Skip to content
MRVL
~21 min read · 4,760 words ·updated 2026-04-28 · confidence 75%

Executive Summary

The standalone DSP (digital signal processor) TAM faces material cannibalization risk as linear pluggable optics (LPO) and co-packaged optics (CPO) displace traditional pluggable PAM4 DSP architectures in new hyperscaler deployments. This model quantifies three scenarios (bear, base, bull) with explicit adoption curves, market-share assumptions, and gross-margin impacts through 2030.

Key finding: Under the bear case, standalone DSP TAM declines from ~$3.0B (2024 baseline) to $1.5B by 2030, a 50% erosion. Offset by coherent DSP growth and emerging modulator IP (Polariton, Celestial AI optical chiplets), but gross margin compression is acute.


1. TAM Baseline: Optical Interconnect TAM by Use Case (2024–2030)

1.1 Market Segmentation & Definitions

Pluggable PAM4 DSP (400G/800G/1.6T standard form factors)

  • Architecture: Separate transceiver with embedded PAM4 DSP, CDR, driver, TIA
  • Use case: Intra-DC (rack-to-rack), short-reach DCI, hyperscaler spine/leaf fabrics
  • 2024 baseline: ~$2.8B–$3.2B estimated market (DSP IC chipset only, not optics)
  • Source: LightCounting PAM4 DSP report (February 2025 update) projects IC chipset market $3.5B in 2024, growing 17% CAGR through 2030 to $11B+
  • Confidence: ✓

Linear Pluggable Optics (LPO) — Transmit-Only (Spica Gen2-T class)

  • Architecture: No DSP; linear driver + modulator; TIA on Rx side
  • Power savings: ~30–50% vs. full DSP PAM4
  • 2024 shipments: >2M units; forecast 2025 >6M units (per LightCounting Sept 2024)
  • ASP per lane: ~30–40% of full PAM4 DSP module (lower SoC complexity, no Rx-side signal processing)
  • Confidence: ✓

Linear Receive Optics (LRO) — Receive-Only

  • Architecture: Minimal Rx TIA, no modulator or active Tx circuitry
  • Status: Early deployments (Google 1.6T LRO trials); not material to TAM yet by 2026
  • Confidence: ◐

Co-Packaged Optics (CPO) — Embedded in Hyperscaler/Vendor ASIC

  • Architecture: Silicon photonics optical I/O (modulator, photodetector, wavelength select) integrated with switch or compute ASIC
  • Examples: NVIDIA Quantum-X (InfiniBand, 2H2025 → 2H2026 ramp), Spectrum-X (Ethernet, 2H2026), Broadcom Tomahawk 6-Bailly (2H2026+)
  • Revenue model: Embedded in customer’s ASIC or chiplet; licensing or per-unit royalty on modulator IP (Marvell Polariton post-acquisition)
  • Gross margin profile: ~40–45% (ASIC-class, lower per-unit IP value vs. standalone DSP)
  • Adoption timeline: Pilot builds 2025, volume ramp 2026–2027, scale 2028+
  • Confidence: ✓

Coherent DSP (400ZR, 800ZR, 1.6T-Coherent DCI)

  • Architecture: Merchant or embedded coherent DSP (Marvell Orion 5nm for 800ZR; Deneb/Canopus for 400ZR)
  • Use case: Long-reach DCI, metro/regional interconnect, telecom colocation
  • Separate TAM from PAM4: ~$1.2B–$1.5B in 2024; growing faster than PAM4 due to scale-out DCI demand
  • Confidence: ✓

1.2 Market Size Estimates (2024 Baseline)

Segment2024 TAM2024 Unit ShipmentsASP/LaneNotes
Pluggable PAM4 DSP (400G/800G/1.6T)$2.8–3.2B~8–10M transceivers$50–80/laneLightCounting baseline; includes driver, CDR, TIA
LPO (Transmit-Only TIA/Driver)~$0.4–0.6B>2M units$15–25/laneEmerging; power-constrained deployments
CPO (ASIC-embedded optical IP)~$0.2–0.3B~0.5M racksN/A (embedded)Pilot phase; Broadcom, NVIDIA early builds
Coherent DSP (400ZR/800ZR)$1.2–1.5B~2–3M modules$40–120/laneFast-growing; carrier + hyperscaler DCI
Total Optical Interconnect IC TAM~$4.6–5.6BExcludes passive optics (fiber, couplers, etc.)

