Skip to content
MRVL
~14 min read · 3,234 words ·updated 2026-04-29 · confidence 63%

Comparable M&A Transactions — Valuation Framework Reference

This file builds an M&A comp set for grounding Marvell’s standalone DCF and any forward “what would MRVL be worth in a strategic transaction” thought experiments. It also informs how to value Marvell’s own acquisition cadence (Celestial AI Nov 2025; Polariton Apr 2026) and the implied “premium for photonic interconnect IP” market price.

Today’s date: 2026-04-29 · MRVL ticker: NASDAQ:MRVL · CIK: 0001835632

Confidence flags: ✓ verified from primary press release / SEC filing · ◐ widely reported but not primary-sourced · ⚠ figure inferred or contested

Section 1 — Marvell’s Own M&A History

Marvell has executed nine acquisitions of $100M+ since 2017 under CEO Matt Murphy. The pattern: large transformational deals (Cavium, Inphi) interspersed with bolt-on technology acquisitions (Aquantia, Innovium, Tanzanite). Every deal has been data-center-aligned; nothing in storage / consumer / auto since the platform pivot. The 2025-2026 cadence (Celestial AI + Polariton) signals a doubling-down on photonic interconnect.

Cavium — closed July 2018

  • Headline price: $6.0B (cash + stock; ~$40/share + 2.1757 MRVL shares per CAVM share) ✓
  • Cavium last-12-month revenue: ~$960M (CY2017) ◐
  • EV/Sales: ~6.3x ◐
  • EV/EBITDA: ~25x ◐ (Cavium adj. EBITDA ~$240M)
  • Strategic logic: ARM CPU expertise (ThunderX2), networking processors, security IP. Created the foundation for what became Marvell’s data center compute and infrastructure roadmap.
  • Post-deal trajectory: ThunderX server CPU ultimately wound down (~2020); networking IP became core to OCTEON DPU line.
  • Confidence: ✓ (closing announced 2018-07-06, 8-K filed)

Avera Semiconductor — closed September 2019

  • Headline price: $740M cash (+ up to $90M earnout) ✓
  • Acquired from: GlobalFoundries (Avera was GF’s ASIC design business)
  • Strategic logic: Custom ASIC design services capability — the seed organization that ultimately became Marvell’s custom silicon (XPU) franchise that scaled to $1.5B in FY26.
  • EV/Sales / EBITDA: Not disclosed at standalone level; Avera was a services/design business, not a product P&L.
  • Confidence: ✓ (closing 2019-09-04, 8-K filed)
  • Retrospective note: Avera was the most strategically consequential acquisition relative to its price tag — without Avera, the Trainium/Maia/etc. custom XPU franchise does not exist.

Aquantia — closed September 2019

  • Headline price: ~$452M cash ✓
  • Aquantia LTM revenue: ~$104M ◐
  • EV/Sales: ~4.4x ◐
  • Strategic logic: Multi-Gig (2.5G/5G/10G) ethernet PHY IP for enterprise and automotive ethernet. Auto-ethernet portion sold to Infineon Aug 2025 for $2.5B.
  • Confidence: ✓

Inphi — closed April 2021

  • Headline price: $10.0B (cash + stock; ~$66/share cash + 2.323 MRVL shares per IPHI) ✓
  • Inphi LTM revenue: ~$683M (CY2020) ◐
  • EV/Sales: ~14.6x ✓ (very rich — explained by Inphi’s dominant PAM4 DSP / electro-optics TAM trajectory)
  • EV/EBITDA: ~50x ◐ (Inphi adj EBITDA ~$200M)
  • Strategic logic: Best-in-class PAM4 DSP for 200G/400G/800G optical interconnects. Created Marvell’s optical leadership franchise.
  • Premium analysis: Inphi traded at ~$176/share pre-announcement; deal struck at implied ~$222/share = +26% premium ✓.
  • Retrospective note: This is the most expensive deal but arguably the best on a strategic value basis. The PAM4 DSP franchise generates ~$1.2B+ revenue annually in FY26 at ~70% gross margins — already paying back the $10B headline.
  • Confidence: ✓ (closing 2021-04-20, 8-K filed)

Innovium — closed October 2021

  • Headline price: $1.1B in stock ✓
  • Innovium business: TERALYNX cloud-scale ethernet switch silicon (12.8T/25.6T/51.2T)
  • Strategic logic: Ethernet switch silicon for hyperscale / cloud — combined with Marvell custom ASIC + Inphi DSP creates an end-to-end data center silicon stack.
  • Confidence: ✓
  • Retrospective: Switch silicon competition vs. Broadcom Tomahawk/Trident has been intense. Innovium’s TERALYNX 10/12 (51.2T) is in production but market share remains modest (~10-15% of hyperscale ethernet switching ◐).

