Celestial AI close: regulatory progress toward the Q1 FY2027 close; any customer, ecosystem, or competitive response (UALink/ESUN positioning, incumbent fabric moves); whether the AWS relationship expands publicly.
- FY2027 guide formalization: how the ~$10B framework converts into an official annual outlook at the Q4 call; whether custom +20% gets raised, held, or hedged; Q1 FY2027 sequential growth (Murphy promised growth every quarter).
- Custom H2 weighting: evidence through the year that the second-half ramp is building — lead-customer next-gen production milestones and any 2nm program disclosure (Sur's sub-3nm pipeline question got only "heads-down execution").
- Switch revenue trajectory toward >$500M FY2027 and 51.2T ramp quantification; UALink 115T/57T sampling on schedule in H2 FY2027.
- AEC/retimer revenue doubling: first revenue evidence and confirmation of the H2 CY2026 retimer production start.
- Enterprise ~$1B annualized run rate in Q4 and carrier's near-doubling — and how these remain trackable inside the consolidated comms/other segment.
- Capital allocation: buyback cadence post-ASR with Celestial funding pending; any authorization expansion; cash vs. stock mix at close.
- XPU-attach milestones: CXL socket #2 entering production (~a year out), NIC attach-rate data, and whether the ">$2B by FY2029" line of sight gets updated.
|
| Aug 28, 2025 | -18.59% | Q2 FY2026 | Read transcript briefingQuarter in one view- Q2 FY2026 revenue $2.006B, a record, +6% QoQ and +58% YoY — at the midpoint of the $2.0B ±5% guide. Non-GAAP EPS $0.67 (guide $0.62–$0.72, above midpoint), +123% YoY. GAAP EPS $0.22 (guide $0.16–$0.26), GAAP operating margin 14.5% — third consecutive GAAP-profitable quarter.
- The quarter's defining disclosure: a guided Q3 custom revenue decline. Data center guided flat sequentially in Q3 — double-digit optics growth offset by lower custom revenue — with Q4 "substantially stronger" and custom up H2 over H1. This is the first guided sequential custom decline of the AI ramp, framed as one-quarter "digestion" on existing programs.
- The June custom AI event recap entered the financial narrative: $94B CY2028 data center TAM (+26% vs. prior), 18 multigenerational XPU/XPU-attach sockets, 50+ pipeline opportunities worth ~$75B lifetime revenue, and a share target of 20% of $94B by CY2028 (from 13% of $33B in CY2024). New wins since June described as "multibillion-dollar lifetime revenue potential" — "18 plus."
- Auto Ethernet divestiture closed early August (ahead of the year-end expectation), $2.5B all-cash; Q3 GAAP EPS guide of $1.98–$2.08 is dominated by the ~$1.8B divestiture gain. Reporting consolidates to two segments (data center; communications and other) starting with the Q4 guide.
- Q3 FY2026 guide: $2.06B ±5% (+36% YoY; ~+40% ex-auto Ethernet), non-GAAP EPS $0.69–$0.79 (+10% sequential at midpoint), non-GAAP GM 59.5–60%.
What management is focused on- Controlling the custom-lumpiness narrative: Murphy repeatedly framed the Q3 custom decline as normal hyperscale-build timing on existing programs — "a 1 quarter digestion with the recovery in Q4" — and pre-committed to custom up H2 over H1 and a "strong fourth quarter." This was the call's central defense.
- Pivoting from single-socket defense to portfolio breadth: when Sur pressed directly on the lead customer's 3nm follow-on and the Asia competitor "claiming victory," Murphy refused socket-level comment for the first time — "commenting on just the individual sockets at this point is only probably increasing the noise level" — and redirected to the 18-plus socket count and 20% share target. A deliberate disclosure strategy change.
- Scale-up as the next leg: Ethernet and UALink scale-up switches positioned as a major XPU-attach category, leveraging Innovium IP and SerDes; product introductions "in the next 2 years" (Bharathi). AEC/AOC DSPs already in market, retimers in customer evaluation, 6.4T silicon photonics light engines demonstrated for NPO/CPO.
- Organizational consolidation: Koopmans promoted to President/COO (sales, non-data center, corporate development); Bharathi to President, Data Center Group with full lifecycle ownership. Framed as execution scaling; also puts new voices on the call.
- Capital allocation post-divestiture: $2.5B proceeds for buybacks plus opportunistic tuck-ins; Murphy disclosed the 10th annual strategic review just completed and R&D now "well north of 80%" in AI/data center (from ~60% a few years ago).
Key numbers and quarter mechanics- Revenue $2.006B (+58% YoY, +6% QoQ). Mix: data center $1.49B (74% of revenue, +3% QoQ, +69% YoY — vs. "slightly faster than Q1's +5%" signal, slightly below); enterprise $194M + carrier $130M = $324M (+2% QoQ vs. mid-single-digit guide — slightly below; +43% YoY); consumer $116M (+84% QoQ vs. +~50% guide — big beat; +30% YoY); auto/industrial $76M (flat QoQ vs. flat guide — in line).
- Non-GAAP: GM 59.4% (vs. 59–60% guide — low end; down from 59.8%); OpEx $493M (vs. ~$495M guide — slightly below); operating margin 34.8% (+870bps YoY); EPS $0.67. GAAP: GM 50.4% (vs. 50–51% guide); OpEx $721M (vs. ~$735M guide — below); operating margin 14.5%; EPS $0.22.
- Cash flow/balance sheet: CFO $462M, +$129M QoQ — a direct answer to last quarter's unexplained drop to $333M. Inventory $1.05B, -$20M QoQ (first decline in several quarters). Cash $1.2B; debt $4.5B after a $1B notes offering in June (mostly refinanced existing debt); leverage improved to 1.63x gross / 1.19x net (from 1.8x/1.42x). Dividends $52M; Q2 buyback $200M (down from $340M in Q1); $540M repurchased in H1; ~$2B authorization remaining.
- Q3 FY2026 guide: revenue $2.06B ±5% (includes mid-single-digit millions of auto Ethernet; ~$60M more if held for the full quarter); GAAP GM 51.5–52%; non-GAAP GM 59.5–60%; GAAP OpEx ~$719M; non-GAAP OpEx ~$485M; GAAP other income ~+$1.8B (divestiture gain); non-GAAP other expense ~$33M; non-GAAP tax 10%; basic 863M / diluted 870M shares; GAAP EPS $1.98–$2.08; non-GAAP EPS $0.69–$0.79.
- Q3 end-market guides: data center flat sequentially (optics double-digit growth offset by lower custom), +mid-30s% YoY; enterprise+carrier +~30% sequentially (implies ~$1.7B annualized run rate vs. ~$900M trough in Q1 FY2025; $2B annual ambition reiterated); consumer down low single digits; auto/industrial ~$35M including a mid-single-digit-million auto Ethernet stub.
- Segment reporting change: Q3 is the last quarter of the current five-segment view; Q4 onward reports data center and "communications and other." Steady-state targets given: enterprise+carrier ~$2B/year, consumer ~$300M/year, industrial ~$100M/year post-divestiture.
Product and launch scorecard- Custom XPU/attach portfolio: 18 sockets, "several" in volume production since late last year, all 18 expected in production between now and the next ~18–24 months; new wins since June add "multibillion-dollar" lifetime revenue ("18 plus"). XPU-attach deal sizes have grown from "several hundred million" lifetime to "much, much larger" (Koopmans). However, the lead-customer 3nm follow-on — confirmed last quarter with capacity secured and CY2026 production — received no update; Murphy declined socket-level comment under direct pressure. The third-customer 3nm program (H2 CY2026 ramp) also went unaddressed.
- Optics: 800G demand strong "with a long life cycle still ahead"; 1.6T 200G/lane PAM now in volume shipments to multiple customers with adoption accelerating "over the next several quarters" — an upgrade from last quarter's "shipping" language. 400G/lane PAM demo reiterated as the 3.2T path. Q3 optics guided double-digit sequential growth — the first explicit optics acceleration after two quarters of implied flatness. LPO acknowledged as real but niche: Marvell has "active wins" in LPO modules, yet pluggables remain "the vast majority... for the foreseeable future."
- Scale-up switching: new product category emphasis — Ethernet and UALink scale-up switches in development, product introductions within 2 years, positioned as a key XPU-attach driver. No revenue yet; timing deliberately vague (Bolton's CY2026 vs. CY2027 question got "next 2 years").
- Switching (scale-out): 12.8T shipping in high volume; 51.2T now ramping and expected to be "a major driver of switch revenue growth in the next fiscal year" — new, checkable commitment.
- Storage: "improved significantly" with SSD and HDD markets returning to health — new positive color after a long downturn.
- Security: expanded Microsoft Azure HSM collaboration — named-customer datapoint, small but concrete.
- DCI: adoption "proliferating across large hyperscalers" — qualitative only.
- Not updated: NVLink Fusion (no mention this quarter), Vault multi-die packaging, custom HBM, PCIe retimer sizing under the AWS agreement (retimers only "in customer evaluation"), carrier L2 socket.