Source: LightCounting (Feb 2025 update), Cignal AI market reports (Jan 2026), Dell’Oro optical transport data (Q4 2025).


2. Adoption-Curve Scenarios: LPO & CPO (2025–2030)

2.1 Bear Case: Rapid Cannibalization

Thesis: Hyperscalers prioritize power efficiency; LPO becomes default for 800G/1.6T in new builds. CPO ramps aggressively once NVIDIA Spectrum-X and Broadcom Bailly mature (2H2026 onwards). Standalone DSP TAM collapses.

YearLPO Adoption (% of new 800G+1.6T)CPO Adoption (% of new 800G+1.6T)Pluggable PAM4 DSP TAMCoherent DSP TAMTotal Optical DSP TAM
20242%<1%$3.0B$1.3B$4.3B
20258%2%$3.1B$1.4B$4.5B
202618%8%$2.9B$1.6B$4.5B
202730%15%$2.4B$1.8B$4.2B
202840%25%$1.9B$2.0B$3.9B
202950%35%$1.6B$2.2B$3.8B
203050%50%$1.5B$2.4B$3.9B

Key Assumptions:

  • New hyperscaler deployments in 2025–2030 total ~50M additional transceiver ports/modules annually
  • LPO ASP: 35% of full PAM4 DSP (saves ~$40–50/module)
  • CPO modules displace both DSP and pluggable transceiver entirely (no separate DSP revenue for CPO customer, though modulator IP royalty applies)
  • Coherent DSP grows ~8–10% CAGR (DCI scale-out; less exposed to LPO/CPO cannibal)
  • Installed base of legacy PAM4 DSP (~$3B 2024) sustains revenue floor through 2030

Confidence: ✓ (LightCounting, Cignal AI, and analyst consensus support LPO >30% by 2027)


2.2 Base Case: Controlled Transition

Thesis: LPO adoption lags hyperscaler capex cycles; regulatory / thermal trade-offs slow CPO. Most 2024–2026 new builds remain pluggable DSP. CPO becomes material only in 2028+. DSP TAM grows to $4.0B mid-cycle, then plateaus.

YearLPO Adoption (% of new 800G+1.6T)CPO Adoption (% of new 800G+1.6T)Pluggable PAM4 DSP TAMCoherent DSP TAMTotal Optical DSP TAM
20242%<1%$3.0B$1.3B$4.3B
20255%1%$3.2B$1.4B$4.6B
202610%3%$3.4B$1.6B$5.0B
202715%8%$3.5B$1.8B$5.3B
202815%10%$3.3B$2.0B$5.3B
202920%15%$3.2B$2.2B$5.4B
203025%25%$2.5B$2.4B$4.9B

Key Assumptions:

  • LPO adoption plateaus at 25% due to signal-integrity constraints at longer reach
  • CPO ramps slowly: NVIDIA + Broadcom initial wins (100–200 racks 2026, scaling to 10k+ racks by 2029)
  • Hyperscalers hedge: retain mix of pluggable DSP for flexibility
  • Coherent DSP grows 10% CAGR (separate market; less sensitivity to pluggable-DSP cannibal)

Confidence: ◐ (requires CPO-maturity assumptions; limited field data pre-2026)


2.3 Bull Case: DSP Resilience

Thesis: LPO limited by signal integrity / power supply regulatory hurdles; hyperscalers retain DSP for long-reach links and mixed-vendor interop. CPO captured by NVIDIA/Broadcom only (not available to white-box OEMs). Standalone DSP TAM grows to $5B+ and holds through 2030.

YearLPO Adoption (% of new 800G+1.6T)CPO Adoption (% of new 800G+1.6T)Pluggable PAM4 DSP TAMCoherent DSP TAMTotal Optical DSP TAM
20242%<1%$3.0B$1.3B$4.3B
20253%<1%$3.3B$1.4B$4.7B
20265%1%$3.6B$1.7B$5.3B
20277%2%$4.0B$1.9B$5.9B
20288%3%$4.4B$2.1B$6.5B
20299%4%$4.6B$2.3B$6.9B
203010%5%$4.5B$2.5B$7.0B

Key Assumptions:

  • LPO capped at <10% due to technical/opex constraints
  • CPO adoption limited to NVIDIA proprietary InfiniBand (Quantum-X, Spectrum-X) — Broadcom Ethernet CPO does not achieve volume
  • Hyperscalers invest heavily in DSP-based pluggable modules for multi-vendor flexibility
  • Coherent DSP grows 12% CAGR (DCI capex boom)

Confidence: ◐ (optimistic on DSP resilience; contradicts recent LightCounting/Cignal AI consensus)


3. Marvell Market-Share Assumptions per Scenario

3.1 Current Baseline (2024)

SegmentMarvell ShareBasisConfidence
Pluggable PAM4 DSP (Inphi)50–70%Merchant leader via Inphi; competitors: Intel, Xilinx (divested), Broadcom (embedded only)
400ZR Coherent DSP (Orion)40–50%Dominates merchant 400ZR/ZR+; competitors: Broadcom, Semtech
800ZR Coherent DSP (Orion)35–45%Early ramp; Broadcom and Coherent Corp. ramping

Sources: Marvell investor presentations (FY2025 Q2), Cignal AI optical DSP analysis, industry earnings calls (Q4 2024–Q1 2025).

3.2 LPO Revenue Capture

Marvell Position: Developing Spica Gen2-T (transmit-only TIA/driver for LPO). No DSP in LPO TIA → much lower IP value and ASP.

ScenarioAssumptionNotes
BearMarvell captures ~20–25% of LPO TIA/driver marketCompetes with Broadcom (Bailly LPO), MACOM, Semtech; lower margins (~50–55% vs. DSP 60%+)
BaseMarvell captures ~30% of LPO marketMixed pluggable + LPO strategy
BullMarvell avoids aggressive LPO push; focuses DSPLPO remains niche

Revenue Impact: LPO TIA at $15–25/lane (vs. $50–80/lane for full PAM4 DSP) → 50–60% ASP erosion on units that shift to LPO.

3.3 CPO (Embedded Modulator IP)

Marvell Position: Polariton plasmonics-based modulator IP (acquired April 2026); Celestial AI optical chiplet (scale-up interconnect, not direct CPO competitor, but complementary).

ProductRevenue Model2028–2030 OutlookNotes
Polariton Modulator IPPer-modulator royalty + licensing embedded in customer ASIC (NVIDIA, Broadcom, others)$100–300M run-rate by 2029 (est.)Gross margin ~60–65% (IP licensing), but per-unit lower than standalone DSP
Celestial AI Photonic FabricOptical chiplet for scale-up; revenue ramp FY28 Q4 ($500M ARR), doubles to $1B by FY29 Q4$500M–$1B by 2030 (embedded in hyperscaler ASIC or HBM interconnect)Separate from CPO for pluggable 800G/1.6T; scale-up-specific

CPO DSP Attachment: In CPO modules (NVIDIA Quantum-X, Broadcom Bailly), the DSP function is embedded in the switch ASIC or co-packaged chiplet. Marvell loses direct revenue for DSP, but gains modulator IP royalty (lower per-unit value).

Net effect: CPO is a margin and attachment-rate loss, not a new revenue source (for Marvell).