Sell-side read-through- The custom Q3 decline drew the most questions (Seymore, Rakers, O'Malley, Arya). Management held a consistent line: timing/digestion on existing programs, not a product transition or lost business (O'Malley asked explicitly whether a program was winding down — answer: no, timing only). The street's key modeling takeaway: Q4 must deliver a "substantially stronger" custom quarter for the H2-over-H1 claim to hold.
- Sur's question was the call's pressure point: Asia supply-chain noise on the lead customer's 3nm follow-on, competitor "claiming victory," and a direct ask on both the follow-on and the third-customer 3nm program. Murphy's refusal to comment on individual sockets — after last quarter's detailed node/capacity/timing disclosure — is a disclosure retreat the street will read as either discipline or reduced visibility. It was the last question taken, and the call ended shortly after.
- Arya's CY2026 growth question (competitor suggesting 60% custom growth; NVIDIA ~50% industry framing) got no annual commitment — Murphy reiterated quarter-at-a-time guidance and deferred any annual view to "later in the year." No pushback on industry growth rates, but no endorsement either.
- Arcuri's AI-mix question exposed a soft spot: last quarter's "AI majority of whole company" trajectory could not be confirmed for Q3 or Q4 — Murphy: "I'd have to give you a follow-up... I don't have the spreadsheet right in front of me." The Q4 mix frame (optics ~half, custom ~quarter, other ~25% of data center) was restated as the last update, with both optics and custom "come up since then."
- Kumar's "are all your wins on track" question — citing investor "controversy" — got a portfolio-level answer (design momentum, AI Day framing) rather than a program-level affirmation. Notably, no one received a direct "yes, all on track."
- Schneider's capital-allocation question confirmed openness to further divestitures "at the right price" was not directly answered for carrier/consumer — Murphy described the framework, not future sales; proceeds split between buybacks and tuck-ins is undecided.
- Notably absent: no questions on the consumer beat's sustainability, the $1B debt raise, SBC, or the segment-reporting change's optics (it will obscure enterprise/carrier/consumer detail just as they recover).
Management credibility- Delivery against last quarter's checkable items: revenue $2.006B vs. $2.0B guide (midpoint); non-GAAP EPS $0.67 above midpoint; non-GAAP OpEx $493M vs. ~$495M; GAAP OpEx $721M vs. ~$735M; consumer +84% vs. +~50% guided (well above); auto/industrial flat as guided; CFO rebounded to $462M as the bear-case watchlist item required; inventory declined. Debits: data center +3% vs. Murphy's "slightly faster than Q1's +5%" signal; enterprise+carrier +2% vs. mid-single-digit guide; non-GAAP GM 59.4% at the low end of 59–60%.
- Multi-quarter promises kept: the auto divestiture closed ahead of schedule (early August vs. "within calendar 2025"); the June 17 event happened and delivered the TAM/share/pipeline disclosures deferred to it last quarter; 1.6T moved to volume shipments at multiple customers; enterprise/carrier recovery now guided to a ~$1.7B run rate, validating the repeated recovery narrative.
- Credibility debits: the lead-customer 3nm program — last quarter's centerpiece disclosure — went from specific (node, capacity, CY2026) to no-comment in one quarter, precisely when Asia noise is loudest. The "AI majority of whole company" claim from last quarter could not be mileposted when asked directly. The Q3 custom decline was not pre-signaled in the Q1 guide (data center was guided mid-single-digits up; actual +3%, then flat guided) — the lumpiness framing is new. GAAP EPS guide of $1.98–$2.08 requires readers to strip a $1.8B gain; mechanically fine but headline-misleading.
- New specificity where management chose: 18-plus sockets, multibillion-dollar new wins, $75B pipeline, 20%-of-$94B share target, ~$500M on-prem run rate, enterprise+carrier ~$1.7B Q3 run rate, 51.2T as next-year switch driver, scale-up products within 2 years.
- Koopmans and Bharathi took real Q&A load for the first time and were consistent with Murphy's messaging — a deliberate show of bench depth, though Bharathi's "next 2 years" scale-up timing was vaguer than Bolton's question invited.
What changed versus the prior quarter- Custom went from steady growth to a guided Q3 decline: last quarter's "custom growing quite strongly" framing is now "nonlinear," with Q3 down and Q4 "substantially stronger." The lumpiness caveat is new language for a business previously presented as a smooth ramp.
- Lead-customer disclosure reversed direction: from node/capacity/timing specificity (Q1) to an explicit refusal to discuss individual sockets (Q2) — while simultaneously expanding the portfolio frame to "18 plus" sockets. The information shifted from program depth to portfolio breadth.
- Optics outlook inflected positively: from implied flat-to-modest near-term (Q1 read-through) to guided double-digit sequential growth in Q3, with 1.6T now in volume shipments to multiple customers.
- Auto Ethernet closed (early August) rather than pending; $2.5B cash in hand; segment reporting collapses to two categories starting with the Q4 guide — materially reducing non-data-center visibility going forward.
- Cash flow repaired: CFO $462M vs. $333M; inventory inflected down $20M; leverage down to 1.19x net. Last quarter's unexplained CFO drop resolved without commentary on the cause.
- Buyback moderated to $200M from $340M despite the incoming $2.5B — Meintjes framed proceeds as enabling "opportunistic" buybacks plus tuck-ins; no step-up committed.
- Scale-up switching elevated from a passing mention to a headline investment area (Ethernet + UALink, Innovium IP, 2-year product window) — the clearest new TAM narrative since NVLink Fusion, which itself went unmentioned this quarter.
- Enterprise+carrier guided +~30% sequentially — a step-change from five quarters of mid-single-digit recovery, attributed to inventory normalization plus refreshed advanced-node products.
- AI-mix quantification regressed: the Q4 frame (optics ~half/custom ~quarter) was restated as stale, and the "majority of whole company" timing went unconfirmed.
Bull case- The custom portfolio is now quantified at a different scale: 18-plus multigenerational sockets, 50+ pipeline opportunities worth ~$75B lifetime, new wins since June alone in the billions, and a 20%-of-$94B CY2028 share target — the Q3 custom dip is framed, plausibly, as build-timing digestion within a broadening ramp, with H2 guided above H1 and Q4 "substantially stronger."
- Optics is re-accelerating now, not next year: double-digit Q3 sequential growth guided, 1.6T in volume at multiple customers, 800G longevity affirmed, and LPO dismissed as niche with Marvell participating anyway. The optics-flattish bear math from last quarter is broken.
- The non-AI recovery arrived with force: enterprise+carrier guided +~30% sequentially to a ~$1.7B run rate (vs. $900M trough), with a reiterated path to ~$2B — plus storage recovering and 51.2T switching ramping into next fiscal year. Growth is no longer single-engine.
- Cash and capital: $462M CFO, $2.5B divestiture proceeds, 1.19x net leverage, ~$2B buyback authorization — capacity for both aggressive repurchase and AI tuck-ins at "this historic moment."
- Operating leverage compounding: non-GAAP EPS +123% YoY on +58% revenue; Q3 guided EPS +10% sequentially on ~+2.7% revenue; GM guided back up to 59.5–60%.
- Scale-up optionality: Ethernet/UALink switches, AEC/AOC, retimers, and 6.4T silicon photonics position Marvell for the copper-to-optics scale-up transition regardless of which fabric wins.
Bear case- The lead-customer question is now a no-comment. After last quarter's detailed 3nm reassurance, Murphy declined to confirm the follow-on program is tracking, under direct questioning about a competitor "claiming victory." Combined with the Q3 custom decline, the street's dual-sourcing/socket-share concern has more room to run, not less — and management chose portfolio statistics over program reassurance.
- Custom revenue is declining in Q3 with the recovery entirely back-loaded into Q4 — a single-quarter promise that now carries the full weight of the H2-over-H1 claim. If Q4 disappoints, both the lumpiness framing and the FY2027 growth commitment take damage.
- AI-mix transparency is degrading: the optics/custom/other split hasn't been updated since Q4, the "AI majority of company" milestone couldn't be confirmed, and the segment consolidation will further reduce visibility into the non-data-center businesses just as they recover.
- Gross margin printed at the low end (59.4%) and the Q3 guide (59.5–60%) still caps below the historical ~60%+ norm; custom dilution persists and the consumer beat (below-average margin) added mix pressure.
- Data center +3% QoQ came in below Murphy's own "slightly faster than Q1" signal, and Q3's flat guide means two consecutive quarters of no sequential data center growth before the promised Q4 step-up — YoY decelerates to mid-30s.
- Buyback fell to $200M in Q2 despite the CFO rebound and pending $2.5B — capital-return cadence is less aggressive than the "flexibility" language implies, and uses of proceeds are explicitly undecided.
- Scale-up switching, the call's biggest new TAM claim, has no products before "the next 2 years" and no revenue timeline — narrative ahead of evidence, in a market where the lead GPU player's proprietary fabric dominates today.
Next-quarter watchlist- Q4 custom recovery: the single most important checkable — Q4 must be "substantially stronger" than Q3 for custom to grow H2 over H1. Any wobble invalidates the digestion framing and reopens the socket-loss debate.
- Lead-customer 3nm program: whether management restores any program-level confirmation (qualification milestones, CY2026 production) or maintains the no-comment posture; any Asia supply-chain or competitor announcements forcing disclosure.
- Q3 delivery vs. $2.06B ±5%, 59.5–60% non-GAAP GM, ~$485M non-GAAP OpEx, $0.69–$0.79 non-GAAP EPS — and whether optics delivers the guided double-digit sequential growth.
- AI revenue mix: whether the "majority of whole company" crossover is confirmed and dated (Arcuri's unanswered milepost), and whether the optics/custom/other split gets refreshed.
- Segment transition: how the Q4 guide presents "communications and other," and whether enterprise+carrier's ~$1.7B run rate and path to $2B remain trackable under the new reporting.
- Capital deployment: size of Q3/Q4 buybacks now that $2.5B is in hand; any tuck-in M&A announcements; whether the ~$2B authorization is expanded.
- Scale-up switching: first disclosed design engagements or product timelines firmer than "next 2 years"; UALink ecosystem milestones; AEC/retimer revenue evidence.
- 51.2T switching ramp and storage recovery: quantification as these become next-fiscal-year growth drivers.
- Consumer: whether the +84% beat holds at the new ~$300M annual run-rate frame after the guided low-single-digit Q3 decline.
- New wins conversion: whether "18 plus" gets a number, and whether any of the 50-opportunity/$75B pipeline converts to disclosed sockets at the next update.
|
| May 29, 2025 | -5.55% | Q1 FY2026 | Read transcript briefingQuarter in one view- Q1 FY2026 revenue $1.895B, above the midpoint of the $1.875B ±5% guide, +4% QoQ and +63% YoY — a record. Non-GAAP EPS $0.62 (guide $0.56–$0.66, above midpoint), +158% YoY, again ~2x revenue growth. GAAP EPS $0.20 (guide $0.14–$0.24), GAAP operating margin 14.3% — second consecutive GAAP-profitable quarter.
- The quarter's defining disclosure: Murphy confirmed the follow-on XPU at the lead customer is 3nm, with wafer and advanced packaging capacity secured and production expected to start calendar 2026; the architecture team is already working on the generation after that. He also conceded — for the first time — that given XPU volumes, it is "certainly possible and likely" the customer is pursuing multiple paths. The dual-sourcing question moved from "declined to comment" to an implicit acknowledgment, paired with a revenue-growth commitment through FY2027 and beyond.
- Portfolio reshaping: sale of the automotive Ethernet business to Infineon announced — all-cash, $2.5B, expected to close within calendar 2025. Buyback stepped up to $340M from $200M.
- Q2 FY2026 guide: $2.0B ±5% (+57% YoY, ~+5.5% QoQ at midpoint), non-GAAP EPS $0.62–$0.72, GAAP EPS $0.16–$0.26. Data center guided mid-single-digits sequential again; enterprise+carrier mid-single-digits (fifth straight sequential growth quarter); consumer +~50% (seasonal snapback); auto/industrial flat.
- New platform announcements: NVIDIA NVLink Fusion partnership (chiplet-based scale-up integration for custom XPUs) and a multi-die packaging platform ("Vault" interposer alternative) already qualified and in production supporting a customer-specific XPU.
What management is focused on- Neutralizing the Asia supply-chain "socket loss" narrative: Murphy's first Q&A answer was a prepared, direct rebuttal — "these Asia supply chain sources have an incomplete view... they simply have no idea what we're doing for our customer" — followed by the 3nm capacity-secured fact pattern. This was the call's clear priority.
- Reframing AI mix: rather than updating the AI percentage of data center, Murphy introduced a new frame — AI is on a path to become the majority of the whole company, "not just the data center end market." Deliberately directional, not quantified.
- Custom platform breadth beyond XPUs: custom HBM architecture, CPO light-engine integration, multi-die packaging, NVLink Fusion — all positioned as a "suite" for rack-level custom infrastructure, with the June 17 custom AI investor event as the venue to detail pipeline and market-share progress.
- Capital allocation: the $2.5B Infineon divestiture plus the $340M buyback step-up signal a shift toward portfolio focus and shareholder returns now that leverage is repaired (1.8x gross / 1.42x net, down from 2.06x/1.58x).
- Setting up June 17: Murphy referenced the event at least four times, deferring multiple answers (customer breadth, business models, design-win pipeline, share goals) to it. It is now the de facto next catalyst.
Key numbers and quarter mechanics- Revenue $1.895B (+63% YoY, +4% QoQ). Mix: data center $1.44B (76% of revenue, +5% QoQ, +76% YoY — vs. mid-single-digit guide, in line); enterprise $178M; carrier $138M (combined +14% QoQ vs. ~+10% guide — beat); consumer $63M (-29% QoQ vs. ~-35% guide — beat); auto/industrial $76M (-12% QoQ vs. down high-single-digits guide — slightly worse; auto grew but industrial lumpiness more than offset).
- Non-GAAP: GM 59.8% (vs. ~60% guide — slightly below; custom mix cited); OpEx $486M (vs. ~$490M guide — below); operating margin 34.2% (from 33.7%); EPS $0.62. GAAP: GM 50.3% (vs. ~50.5% guide); OpEx $682M (vs. ~$712M guide — well below); operating margin 14.3%; EPS $0.20.
- Cash flow/balance sheet: CFO $333M — down sharply from $514M in Q4 despite higher revenue (no explanation offered on the call; not asked about). Inventory $1.07B, +$42M QoQ (much slower build than Q4's +$170M). Cash $886M; debt $4.2B; leverage 1.8x gross / 1.42x net. Dividends $52M; buyback $340M.
- Q2 FY2026 guide: revenue $2.0B ±5%; GAAP GM 50–51%; non-GAAP GM 59–60% (a step down from the ~60% norm — Meintjes attributed H2 GM to custom growth plus the consumer snapback, both below-average margin); GAAP OpEx ~$735M; non-GAAP OpEx ~$495M; other expense ~$49M; non-GAAP tax 10%; basic shares 864M / diluted 874M; GAAP EPS $0.16–$0.26; non-GAAP EPS $0.62–$0.72.
- Q2 end-market guides: data center mid-single-digits sequential (Murphy: Q2 "looking like it's gonna grow slightly faster than Q1"); enterprise+carrier mid-single-digits; consumer +~50%; auto/industrial flat.
- Note: the prior quarter's diluted share count anomaly is resolved — Q2 guide has diluted (874M) above basic (864M), confirming last quarter's 818M figure was a transcription error.
Product and launch scorecard- Lead XPU program: high-volume production, "zero silicon" first-time success, met a steep ramp. Follow-on: 3nm wafer and advanced packaging capacity secured, production start expected CY2026, subject to customer qualification cycles. Generation-after-next already in architecture definition. Revenue growth committed for FY2027 and beyond. This is the most concrete next-gen disclosure to date — node, capacity, and timing window all new.
- Customer C XPU (AI Day 2024 win): joint development "progressing well"; notably, Marvell is already engaged with this customer on the follow-on generation's architecture — an upgrade from last quarter's "milestones completed, on track for CY2026."
- Multi-die packaging platform ("Vault" interposer): announced same day; already qualified and in production for a customer-specific XPU — a rare announce-and-in-production launch, with proliferation to next-gen designs planned.
- NVLink Fusion: partnership announced; initial model is a chiplet solution (Marvell XPU + IO chiplets interfacing to an NVLink chiplet). Murphy reports "a whole bunch of interest from customers" already. No revenue timing given.
- Optics: 1.6T 5nm shipping with "very strong demand"; 3nm 1.6T (Ara) positioned as where 1.6T volume concentrates, with the big ramp "next year" (FY2027); 800G still dominates this year and demand remains strong. 400G/lane PAM demonstrated (industry first) as the path to 3.2T. Murphy claimed maintained "commanding" 800G share and first-to-market at both 1.6T nodes; LPO dismissed as a smaller, scale-up-adjacent market Marvell will participate in.
- AEC: disclosed as "ramping this year" — new revenue data point. Switching: "continued revenue growth" (no quantification this quarter).
- Custom HBM and CPO: strong customer interest claimed for custom HBM; CPO framed as copper-to-optical scale-up transition expanding interconnect TAM. No revenue attached to either yet.
- Not updated: PCIe retimer/cable sizing under the AWS agreement (O'Malley asked directly; Murphy gave "progressing well, no major updates"), carrier L2 socket, storage, enterprise product detail.
Sell-side read-through- The socket question was asked three times in escalating directness (Arya: content and exclusivity; Bolton: dual-sourcing and guarantees; Bacalter: what multiple paths look like). Murphy's answers shifted the disclosure: he acknowledged multiple paths are "certainly possible and likely" given volumes, while insisting Marvell's revenue grows through program transitions regardless. Bolton's follow-up — "do you have guarantees that you don't get blanked?" — got no contractual answer, only execution language. Read-through: the street now treats dual-sourcing as probable; Marvell's defense is revenue continuity, not exclusivity.
- AI quantification was refused again. Arcuri floated $3.5–4B for FY2026 AI and got the new "majority of whole company" trajectory framing instead of a number. Fourth consecutive quarter without a raised formal target.
- Kumar pressed the investor complaint that data center growth (mid-single-digits) feels slow; Murphy's answer leaned on the Q3→Q4 20%+ step-up, slightly-faster Q2, and 70%+ YoY comps — context, not acceleration. He also disclosed on-prem is now "pretty small," down slightly in Q1, "relatively stable" — reducing the mix-drag explanation going forward.