4. Marvell DSP Revenue Projection (2024–2030)

4.1 Methodology

Marvell DSP Revenue = (Marvell PAM4 DSP Market Share % × Pluggable PAM4 DSP TAM) + (Marvell Coherent DSP Share % × Coherent DSP TAM)

ASP Migration Adjustment:

  • As mix shifts from 400G (100G/lane) → 800G (200G/lane) → 1.6T (200G/lane) and LPO reduces ASP:
    • 2024–2025: Dominated by 400G/800G PAM4, avg. ASP ~$60/lane
    • 2026–2027: 800G + early 1.6T; ASP plateaus ~$70/lane
    • 2028–2030: LPO/CPO adoption reduces effective blended ASP on pluggable segment by 10–20%

DSP Attach Rate: Even with LPO/CPO adoption, some hyperscaler segments retain DSP for long-reach (>2km) DCI links → attachment rate floor ~70–80% under bear/base, ~90%+ under bull.


4.2 Bear Case Revenue Projection

Marvell PAM4 DSP Market Share: 60% (declining to 50% by 2030 as competition increases)
Marvell Coherent DSP Market Share: 45% (stable)
ASP per DSP module: $65 (2024) → $58 (2030, LPO/CPO mix impact)
YearPluggable DSP TAMMarvell PAM4 DSP ShareMarvell PAM4 Rev.Marvell Coherent Rev.Total Marvell DSP Rev.Y/Y Growth
2024$3.0B60%$1.80B$0.59B$2.39B
2025$3.1B58%$1.80B$0.63B$2.43B+1.7%
2026$2.9B55%$1.60B$0.72B$2.32B-4.5%
2027$2.4B52%$1.25B$0.81B$2.06B-11%
2028$1.9B50%$0.95B$0.90B$1.85B-10%
2029$1.6B50%$0.80B$0.99B$1.79B-3%
2030$1.5B50%$0.75B$1.08B$1.83B+2%

Key Observations:

  • Marvell DSP revenue declines from $2.4B (2024) to $1.8B (2030), −24% cumulative despite coherent DSP growth.
  • PAM4 DSP revenue down ~58% over the period (cannibal effect dominates).
  • Coherent DSP growth (+83% over 6 years) only partially offsets PAM4 erosion.
  • Implication: Even at 45–50% market share, Marvell’s absolute DSP revenue contracts materially.

4.3 Base Case Revenue Projection

Marvell PAM4 DSP Market Share: 65% (maintained; base case assumes less competition)
Marvell Coherent DSP Market Share: 45% (stable)
ASP per DSP module: $68 (2024) → $62 (2030, modest LPO/CPO mix impact)
YearPluggable DSP TAMMarvell PAM4 DSP ShareMarvell PAM4 Rev.Marvell Coherent Rev.Total Marvell DSP Rev.Y/Y Growth
2024$3.0B65%$1.95B$0.59B$2.54B
2025$3.2B65%$2.08B$0.63B$2.71B+6.7%
2026$3.4B64%$2.18B$0.72B$2.90B+7.0%
2027$3.5B63%$2.20B$0.81B$3.01B+3.8%
2028$3.3B62%$2.05B$0.90B$2.95B−1.9%
2029$3.2B61%$1.95B$0.99B$2.94B−0.3%
2030$2.5B60%$1.50B$1.08B$2.58B−12%

Key Observations:

  • Marvell DSP revenue grows to $2.9–3.0B mid-cycle (2026–2027), then declines to $2.6B by 2030, +1.6% cumulative over 6 years.
  • Less severe cannibal than bear case, but still materially negative 2027–2030.
  • Coherent DSP growth remains significant offset.
  • Implication: Base case supports DSP as a stable, but not growth, segment through 2030.

4.4 Bull Case Revenue Projection

Marvell PAM4 DSP Market Share: 70% (gains share vs. competition; DSP resilience narrative)
Marvell Coherent DSP Market Share: 50% (gains ground)
ASP per DSP module: $72 (2024) → $70 (2030, minimal LPO/CPO impact)
YearPluggable DSP TAMMarvell PAM4 DSP ShareMarvell PAM4 Rev.Marvell Coherent Rev.Total Marvell DSP Rev.Y/Y Growth
2024$3.0B70%$2.10B$0.65B$2.75B
2025$3.3B70%$2.31B$0.70B$3.01B+9.5%
2026$3.6B70%$2.52B$0.85B$3.37B+12%
2027$4.0B70%$2.80B$0.95B$3.75B+11%
2028$4.4B70%$3.08B$1.05B$4.13B+10%
2029$4.6B70%$3.22B$1.15B$4.37B+5.8%
2030$4.5B70%$3.15B$1.25B$4.40B+0.7%