- O'Malley surfaced a real question: if custom is growing "quite strongly" and data center is only mid-single-digits, is optics flattening? Murphy said optics grows "throughout the year" — but the math implies optics is roughly flat-to-modest sequentially near term, with 800G dominance persisting until the 3nm 1.6T ramp next year.
- Gross margin drew two questions (Arcuri, Curtis). Key takeaways: custom runs below corporate average and is dilutive as it grows; Q2's 59–60% guide and H2 "similar range" language from Meintjes signal no near-term GM recovery; offset is operating-margin dollars and EPS leverage ("very operating margin rich").
- Svanberg's SerDes question gave Murphy a venue to assert 200G leadership and the only 400G/lane demo — a direct response to competitive noise.
- Notably absent: no one asked about the CFO decline ($514M → $333M), the Infineon deal's dilution/uses beyond "flexibility," SBC, or the consumer business cap. The auto/industrial miss vs. guide went unchallenged.
Management credibility- Delivery against last quarter's checkable items: revenue above midpoint ($1.895B vs. $1.875B); data center +5% vs. mid-single-digit guide; enterprise+carrier +14% vs. ~+10%; consumer -29% vs. ~-35% (better); non-GAAP OpEx $486M vs. ~$490M; GAAP OpEx $682M vs. ~$712M (much better); GAAP profitability sustained as promised. Debits: non-GAAP GM 59.8% vs. ~60% (marginal miss); auto/industrial -12% vs. down high-single-digits (worse); CFO fell to $333M with no commentary.
- Multi-quarter promises kept: the follow-on XPU disclosure from last quarter ("planning for a production ramp once it completes sampling and qualification") firmed into node + capacity + CY2026 production — Murphy delivered specificity he previously withheld. The 3nm Ara timeline (production H2 CY2025) was reiterated implicitly via "shipping now" 5nm and 3nm ramp next year. Buyback not only held but stepped up.
- New specificity where management chose: 3nm node confirmation, capacity secured, CY2026 production start, generation-after-next engagement, customer C follow-on engagement, NVLink chiplet mechanics, Vault in production, AEC ramping, on-prem now small and stable.
- Credibility debits: the AI revenue target remains formally unraised a fourth quarter despite "very significantly exceed" and now "majority of whole company" language — the adjective escalation without a number is becoming a pattern. The dual-sourcing acknowledgment, while honest, retroactively reframes prior quarters' carefully worded denials-by-omission. CFO decline unaddressed. Multiple substantive questions (AWS attach sizing, H2 guidance, business-model detail) deferred to June 17 — the event now carries a heavy burden of proof.
- Murphy's "as clear as it could be" answer to Arya was clear on Marvell's position but deliberately silent on exclusivity — consistent, disciplined messaging that investors should read as confirmation that exclusivity is not claimable.
What changed versus the prior quarter- The 3nm socket question moved from partial answer to near-full answer: node confirmed, wafer and packaging capacity secured, CY2026 production start, and a third generation in architecture definition. Simultaneously, dual-sourcing moved from "declined to comment" to "certainly possible and likely" — both sides of the risk are now on the record.
- Portfolio action: the automotive Ethernet business is being sold to Infineon for $2.5B cash (close expected within CY2025) — the first major divestiture of the AI-first era and a material capital-allocation event. Auto/industrial will shrink as a reported segment.
- AI framing shifted from "% of data center" to "path to majority of whole company" — a directional upgrade without new quantification; the optics/custom/other mix split from last quarter was not updated.
- Gross margin guidance stepped down: Q2 non-GAAP GM guided 59–60% vs. the prior ~60% norm, with Meintjes explicitly guiding H2 to a "similar range" — custom mix dilution is now a guided, persistent feature rather than a quarterly variable.
- Buyback stepped up 70% to $340M; leverage improved to 1.42x net.
- CFO dropped to $333M from $514M — the weakest in several quarters relative to revenue; unexplained.
- The AWS agreement returned to the narrative (O'Malley's question) but with "no major updates" — attach products (retimers, AEC) still unsized, though AEC is now "ramping."
- NVIDIA moved from competitor-adjacent topic to partner: NVLink Fusion validates custom XPUs as complementary to merchant GPU infrastructure — a strategic framing win Murphy used repeatedly.
- Consumer guided +50% for Q2 (seasonal snapback) but Meintjes flagged it as a gross-margin headwind — new color on consumer's below-average margin.
Bull case- The lead-customer moat is now documented, not just asserted: 3nm capacity secured, CY2026 production, generation-after-next in definition, revenue growth committed through FY2027+ — and customer C is already on its own follow-on architecture, making both ramping programs multigenerational.
- Even under dual-sourcing, Murphy's framing holds: XPU volumes are large enough that multiple paths still leave Marvell's revenue growing. The bear scenario shifted from "socket loss" to "socket share," which the growth commitment explicitly absorbs.
- $2B Q2 guide (+57% YoY) with data center re-accelerating slightly and five straight quarters of enterprise/carrier recovery — the non-AI businesses are now contributing growth, not just stabilizing.
- Platform breadth is converting to production: Vault multi-die packaging already qualified and shipping in an XPU program; NVLink Fusion opens NVIDIA-ecosystem scale-up sockets; custom HBM and CPO extend content per accelerator. Content per XPU generation looks up, not flat.
- Optics setup for FY2027: 800G still dominant and strong, 1.6T 5nm shipping, 3nm Ara positioned as the volume node with >20% power savings and a "much stronger ramp next year," plus 400G/lane leadership toward 3.2T. Share claimed as maintained at every node.
- Operating leverage intact: EPS +158% YoY vs. revenue +63%; non-GAAP operating margin 34.2%; management reiterates custom programs are "very operating margin rich" in dollars even at lower gross margin.
- Balance sheet plus divestiture: $2.5B incoming cash, 1.42x net leverage, $340M buyback — capacity for aggressive capital return or reinvestment.
Bear case- Dual-sourcing is now effectively conceded. "Certainly possible and likely" that the lead customer pursues multiple paths, with no contractual protections disclosed (Bolton's "guarantees" question unanswered). FY2027+ custom revenue depends on Marvell's share of a split socket and on AI capex continuing to grow — Murphy's own caveat.
- Gross margin is drifting down structurally: 59.8% printed, 59–60% guided, H2 "similar range." Custom dilution plus a below-average-margin consumer snapback means no GM recovery in sight; the 38–40% operating margin target must be reached entirely through OpEx leverage.
- Data center sequential growth has been mid-single-digits for two consecutive guided quarters after the 20%+ step-ups; optics appears roughly flat near-term (custom carrying the growth), and the 1.6T/3nm volume ramp is a next-year story. Investors pressing for faster growth got comps and context, not acceleration.
- CFO fell 35% sequentially to $333M while inventory sits at $1.07B and buybacks tripled — cash conversion is funding less of the capital return than the headline suggests. Unexplained on the call.
- AI revenue remains formally unquantified a fourth quarter; if FY2026 AI were clearly tracking to $3.5–4B, declining to say so implies back-half visibility limits or deliberate anchoring ahead of June 17.
- Auto/industrial missed its guide (-12% vs. down high-single-digits), and the segment's strategic future is now the divestiture; industrial lumpiness persists.
- The AWS attach opportunity (retimers, cables, AEC) remains unsized despite competitors "putting up bigger numbers" — breadth narrative still lacks revenue evidence outside XPUs and optics.
Next-quarter watchlist- June 17 custom AI investor event: the make-or-break disclosure venue — design-win pipeline, market-share progress vs. AI Day 2024 goals, customer breadth beyond the top four, business models (including dual-source structures), and whether the FY2026 AI target is finally raised.
- Q2 delivery vs. $2.0B ±5%, 59–60% non-GAAP GM, ~$495M non-GAAP OpEx, $0.62–$0.72 non-GAAP EPS — and whether data center lands above Q1's +5% as Murphy signaled.
- Lead-customer 3nm program: qualification milestones against the CY2026 production start; any competitive announcement or supply-chain reporting that forces further disclosure; content direction on the next gen (Arya's question was never answered).
- Cash flow: whether CFO rebounds from $333M toward the prior $500M+ run rate, and whether inventory ($1.07B) converts as the custom/optics ramp continues.
- Optics trajectory: evidence that optics returns to sequential growth in H2 FY2026; 800G durability vs. the 1.6T transition; Ara 3nm production on schedule in H2 CY2025; AEC ramp quantification.
- NVLink Fusion: first customer engagements converting to programs; whether the chiplet model produces revenue timelines.
- Infineon transaction: closing progress within CY2025, use of proceeds, and how auto/industrial is reported post-divestiture.
- Gross margin: whether 59–60% is a floor or continues drifting with custom mix; any update to the 38–40% operating margin path.
- Enterprise+carrier: sixth consecutive growth quarter and progress toward the prior $2B run-rate ambition; consumer's +50% snapback delivery.