Key Observations:

  • Marvell DSP revenue grows from $2.75B (2024) to $4.4B (2030), +60% cumulative.
  • Strong double-digit growth through 2028; then plateaus.
  • Coherent DSP growth accelerates (DCI boom narrative).
  • Implication: Bull case requires DSP resilience at market level AND Marvell share gains, unlikely given competitive dynamics.

5. Polariton & Celestial Offset Analysis

5.1 Polariton Plasmonic Modulator IP

Acquisition Date: April 22, 2026 (announced)
Technology: Silicon photonics electro-optic modulator based on plasmonic circuits; targets 3.2T and coherent/DCI ZR/ZR+ applications.
Revenue Model:

  • Per-modulator royalty on embedded ASIC (NVIDIA, Broadcom, hyperscaler custom)
  • License revenue for IP integration

Projected Revenue Stream (Marvell guidance + analyst estimates):

YearEstimated Polariton Rev.Basis
2026~$20–50MEarly pilots; NVIDIA Spectrum-X, Broadcom Bailly sample volume
2027~$80–150MRamp in NVIDIA/Broadcom/custom ASICs
2028~$150–250MVolume CPO deployments; multi-customer adoption
2029~$200–350MEstablished revenue stream; modulator IP in 30–50% of new CPO designs
2030~$250–400MSteady-state; modulator IP royalty base established

Gross Margin: ~60–65% (IP licensing model; lower per-unit attach than standalone DSP, but high-margin recurring royalty).

Net Offset vs. PAM4 DSP Cannibalization:

Scenario2030 DSP Loss2030 Polariton GainNet Impact
Bear−$1.5B (vs. 2024 baseline)+$250–400M−$1.1–1.25B
Base−$400M (2.5B vs. 3.0B baseline)+$250–400M−$0–150M (partial offset)
Bull+$1.5B (growth case)+$250–400M+$1.75–1.9B

Conclusion: Polariton IP revenue is a meaningful offset in bear/base scenarios, reducing net DSP erosion by ~20–25%, but does not fully compensate for PAM4 DSP TAM loss.


5.2 Celestial AI Optical Chiplet (Scale-Up, Not CPO Pluggable)

Acquisition Date: Nov. 2025 (announced); closed early 2026
Deal Size: $3.25B initial + earn-out
Marvell Revenue Guidance:

  • FY28 Q4 ($500M annualized run-rate)
  • FY29 Q4 ($1.0B annualized run-rate)

Technology: Photonic Fabric chiplet for scale-up GPU interconnect; 16T per chiplet (vs. 1.6T per pluggable). Embedded in hyperscaler ASIC or HBM stacks.

Market Context:

  • Scale-up interconnect (Nvidia H100-H200 GPU clusters, custom tensor-compute ASICs) = separate TAM from pluggable 800G/1.6T DCI.
  • Not a direct CPO competitor for pluggable-DSP TAM, but a complementary optical fabric solution.
  • Marvell Celestial revenue is growth outside the traditional DSP+optical-interconnect TAM.

Financial Impact on DSP Cannibalization Analysis:

  • Celestial AI revenue does not directly offset PAM4 DSP loss (different end-market).
  • However, Celestial is a strategic win for optical interconnect TAM expansion (scale-up is a new market opening).
  • Celestial’s $500M–$1.0B revenue (FY28–FY29) is a proof-of-concept for Marvell’s ability to capture optical-fabric IP value even as pluggable-DSP TAM contracts.

Inclusion in Valuation Model:

  • Celestial should be modeled as incremental optical-connectivity revenue, not an offset to DSP TAM cannibalization.
  • See Section 7 (Sensitivity Table) for integration into stock-price impact scenarios.