- Customer C: production start still on track for CY2026, and whether the follow-on architecture engagement converts to a disclosed win.
|
| Mar 5, 2025 | -19.81% | Q4 FY2025 | Read transcript briefingQuarter in one view- Q4 FY2025 revenue $1.817B, above the midpoint of the $1.8B ±5% guide, +20% QoQ and +27% YoY — the third consecutive ~20%+ sequential step. Non-GAAP EPS $0.60 (midpoint of $0.54–$0.64), +40% QoQ, again 2x revenue growth. GAAP profitability delivered as guided: GAAP EPS $0.23 vs. $0.11–$0.21 guide (above the top end), GAAP operating margin 12.9%.
- FY2025 totals: revenue $5.767B, data center +88% YoY, record CFO $1.68B, $933M returned via buybacks/dividends. Non-GAAP operating margin expanded >1,000bps through the year (23.3% Q1 → 33.7% Q4).
- The quarter's most important disclosure: Murphy confirmed Marvell is "fully engaged" with the lead XPU customer on the follow-on generation, "planning for a production ramp once it completes sampling and qualification," and expects custom XPU revenue from this customer to grow in FY2026, FY2027 "and beyond." This is the closest thing yet to an answer on the 3nm socket question left open last quarter — but it remains carefully worded around "products we will be building," and Murphy explicitly declined to comment on whether the customer is also working with someone else.
- AI quantification moved again: FY2025 AI ended "substantially above" the $1.5B target; FY2026 will "very significantly exceed" $2.5B — but Murphy refused to put a number on it even when Arcuri floated ~$3.5B. New mix disclosure: AI is now >50% of total company revenue; within data center, ~50% electro-optics, ~25% custom, ~25% everything else.
- Q1 FY2026 guide: $1.875B ±5% (+3% QoQ, >60% YoY at midpoint), non-GAAP EPS $0.56–$0.66, GAAP EPS $0.14–$0.24. The sequential deceleration (from +20% to +3%) is mix-driven: data center guided only mid-single-digits (cloud/AI double-digits offset by on-prem seasonal decline), consumer -35%, auto/industrial down high-single-digits.
What management is focused on- Locking in the multi-year custom narrative: Murphy's prepared remarks and his first two Q&A answers were built around one message — the lead XPU relationship extends through the next generation and revenue grows FY26, FY27 and beyond. He pre-emptively addressed the "competitor taking sockets" debate before being asked in earnest.
- Broadening the custom story beyond the two ramping programs: custom engagements now disclosed with all four major hyperscalers (two in compute, a third coming in CY2026, the fourth being the Meta custom NIC in production); new wins include multiple custom NICs and a follow-on custom CXL memory solution.
- Scale-up connectivity as incremental TAM: CPO (6.4T 3D silicon photonics engine), LPO, AEC, coherent-lite — repeatedly framed as upside, not cannibalization ("that's not... Marvell's DSP revenue goes down... that is a revenue upside"). Murphy was explicit that scale-out CPO hype "came out of the balloon" at OFC and is long-term; the near-term opportunity is in-rack/scale-up.
- Organizational proof of the AI-first pivot: all hyperscale products consolidated under Raghib Hussain (cloud data center group); everything else merged under Chris Koopmans (multi-market group). This operationalizes last quarter's restructuring rationale.
- Operating leverage and margin trajectory: "significant progress towards our long-term non-GAAP operating margin target of 38% to 40%" in FY2026; Q1 OpEx up only ~2% despite normal seasonal payroll/merit increases, credited to custom NRE (contra-OpEx) — a new, concrete leverage mechanism.
- Balance-sheet normalization: Fitch investment-grade upgrade (January), leverage down to 2.06x gross / 1.58x net, cash $948M (+$80M), $200M buyback again.
Key numbers and quarter mechanics- Revenue $1.817B (+27% YoY, +20% QoQ). Mix: data center 75% ($1.37B, +78% YoY, +24% QoQ — vs. low-to-mid-20s guide, in line), enterprise $171M, carrier $106M (combined +18% QoQ vs. mid-teens guide — beat), consumer $89M (-8% QoQ vs. down-mid-teens guide — beat), auto/industrial $86M (+3% QoQ vs. low-to-mid-single-digit guide — in line).
- Non-GAAP: GM 60.1% (vs. ~60% guide); OpEx $479M (vs. ~$480M guide); operating margin 33.7% (from 29.7%); EPS $0.60. GAAP: GM 50.5% (vs. ~50% guide); OpEx $682M (vs. ~$710M guide — below); operating margin 12.9%; EPS $0.23. No new restructuring charges — consistent with "largely behind us."
- Cash flow/balance sheet: CFO $514M (vs. $536M in Q3 — held near the level despite the bigger revenue base); inventory $1.03B, +$170M QoQ (+20%), though flat on a days basis; DSO 51 days, down 9; cash $948M; debt $4.06B; $52M dividends; $200M buyback.
- Q1 FY2026 guide: revenue $1.875B ±5%; GAAP GM ~50.5%; non-GAAP GM ~60%; GAAP OpEx ~$712M; non-GAAP OpEx ~$490M (+2% QoQ, below normal seasonality due to higher NRE); other expense ~$43M; non-GAAP tax 10%; GAAP EPS $0.14–$0.24; non-GAAP EPS $0.56–$0.66.
- Q1 end-market guides: data center mid-single-digits sequential (cloud/AI double-digits — Murphy clarified this means "above 10%," not ~20% — offset by on-prem seasonal decline); enterprise+carrier ~+10%; consumer ~-35% (gaming seasonality; ~$300M/year long-term run rate reiterated); auto/industrial down high-single-digits (industrial lumpiness offsetting auto growth).
- Transcript artifacts to verify against the press release: diluted share count stated as 818M vs. basic 867M (almost certainly a transcription error — diluted should exceed basic); Murphy's "$4.6 million" Q1 FY25 run-rate reference is plainly billions.
Product and launch scorecard- Custom XPU program 1 (lead customer): in high-volume production, "performing extremely well with significant volume production ahead." Follow-on generation: fully engaged, ramp planned after sampling/qualification; revenue expected to grow FY26, FY27 and beyond. Under Sur's direct questioning (training? CY26 ramp? 5nm or 3nm?), Murphy gave only: "very high volume program," "continuation of what we're doing," assume node transitions on every next-gen, and "we'll be ready to ramp when it's time." Node and timing remain undisclosed.
- Custom ARM CPU (program 2): in production; Murphy now expects "expanding adoption in our customers' data centers" — an upgrade in language from prior quarters.
- Customer C (AI Day 2024 XPU win, second US hyperscaler): "tremendous progress," key technical milestones completed, "well on track" for production start in calendar 2026; engagement described as multigenerational with "very significant" incremental revenue over several years. On schedule per the prior commitment.
- Fourth hyperscaler: confirmed as the Meta custom NIC (in production), with the opportunity set "bigger than just that product." Multiple additional custom NIC wins plus a follow-on custom CXL memory solution disclosed.
- Optics: 800G remains "the workhorse" with "very healthy demand" into FY2026; 1.6T 5nm shipping; 3nm "Ara" 1.6T DSP now has a production timeline — second half of this year (CY2025) — new information versus last quarter's announcement-only status. Murphy directly rebutted 800G inventory/pricing "noise": strong orders in H2 FY25, being fulfilled, demand strong this year.
- CPO: 6.4T 3D silicon photonics engine positioned for custom XPU integration; customer evaluations underway; Murphy was unusually candid that wide-scale adoption is multi-year and scale-out CPO is "very long-term." Framed as TAM expansion, not DSP cannibalization.
- 2nm: industry's first 2nm silicon IP demonstrated on TSMC N2 (announced this week) — working silicon, platform-level, no wins attached yet.
- Teralynx switching: revenue growing double-digits sequentially in Q4 — the first time switching contributed a quantified growth data point; still not dollar-sized.
- Not updated: PCIe retimers, the carrier L2 base station socket's revenue contribution, storage targets, enterprise product detail.
Sell-side read-through- The socket question dominated and was partially defused. Seymore opened on reconciling Marvell's confidence vs. competitor claims; Murphy gave the FY26/FY27-and-beyond growth commitment but explicitly declined to comment on whether the customer works with another vendor. Sur then pressed for program specifics (training, CY26, node) and got none. Read-through: Marvell has a follow-on program in development, but dual-sourcing risk is neither confirmed nor denied — the disclosure is structured to be true even if the customer dual-sources.
- AI sizing remains deliberately open-ended. Arcuri directly proposed ~$3.5B for FY2026 ("is that what you mean by very significantly?") and Murphy declined to anchor: "I'm not putting a number on it just yet." The $2.5B target remains formally unraised for a third consecutive quarter despite "very significantly exceed" language.
- Arya's question on why beats look modest versus the largest customer's capex went essentially unanswered — Murphy responded with sequential growth rates (25%, 25%, double-digits) rather than addressing the supply-vs-early-ramp framing. A mild evasion worth noting.
- New mix math from Arya/Arcuri's exchange: AI >50% of total revenue; data center ~50% optics / ~25% custom / ~25% other. This lets the street back into custom at roughly $450M+/quarter exiting FY2025 — consistent with the $1B+ FY2026 custom marker.
- Kumar's stickiness question drew the fullest competitive framing yet: engagements with all four hyperscalers, sticky sockets but "you've got to win them each time," still a two-player viable field, and new custom designs closed as recently as Q4.