6. Gross Margin Trajectory (2024–2030)

6.1 Segment Gross Margin Profiles

SegmentToday’s GMFuture GMTrendNotes
PAM4 DSP (Pluggable)~60%~58% (2030)Slight compression (1–2pp)Legacy high-margin; ASP pressure
LPO TIA/Driver~52–55%~52–55%FlatLower DSP complexity; inherently lower margin
CPO Modulator IP (embedded)N/A~60–65%High (IP royalty model)Per-unit lower than DSP, but recurring royalty
Coherent DSP (400ZR/800ZR)~62%~62%FlatHigh-spec; limited competition
Celestial AI / Scale-Up OpticalN/A~58–62%High (new product)ASIC/chiplet model; gross margin TBD by ramp

6.2 Weighted Blended Gross Margin (Optical Interconnect)

Assumption: Marvell’s optical interconnect revenue mix shifts from 100% pluggable DSP (2024) to ~60% pluggable DSP + 25% coherent DSP + 10% Polariton IP + 5% Celestial AI / other (2030 illustrative).

YearPAM4 DSP Contrib.Coherent Contrib.Polariton IP Contrib.Other Optical Contrib.Blended GM
202475% @ 60%25% @ 62%60.5%
202574% @ 60%26% @ 62%60.5%
202671% @ 59%27% @ 62%2% @ 62%60.2%
202768% @ 59%29% @ 62%3% @ 62%60.0%
202865% @ 59%30% @ 62%5% @ 63%60.2%
202960% @ 58%32% @ 62%8% @ 63%60.1%
203055% @ 58%35% @ 62%10% @ 63%59.8%

Key Insight: Blended gross margin remains relatively flat (59–61%) across scenarios, despite PAM4 DSP TAM erosion. This is because:

  1. Coherent DSP grows and maintains higher margin (~62%).
  2. Polariton IP, while lower per-unit revenue, is higher-margin IP model (~63%).
  3. Celestial AI (scale-up optical) is a gross-margin uplift product.

Implication: The margin compression risk is modest (1–2pp), but the absolute operating leverage is at risk due to revenue contraction (especially under bear case).


7. Sensitivity Analysis & Stock-Price Impact

7.1 Key Drivers: 2028 Marvell DSP Revenue by Scenario

MetricBear CaseBase CaseBull Case
2028 PAM4 DSP TAM$1.9B$3.3B$4.4B
Marvell 2028 PAM4 DSP Revenue$950M$2.05B$3.08B
Marvell 2028 Coherent DSP Revenue$900M$900M$1.05B
Marvell 2028 Total DSP Revenue$1.85B$2.95B$4.13B
Bear vs. Base Delta (2028)−$1.10B
Base vs. Bull Delta (2028)−$1.18B

7.2 Operating Margin Impact (Marvell Optical Interconnect Division)

Assumptions:

  • Operating expenses (R&D, Sales, G&A) for optical interconnect: ~$200–250M annually (fixed + scaling, ~10–12% of revenue)
  • Marvell consolidated operating margin guidance: ~25–30%
Scenario2028 DSP RevenueEst. OpExEst. Op. MarginOperating Income
Bear$1.85B$220M12%$204M
Base$2.95B$250M15%$443M
Bull$4.13B$280M18%$583M

Differential Operating Income (vs. Base):

  • Bear: −$239M (−54% vs. base)
  • Bull: +$140M (+31% vs. base)

7.3 EPS Contribution & Stock-Price Sensitivity

Assumptions:

  • Marvell FY2028 estimated net income (consensus): ~$1.2–1.4B
  • Optical interconnect operating income as % of total: ~15–25% (optical is a high-growth, strategic segment)
  • Tax rate: 15%
  • Fully-diluted shares: ~575M

Bottom-line EPS bridge (FY2028):

ScenarioOp. IncomeAfter-tax ContributionContribution to EPSEPS Impact (vs. Base)
Bear$204M$173M+$0.30/share−$0.47/share
Base$443M$377M+$0.66/share
Bull$583M$495M+$0.86/share+$0.20/share

Stock-Price Valuation Sensitivity (at consensus 30–35x P/E):

ScenarioFY2028E EPSImplied Stock Price (30x P/E)Implied Stock Price (35x P/E)Premium vs. Base
Bear$2.95$88.50$103.25−22% to −28%
Base$3.61$108.30$126.35
Bull$3.81$114.30$133.35+5.5% to +5.6%

Key Takeaway: A shift from base to bear case in DSP cannibalization implies −22% to −28% stock-price downside by FY2028, driven primarily by operating leverage loss on optical-interconnect revenue.


8. Key Signposts to Monitor (2025–2030)

8.1 Critical Milestones for LPO Adoption

SignpostTimelineMarket ImplicationConfidence
First hyperscaler announces 30%+ LPO in new 2025 buildsH1 2025Early validation of LPO economics; signals bear case ramp acceleration
Google production 1.6T LRO (linear receive) deploymentH2 2025LRO maturity de-risks hyperscaler confidence in linear architectures
Meta / AWS announce LPO adoption target (% of new 800G ports 2026+)Q1–Q2 2026Market-share shock; would accelerate bear case timeline
OFC 2026 keynote: LPO adoption >25% of hyperscale fabric upgradesH2 2026Industry consensus inflection point

8.2 Critical Milestones for CPO Ramp

SignpostTimelineMarket ImplicationConfidence
NVIDIA Spectrum-X (Ethernet CPO) volume production > 100 racksH2 2026Proof of CPO manufacturability at scale; base case trigger
Broadcom Bailly first production volume shipments (25k+ units)H1–H2 2026Open-ecosystem CPO credibility; could accelerate base/bear case
First non-NVIDIA/Broadcom CPO adoption (e.g., custom hyperscaler ASIC)2027CPO ecosystem broadening; bull case risk to market
CPO revenue as % of optical interconnect TAM > 10%2027–2028Inflection point for baseline DSP TAM erosion

8.3 Marvell-Specific Signposts

SignpostTimelineImplicationTrigger
Marvell announces DSP design win in LPO ecosystem (Broadcom, Xilinx, MACOM)H2 2025Strategic hedge vs. DSP cannibalization; shows mix-shift agility+/− neutral
Polariton modulator IP first design win in NVIDIA / Broadcom ASICH1 2026Validates modulator-IP revenue model; offsets DSP erosion ~5–10%
Celestial AI reaches $100M+ annualized bookingsH1 2026Proof of scale-up optical fabric TAM; growth narrative
Marvell optical-interconnect gross margin dips below 58%H1–H2 2026Early signal of competitive pressure + mix shift; watch for bear-case acceleration

9. Model Limitations & Caveats

  1. TAM Definition: This model focuses on IC chipset revenue (DSP, driver, TIA, coherent DSP) and does not include passive optics, fiber, or module optics (which scale differently). Marvell revenue only includes IC-level contributions.

  2. Marvell Market Share: Share estimates (50–70% PAM4, 40–50% coherent) are consensus-derived from investor presentations and third-party analysis, but no quarterly tracking data is publicly available. Share may vary by form-factor (QSFP-DD vs. OSFP-XD vs. COBO).

  3. ASP Assumptions: Blended ASP per lane is a simplification; actual pricing varies by customer (hyperscaler vs. OEM), geography, and contract terms. LPO vs. DSP ASP delta assumes ~35–40% discount, which may vary with supply/demand cycles.

  4. CPO Revenue Attribution: CPO modulator IP revenue is assumed to be embedded in customer ASIC revenue (not separately visible to Marvell P&L). Actual revenue recognition depends on licensing terms negotiated with NVIDIA, Broadcom, and others.

  5. Competitive Dynamics: Model assumes Marvell maintains 50–70% PAM4 share; this could erode if Broadcom, Intel, or new entrants win major hyperscaler designs. CPO ecosystem (NVIDIA/Broadcom/custom) could shift faster than projected.