- Reitzes got Murphy on record that sequential AI/data-center growth continuing through the year is "a fair assumption," and that the CY2028 $15B data center goal is tracking — share gaining '23→'24→'25, market and Marvell both growing above the required CAGR, TAM "if anything way larger" than the April 2024 view.
- Sur's inventory question (inventory +20% QoQ vs. +3% revenue guide) was answered cleanly: build supports custom ramps and optics growth; flat on a days basis. Reasonable, but inventory is now $1.03B and must convert.
- Notably absent: no analyst challenged the Q1 sequential deceleration directly, no one asked about gross margin trajectory beyond the ~60% guide, no follow-up on SBC reduction commitments, and no one pressed the diluted share count anomaly.
Management credibility- Delivery against last quarter's checkable items was comprehensive: revenue above midpoint; first GAAP profit quarter delivered and above the guide's top end; data center +24% vs. low-to-mid-20s guide; enterprise+carrier +18% vs. mid-teens; consumer -8% vs. down-mid-teens; auto/industrial +3% in line; non-GAAP GM 60.1% vs. ~60%; OpEx $479M vs. ~$480M; GAAP OpEx $682M vs. ~$710M (better); no new restructuring charges — "largely behind us" held.
- Multi-quarter promises kept: optics grew sequentially again (four for four on "every quarter this fiscal year"); custom ramped as forecast; FY2025 AI "substantially above" $1.5B as signaled; CFO stayed above $500M; buyback held at $200M; leverage fell again (2.06x/1.58x from 2.23x/1.76x).
- New specificity where management chose: AI >50% of revenue, the optics/custom/other split, Ara production in H2 CY2025, customer C milestones completed, four-hyperscaler engagement map, NRE as a visible OpEx lever, Fitch upgrade.
- Credibility debits: the FY2026 AI number remains unquantified despite "very significantly exceed" — three quarters of escalating adjectives without a revised target invites skepticism about the anchor. The follow-on XPU disclosure is carefully lawyered ("products that we will be building") and does not address dual-sourcing. The Q1 guide's sharp sequential deceleration (+20% → +3%) was not proactively explained in prepared remarks beyond the on-prem/consumer seasonality mechanics; the cloud/AI "double-digits means above 10%, not 20%" clarification came only under questioning and tempers the growth-rate narrative.
- Murphy's long-range claims ($15B CY2028, 20% share, TAM "way larger") are now on the record with specific tracking language — checkable at the June 10 Investor Day.
What changed versus the prior quarter- The 3nm/next-gen socket question moved from non-answer to partial answer: last quarter Murphy pointed to the AWS agreement's multi-generational scope; this quarter he committed to follow-on XPU engagement, a planned production ramp, and FY26/FY27-and-beyond revenue growth at the lead customer. Still no node, timing, or exclusivity confirmation.
- AI disclosure stepped up again: from "hundreds of millions above $1.5B" to "substantially above" with FY2026 "very significantly" above $2.5B, plus the first hard mix data (AI >50% of revenue; custom ~25% of data center). The formal targets remain unraised.
- GAAP profitability arrived — the milestone guided last quarter — and management now expects it to continue through FY2026.
- Growth rate inflected down sequentially: Q1 guided +3% QoQ after three quarters of ~+19-20%, with data center decelerating to mid-single-digits. The YoY frame (>60%) is the new headline; the sequential steepening arc from last quarter's bull case is over, at least for Q1.
- The AWS agreement was not mentioned once — a notable silence one quarter after it anchored the call. The Meta custom NIC and four-hyperscaler framing took its place as the breadth evidence.
- Organizational restructuring completed (Hussain/Koopmans two-group structure); no financial charges this quarter.
- Optics narrative shifted from defense (tariff/inventory questions last quarter) to offense: Murphy volunteered rebuttals of 800G inventory/pricing noise and dated the 3nm 1.6T production ramp.
- Consumer got a long-term ceiling: ~$300M/year "over the next several years" — formally capping that business's recovery expectations.
- FY2026 tax rate firmed at 10% for Q1 (within last quarter's 10–11% revision).
Bull case- The custom moat thesis got its strongest support yet: follow-on XPU engagement at the lead customer with revenue growth committed through FY2027+, customer C on track for CY2026 production with multigenerational scope, four hyperscaler engagements, multiple custom NIC wins, a follow-on custom CXL win, and new designs closing as recently as Q4 — all against a self-described two-player viable field.
- AI revenue is now the majority of the company (>50%) and still accelerating on a dollar basis: FY2025 "substantially above" $1.5B, FY2026 "very significantly" above $2.5B, with Q1 guided >60% YoY and cloud/AI growing double-digits sequentially even in a seasonally weak quarter.
- Operating leverage keeps compounding with a new mechanism: EPS +40% on revenue +20%; non-GAAP operating margin 33.7% (from 23.3% in Q1 FY25); Q1 OpEx up only ~2% against normal seasonal headwinds because custom NRE now offsets OpEx — evidence the custom model is structurally accretive to margins, not just dilutive to gross margin. Management targets "significant progress" toward 38–40% in FY2026.
- Optics has both near-term health and a dated catalyst: 800G demand "very healthy" with orders fulfilled, 1.6T 5nm shipping, and the 3nm Ara entering production H2 CY2025 with >20% power savings as 1.6T adoption accelerates.
- Scale-up/CPO is pure optionality: incremental TAM, customer evaluations underway on the 6.4T photonics engine, and management explicitly framed it as upside rather than DSP substitution.
- Balance sheet is fully repaired: record $1.68B FY25 CFO, Fitch upgrade, leverage at 1.58x net and falling, $200M/quarter buybacks funded by operations.
- Non-AI recovery continues: enterprise+carrier +18% in Q4 and ~+10% guided for Q1, trending back toward the $2B run rate.
Bear case- Sequential momentum broke: +3% total revenue guided for Q1, with data center decelerating from +24-25% to mid-single-digits and cloud/AI clarified as "above 10%, not 20%." Two consecutive quarters of ~25% data center sequential growth set a bar the Q1 guide does not meet; bears will ask whether the custom ramp is lumpy rather than smooth.
- The follow-on XPU disclosure, while reassuring, is compatible with dual-sourcing. Murphy declined to comment on the customer working with another vendor, gave no node, no timing, and no exclusivity — the single largest FY2026/27 revenue risk remains unresolved by design.
- AI targets remain formally unraised for the third straight quarter despite "very significantly exceed" language. If FY2026 AI were tracking near $3.5B, declining to say so is either conservatism or uncertainty about back-half visibility — the transcript cannot distinguish.
- Gross margin is pinned at ~60% (60.1% printed, ~60% guided) with mix cited as the determinant — no path articulated back toward the historical model, and custom growth keeps pressure on it.
- Inventory rose $170M (+20% QoQ) to $1.03B against a +3% revenue guide. Management's days-based defense is reasonable, but if Q1/Q2 revenue underdelivers, this becomes a write-down or margin story.
- Consumer is now formally capped at ~$300M/year and drops 35% in Q1; enterprise/carrier still ship below end consumption; auto/industrial guided down. The non-data-center 25% of revenue offers no near-term upside surprise.
- The AWS agreement — last quarter's centerpiece — went entirely unmentioned, and Arya's question about why Marvell's beats look modest relative to its largest customer's capex was not directly answered. Concentration risk is unchanged even as the narrative broadens.
Next-quarter watchlist- Q1 delivery vs. $1.875B ±5%, ~60% non-GAAP GM, ~$490M non-GAAP OpEx, $0.56–$0.66 non-GAAP EPS, $0.14–$0.24 GAAP EPS — and whether GAAP profitability holds as promised for FY2026.
- Data center sequential growth: does cloud/AI land meaningfully above 10%, and does data center re-accelerate in Q2 after the mid-single-digit Q1? Murphy's "fair assumption" of continued sequential growth through the year is now checkable quarter by quarter.
- Follow-on XPU at the lead customer: sampling/qualification milestones, ramp timing, node disclosure, and any competitive announcement. Also customer C's path to CY2026 production start and whether the fourth hyperscaler engagement expands beyond the NIC.
- FY2026 AI quantification: whether the $2.5B target is formally raised (Investor Day June 10 is the obvious venue) and whether the optics ~50% / custom ~25% / other ~25% data center mix shifts as custom ramps.
- Inventory conversion: $1.03B must turn into revenue in H1 FY2026; watch days inventory and any commentary on custom/optics supply.
- Ara 3nm 1.6T production in H2 CY2025 — sampling traction, customer commitments, and whether 800G demand holds through the 1.6T transition as Murphy claims.
- NRE/OpEx dynamic: whether the ~2% Q1 OpEx step-up proves the custom NRE leverage claim, and the trajectory toward the 38–40% operating margin target.
- Enterprise+carrier ~+10% Q1 delivery and progress toward the $2B run rate; whether "shipping below consumption" language lifts.
- Capital returns and SBC: buyback cadence vs. authorization, and any quantified evidence of SBC declining as a percentage of revenue (promised two quarters ago, still unmeasured).
- Verify the diluted share count (818M stated vs. 867M basic — likely transcript error) against the press release before modeling Q1 EPS.
|
| Dec 3, 2024 | +23.19% | Q3 FY2025 | Read transcript briefingQuarter in one view- Q3 FY2025 revenue $1.516B, $66M above the midpoint of the $1.45B ±5% guide, +19% QoQ and +7% YoY — the promised return to YoY growth delivered, and the beat was driven by a stronger-than-forecasted custom silicon ramp plus double-digit sequential optics growth. Non-GAAP EPS $0.43 ($0.03 above midpoint of the $0.35–$0.45 guide), +43% QoQ — EPS growing more than 2x revenue, the operating-leverage proof point landing again.
- The quarter's defining event is strategic, not financial: a five-year, multi-generational expanded agreement with AWS spanning custom AI, optical DSPs, AEC DSPs, PCIe retimers, DCI modules, and Ethernet switching, plus Marvell using AWS for cloud EDA. Murphy called it "a significant step up in the expected volume of business" between the companies.
- Q4 guide is the accelerant: $1.8B ±5% (+19% QoQ, +26% YoY at midpoint), non-GAAP EPS $0.54–$0.64, and — a milestone — guided return to GAAP profitability ($0.11–$0.21 GAAP EPS).
- The offset: a ~$750M restructuring charge (about three-quarters non-cash), mostly impairments of acquired intangibles and purchased technology licenses tied to redirected enterprise/carrier investment. GAAP GM collapsed to 23% and GAAP operating margin to -46.4%; GAAP EPS -$0.78. Management says the charges are "largely behind us."
- AI quantification finally got a number, sort of: FY2025 AI is tracking ahead of the $1.5B target "on the order of hundreds of millions of dollars," and FY2026 is tracking ahead of $2.5B. Custom silicon specifically: ~$500M+ this year, $1B+ next year, against a reiterated $40B custom TAM and 20% share (~$8B) ambition.
What management is focused on- Declaring the AI-first transformation complete: data center now 73% of revenue (up from 69%), expected to rise again in Q4; Murphy explicitly called Marvell "an AI-first data center semiconductor company" and announced restructuring to "purposefully redirect" R&D intensity toward data center at the expense of enterprise/carrier investment levels.
- Selling the AWS agreement as validation of the full-platform strategy — not just custom, but the entire networking attach (optics, AEC, retimers, DCI, switching) — and as multi-generational revenue visibility. Murphy repeatedly steered questions back to it.
- Operating leverage as the financial narrative: FY-to-date revenue +31% (Q1→Q3) vs. non-GAAP EPS +79% (2.5x leverage); Q4 guide implies another step (revenue +19%, EPS +40% at midpoint). Custom scale "will help accelerate our timeline to achieve our long-term target operating margin model."
- Custom silicon competitive framing: Murphy now says only Marvell and "one other very large, highly scaled-up competitor" can ship these designs at volume, citing IP breadth (SerDes, HBM PHYs, die-to-die, packaging), A0/first-pass-silicon execution on 100B+ transistor chips, and manufacturing/yield capability as the barrier stack.
- Capital returns and SBC discipline: $200M buyback in Q3 (up from $175M, as promised), $525M cumulative FY-to-date against the $3B authorization; new explicit commitment to reduce stock-based compensation as a percentage of revenue with "significant improvement" expected.
- Succession and stability messaging: Loi Nguyen (Inphi co-founder) retiring April 2025 with succession planning underway; Murphy gave an emphatic "I am all-in" denial of press speculation about him leaving for another CEO role.
Key numbers and quarter mechanics- Revenue $1.516B (+7% YoY, +19% QoQ). Mix: data center 73%, enterprise 10%, consumer 6%, carrier 6%, auto/industrial 5%.
- End markets: data center record $1.1B (+98% YoY, +25% QoQ — well above the high-teens guide) on the custom step-up plus double-digit sequential optics growth; enterprise $151M (flat QoQ); carrier $85M (+12% QoQ); enterprise+carrier combined +4% QoQ, in line with the mid-single-digit guide; consumer $97M (+9% QoQ, above the "up slightly" guide); auto/industrial $83M (+9% QoQ, above the mid-single-digit guide — recovering from last quarter's miss).
- Non-GAAP: GM 60.5% (slightly below the ~61% guide — higher custom mix); OpEx $467M (in line with ~$465M); operating margin 29.7% (up from 26.1%); EPS $0.43. GAAP: GM 23%; OpEx $1.052B including restructuring; operating margin -46.4%; EPS -$0.78 (far outside the -$0.09 to +$0.05 guide due to the ~$750M charge).
- Cash flow/balance sheet: CFO $536M, +$230M QoQ — the rebound the prior quarter lacked, finally confirming the earnings ramp; inventory $859M (+$41M, deliberately built for the data center ramp); cash $868M (+$59M); debt $4.1B; leverage 2.23x gross / 1.76x net (first deleveraging in several quarters, from 2.29x/1.84x); $52M dividends; $200M buyback.
- Q4 FY2025 guide: revenue $1.8B ±5% (+19% QoQ, +26% YoY at midpoint); GAAP GM ~50%; non-GAAP GM ~60% (down ~50bps on custom mix); GAAP OpEx ~$710M; non-GAAP OpEx ~$480M; other expense ~$46M; non-GAAP tax 7%; basic 867M / diluted 877M shares; GAAP EPS $0.11–$0.21 (first guided GAAP profit); non-GAAP EPS $0.54–$0.64.
- FY2026 tax rate revised up again: now 10–11% (prior quarter's disclosure was 9%) on higher expected operating income — a small incremental EPS headwind versus what was signaled 90 days ago.
- Q4 end-market guides: data center low-to-mid-20s sequential growth (another "significant" custom step-up plus switch and interconnect growth); enterprise+carrier mid-teens sequential (acceleration, as promised, and better than the "double-digits" framed a quarter ago); consumer down mid-teens (gaming seasonality, consistent with the disclosed pattern); auto/industrial low-to-mid single digits up.
Product and launch scorecard- Custom AI silicon: the quarter's driver. First two programs ramped harder than forecast; "100 billion-plus transistor chips" shipped to high volume on first-pass silicon; supply chain capacity for customer growth forecasts "secured." Murphy put numbers on the record for the first time since AI Day: ~$500M+ custom this year, $1B+ next year, both being overshot. Two large programs (accelerator and compute) at two different customers both "tracking well"; additional programs entering production next year; third large customer ("customer C") reaffirmed as the largest of the three opportunities, tracking well for a 2026 ramp.
- AWS agreement: five-year, multi-generational, spanning custom AI plus the full networking portfolio; includes EDA-in-the-cloud collaboration. Not dollar-sized. Harlan Sur's direct question — is Marvell the ASIC vendor on the customer's next-gen 3nm training chip ramping late CY2025 — was not answered directly; Murphy pointed to the agreement's multi-generational scope instead. That non-denial/non-confirmation is the key open item.
- 1.6T PAM: shipments of the industry-first 5nm 1.6T DSP began in Q3 as guided; "strong design win momentum with leading customers"; production ramp accelerates next year, but 800G remains the volume driver through FY2026.
- New product: "Ara" — industry's first 3nm 1.6T DSP with 200G/lane electrical and optical interfaces, >20% module power reduction vs. predecessor. Announced ~18 months after the 5nm Nova; Murphy framed the faster cadence as deliberate ("hyperscale speed") in response to Svanberg's surprise at the timing.
- AEC: 800G (100G/lane) production ramp "accelerating" with multiple module partners — upgraded from last quarter's "initial shipments"; 1.6T AEC DSPs sampling (carried over).
- Switching: Teralynx 10 (51.2T, 5nm) now explicitly "gone into production" with "very, very strong" interest — firmer than last quarter's "starting to go into production with a lead customer"; still not revenue-sized.
- 2nm platform: "progressing very well," with "tremendous interest" for next-generation designs — positioning for the next custom generation; no wins attached.
- Carrier: the recovery driver is now named — a new layer-2 base station processor socket, won a few years ago, "in production now" after taking longer than expected. This is the first specific product-level explanation of the carrier recovery.
- Not mentioned this quarter: PCIe Gen6 retimer traction (only listed in the AWS scope), CXL devices, the Microsoft HSM win, and the storage ~$200M/quarter target — none updated.
Sell-side read-through- AI quantification: partially answered for the first time. Arya's FY25/FY26 ask drew "hundreds of millions" above the $1.5B target and the $500M+/$1B+ custom figures — more than any prior call — but no total AI dollar figure and no FY2026 growth rate. The $1.5B/$2.5B targets remain formally unraised.
- The 3nm next-gen ASIC question (Sur) was the sharpest moment of the call: citing "competitive noise" around the customer's announced 3nm program, Sur asked if Marvell is the vendor. Murphy praised the AWS agreement's multi-generational nature but did not confirm the socket. Read-through: the win is implied but not secured in disclosure — this is now the single most important competitive question for FY2026.
- Murphy's commitment to Marvell was challenged directly (Seymore, citing press reports of other CEO opportunities). The denial was emphatic and unprompted in length — the question existing at all is a new governance thread.
- Enterprise/carrier recovery quality (Curtis): Murphy recommitted to the ~$2B combined run rate ("we're going to get back... the question is when"), disclosed the carrier L2 base station socket as a company-specific driver, and noted H2 recovery is running ahead of prior expectations — but still shipping below end consumption.