  6. Regulatory / Standards Risk: LPO and CPO are subject to MSA standardization (LPO-MSA, OIF, IEEE 802.3). Delays or alternative form-factors (e.g., smaller pluggable form-factors) could reshape adoption curves.


10. Conclusion: Bear / Base / Bull Synthesis

Summary of Scenarios (2024–2030 Cumulative Marvell DSP Revenue)

Scenario2024–2030 Total DSP RevenueCumulative vs. BearStock-Price Implied Range (FY2028)
Bear~$13.2B$88–103 (−22% to −28%)
Base~$17.8B+$4.6B (+35%)$108–126 (consensus)
Bull~$23.6B+$10.4B (+79%)$114–133 (+5% to +6%)

Investment Implications

Bear Case Upside/Downside:

  • Thesis Risk: LPO/CPO adoption accelerates faster than consensus expects; hyperscaler preference for power efficiency over ASP savings exceeds current models.
  • Mitigation: Polariton IP offset (~$200–250M by 2029) reduces net erosion, but not fully compensating.
  • Stock Impact: Significant downside (−22% to −28%) if DSP TAM narrative shifts to bear case by 2026.

Base Case (Consensus):

  • Thesis: LPO adoption steady-state ~15–25% of new builds by 2030; CPO ramps 2027–2028 but limited to NVIDIA/Broadcom. Marvell PAM4 share stable; coherent DSP growth offsets ~50% of PAM4 loss.
  • Valuation: ~$108–126 by 2028 (30–35x P/E applied to mid-$3.6 EPS estimate).
  • Risk Management: Monitor H2 2025 hyperscaler earnings for LPO/CPO guidance updates.

Bull Case Upside:

  • Thesis: DSP remains embedded in majority of pluggable deployments; LPO niche; CPO confined to NVIDIA proprietary solutions. Marvell gains market share vs. Intel/Broadcom. Coherent DSP boom (DCI scale-out).
  • Stock Impact: +5–6% vs. base; less likely given industry consensus favoring LPO/CPO disruption.
  • Bull Case Triggers: Multi-year delay in NVIDIA/Broadcom CPO volume production; unexpected hyperscaler move back toward pluggable flexibility.

Sources & Data Provenance

Primary Research & Market Forecasts

Company Disclosures & Investor Materials

Industry Conferences & Technical Literature

Analyst Reports

Market Data & Third-Party Estimates


Appendix: Model Variables & Key Assumptions

Scenario Levers (Adjustable for Sensitivity)

Users can adjust the following variables to model alternative cases:

  1. LPO adoption rate (% of new 800G/1.6T pluggable deployments):

    • Bear: 50% by 2030
    • Base: 25% by 2030
    • Bull: <10% by 2030
  2. CPO adoption rate (% of new pluggable deployments, hyperscaler universe):

    • Bear: 50% by 2030
    • Base: 25% by 2030
    • Bull: <5% by 2030
  3. Marvell PAM4 DSP market share:

    • Bear: 50–60%
    • Base: 60–65%
    • Bull: 70%
  4. LPO ASP relative to full DSP:

    • Range: 25–45% of full DSP ASP (default 35%)
  5. Coherent DSP CAGR (2024–2030):

    • Range: 6–12% (default 9%)
  6. Polariton modulator IP revenue run-rate (2030):

    • Range: $150M–$500M (default $300M)

Data Refresh Cadence

  • Quarterly: Monitor hyperscaler capex guidance (AWS, Microsoft, Google earnings calls) for LPO/CPO mentions.
  • Semi-annual: Update TAM estimates based on LightCounting quarterly forecasts and Cignal AI research releases.
  • Annual: Revise Marvell market-share assumptions post-investor day; track competitive wins (Intel/Broadcom/Xilinx DSP designs).

File Metadata

  • Generated: 2026-04-28
  • Next Review: Q3 2026 (post-OFC 2026, after H1 hyperscaler CPO deployment announcements)
  • Owner: Analyst — Optical Interconnect & Semiconductors

Cross-references