- Optics inventory risk surfaced (Hari): tariff fears and potential customer inventory build. Murphy acknowledged the concern, cited strong bookings and orders inside lead time, and said they monitor diligently — a watch item, not a denial of the risk.
- Custom TAM framing (Lipacis): Murphy confirmed $40B custom within $75B data center TAM, 20% share goal, and a two-player viable landscape — and gave the clearest articulation yet of why custom wins (TCO, workload scale, customer network co-optimization, coexistence with merchant).
- Notably absent: no analyst pressed the $750M restructuring charge's composition, cash portion, or whether more enterprise/carrier write-downs could follow; no one asked about the FY2026 tax rate moving from 9% to 10–11%; the old $200M+ custom quarterly exit run-rate metric remains retired without comment, now superseded by the $500M+/$1B+ annual framing.
Management credibility- Delivery against last quarter's checkable claims was strong across the board: revenue and EPS above midpoint; data center's high-teens guide beaten (+25%); enterprise+carrier mid-single digits delivered (+4%); auto recovered (+9% vs. mid-single-digit guide); consumer beat ("slight" growth → +9%); optics grew sequentially as promised ("every quarter this fiscal year" — three for three); 1.6T shipped in Q3 as committed; the buyback increase delivered ($200M vs. $175M).
- The cash-flow question from last quarter resolved cleanly: CFO rebounded to $536M (+$230M), cash rose, and leverage ticked down for the first time in four quarters — the strongest credibility repair of the quarter.
- New specificity where management chose: custom silicon annual numbers ($500M+/$1B+), "hundreds of millions" of AI upside, the carrier L2 socket, the Ara 3nm announcement, and the AWS agreement itself — a named, checkable, multi-year contract.
- Credibility debits: the ~$750M restructuring charge arrived with no prior warning — last quarter's call gave no hint that enterprise/carrier intangibles and licenses were impaired, and the GAAP EPS guide (-$0.09 to +$0.05) missed by a wide margin as a result. Management framed it as "largely behind us," but the lack of signaling is a disclosure miss. The FY2026 tax rate moved from 9% to 10–11% in one quarter without explanation.
- The 3nm ASIC non-answer cuts both ways: Murphy has earned benefit of the doubt on execution (first-pass silicon, on-time ramps), but declining to confirm the next-gen socket while touting a "multi-generational" agreement leaves a gap between language and commitment that a competitor's win would expose.
- The $200M+ custom quarterly exit run-rate commitment — reiterated for three quarters then dropped last quarter — was again absent; the new annual framing implies it was conservative, but management never reconciled the old metric.
What changed versus the prior quarter- The recovery completed its arc: from "first growth quarter" (Q2) to +19% sequential, +7% YoY, and a Q4 guide of +26% YoY with a return to GAAP profitability. All five end markets grew in Q3 as guided.
- AI disclosure took a real step: from "significantly exceed" with no numbers to "hundreds of millions" above target and explicit custom silicon figures ($500M+ FY25, $1B+ FY26). Still no formal target raise.
- Customer concentration got a name and a contract: the AWS five-year multi-generational agreement is the first named, comprehensive custom+networking commitment — and simultaneously sharpens concentration questions.
- The cost of the pivot became visible: a ~$750M restructuring charge writing down enterprise/carrier-era assets, with R&D explicitly redirected to data center. The "AI-first" identity is now financial, not rhetorical.
- Cash flow inflected: CFO $536M vs. $306M; leverage down for the first time in four quarters; buybacks still rising ($200M) — now funded by operations.
- Gross margin trajectory firmed at a lower level: 60.5% printed, ~60% guided for Q4, and "about 60% through next year" — the ~61% "zip code" from last quarter has already stepped down; the old 64%+ model is further away.
- Enterprise/carrier recovery upgraded from order evidence to product evidence (L2 base station socket in production) and the Q4 guide accelerated to mid-teens vs. "double-digits" framed a quarter ago.
- FY2026 tax headwind increased: 10–11% vs. the 9% disclosed last quarter.
- New threads: CEO-departure speculation addressed head-on; Loi Nguyen's retirement announced; Investor Day set for June 10, 2025.
Bull case- The ramp is steepening, not flattening: +10% → +19% → +19% guided sequential growth, with Q4 at +26% YoY and data center guided to another low-to-mid-20s sequential step on a second "significant" custom ramp. Data center nearly doubled YoY in Q3.
- Custom silicon now has numbers and a contract: $500M+ this year, $1B+ next year, both being overshot, a five-year multi-generational AWS agreement, two programs ramping at two customers, more entering production next year, and the largest opportunity (customer C) still ahead in 2026 — against a $40B TAM with a 20% share goal and a self-described two-player viable field.
- Operating leverage is compounding exactly as promised: EPS +43% on revenue +19% in Q3; Q4 guide implies EPS +40% at midpoint on +19% revenue; non-GAAP operating margin 29.7% and guided toward the bottom end of the long-term range by end of next year; GAAP profitability guided for Q4.
- Cash flow now confirms the P&L: $536M CFO, rising cash, falling leverage, and a $200M buyback — the bear case from last quarter (buybacks funded by the balance sheet) is defused.
- The product cadence is accelerating: Ara 3nm 1.6T DSP announced ~18 months after Nova with >20% power savings, 1.6T shipping, 800G AEC ramping with multiple module partners, TL10 51.2T in production with strong interest, 2nm platform drawing "tremendous interest."
- Traditional businesses are now tailwinds: enterprise+carrier guided to mid-teens Q4 growth with a named carrier product driver and a recommitted ~$2B run-rate path; auto recovering; consumer seasonality now well-understood and quantified.
Bear case- Gross margin is grinding down, not stabilizing: 61.9% → 60.5% → ~60% guided → "about 60% through next year," with an explicit caveat that further custom upside pushes it lower. The mix shift is structural and the merchant-recovery offset is still early.
- The $750M restructuring charge — mostly enterprise/carrier-era intangibles and licenses — landed with zero prior signaling, blowing through the GAAP guide. Management says charges are "largely behind us," but the episode demonstrates that GAAP quality and disclosure timing lag the non-GAAP story; ~$250M of the charge is cash-natured obligations.
- The 3nm next-gen ASIC socket at the lead custom customer is unconfirmed. Sur asked directly, citing competitive noise, and Murphy did not claim it. If that socket goes to the "one other very large, highly scaled-up competitor," the multi-generational AWS narrative and the $1B+ FY2026 custom figure both get re-underwritten.
- Concentration is increasing by design: data center at 73% and rising, custom dominated by two customers today, and the AWS agreement deepens the largest relationship. Murphy's own framing — Marvell plus one competitor — means every socket is binary.
- FY2026 tax rate stepped up to 10–11% (from 9% signaled last quarter) — a second, larger EPS headwind that arrived without explanation.
- Optics inventory/tariff risk is now on the record (Hari's question); management's answer was monitoring, not evidence. Enterprise/carrier still ship below end consumption, and consumer declines mid-teens in Q4 — the non-AI businesses remain below trend.
- SBC remains a stated problem management is only now promising to fix ("significant improvement" expected, no numbers), and the GAAP/non-GAAP OpEx gap remains enormous ($1.052B vs. $467M this quarter, inflated by restructuring).
Next-quarter watchlist- Q4 delivery vs. $1.8B ±5%, ~60% non-GAAP GM, ~$480M non-GAAP OpEx, $0.54–$0.64 non-GAAP EPS — and the guided first GAAP profit quarter ($0.11–$0.21). Any further restructuring or impairment would directly contradict "largely behind us."
- The 3nm ASIC question: any confirmation, denial, or competitive announcement regarding the lead customer's next-gen training chip ramping late CY2025. Also watch for customer C progress toward its 2026 ramp and any new program entering production.
- Custom silicon trajectory vs. the new $500M+/$1B+ markers — whether FY2025 custom lands meaningfully above $500M and whether the FY2026 AI target ($2.5B) is formally raised now that "hundreds of millions" of FY2025 upside is on the record.
- AWS agreement economics: any sizing of the revenue step-up, the split between custom and networking, and whether AEC/retimer/switching attach at AWS becomes visible in segment commentary.
- Gross margin: whether ~60% is the floor for FY2026 as Meintjes sketched, and whether optics/merchant recovery offsets custom dilution in actual mix.
- Optics: fourth consecutive quarter of sequential growth as promised; 1.6T ramp acceleration "next year"; Ara 3nm sampling timing; any evidence of customer inventory build or tariff-driven pull-ins (Hari's thread).
- Enterprise/carrier: mid-teens Q4 growth delivered, the L2 base station socket's revenue contribution, and progress toward the ~$2B combined run rate; whether "shipping below consumption" language finally lifts.
- Cash and capital: CFO holding at or above the $536M level as revenue steps to $1.8B; inventory build ($859M, +$41M) converting to revenue rather than stacking; buyback cadence vs. the remaining authorization; SBC-as-%-of-revenue evidence.
- FY2026 setup: tax rate finalization (10–11%), OpEx growth vs. revenue growth, and any preliminary FY2026 framework ahead of the June 10, 2025 Investor Day, where the long-term model gets updated.
|
Corbanu