Tanzanite Silicon Logic — closed March 2023

  • Headline price: Not disclosed; estimated <$200M ⚠
  • Tanzanite business: CXL (Compute Express Link) memory pooling and disaggregation
  • Strategic logic: Memory-fabric expertise for AI/server architecture. CXL is a multi-year roadmap; revenue contribution still negligible.
  • Confidence: ◐ (closing announced; financials confidential)

Celestial AI — closed February 2, 2026

  • Headline price: $3.25B (cash + stock per Marvell IR press release of 2025-12-02) ✓
  • Announcement date: 2025-12-02 (concurrent with Q3 FY26 earnings call)
  • Close date: 2026-02-02
  • Celestial business: Photonic Fabric optical interconnect; scale-up fabric architecture
  • Celestial revenue at deal: Pre-revenue / early-revenue ⚠
  • Strategic logic: Co-packaged optics, scale-up fabric for multi-rack AI training/inference; “transformational milestone” per Murphy.
  • Revenue contribution per management: H2 FY28 start; $500M run rate by Q4 FY28; $1B+ by Q4 FY29 ◐
  • Operating cost: +$50M annually post-close
  • Implied “ramp multiple”: $3.25B / $500M target = 6.5x forward revenue (forward 2.5 years out) ◐
  • Confidence: ✓ (announced 2025-12-02; close confirmed Q4 FY26 call 2026-03-05)

Polariton Technologies — announced April 22, 2026

  • Headline price: Not disclosed (terms TBD) ⚠
  • Announcement date: 2026-04-22
  • Polariton business: Plasmonic modulator IP; CMOS-compatible plasmonic modulators for >3.2T optical interconnects
  • Strategic logic: Fills modulation gap not covered by Celestial (which is fabric-architecture-focused). Plasmonic competes against polymer EO (LWLG), thin-film LiNbO3, and conventional Mach-Zehnder.
  • Polariton revenue: Pre-revenue; research-stage IP company ⚠
  • Implied logic: Marvell paying for IP/team rather than revenue; likely <$500M deal value (no 8-K material announcement at IR press release detail level as of 2026-04-29) ⚠
  • Confidence: ◐ (announcement language verified; deal terms inferred)

Cumulative M&A spend (Murphy era)

  • Cavium $6.0B + Avera $0.74B + Aquantia $0.45B + Inphi $10.0B + Innovium $1.1B + Tanzanite $0.2B + Celestial $3.25B + Polariton TBD = **$22B+** in cumulative deal spend.
  • Largest single deal: Inphi ($10B); most strategically consequential per dollar: Avera ($0.74B → custom silicon franchise ~$1.5B revenue FY26).

Marvell own deal multiples summary

DealEVTarget LTM RevEV/SalesEV/EBITDAConfidence
Cavium 2018$6.0B$960M6.3x~25x◐
Avera 2019$0.74Bn/a (services)n/an/a✓
Aquantia 2019$0.45B$104M4.4xn/a◐
Inphi 2021$10.0B$683M14.6x~50x✓
Innovium 2021$1.1Bnot disclosedn/an/a◐
Tanzanite 2023<$0.2Bn/an/an/a⚠
Celestial 2026$3.25B<$50M ⚠>65xn/a✓ deal terms; ◐ rev
Polariton 2026TBDn/an/an/a⚠

Pattern: Murphy pays high revenue multiples for forward-leaning IP (Inphi 14.6x, Celestial >65x) but has been disciplined on conventional silicon assets (Cavium 6.3x, Aquantia 4.4x). Forward multiple expansion accepted when the technology is structurally enabling for AI workloads.

Section 2 — Peer Optical / Networking M&A (Comparable Multiples)

Lumentum — Oclaro (closed December 2018)

  • Headline price: $1.7B cash + stock ✓
  • Oclaro LTM revenue: ~$580M ◐
  • EV/Sales: ~2.9x ◐
  • EV/EBITDA: ~14x ◐
  • Strategic logic: Telecom/datacom optical components consolidation; positioned Lumentum as #1 in coherent transceiver components.
  • Premium: ~27% over pre-announcement Oclaro share price ◐
  • Confidence: ◐

Coherent Corp (formerly II-VI) — Coherent Inc. (closed July 2022)

  • Headline price: $7.0B (cash + stock); rebranded combined entity to Coherent Corp ✓
  • Coherent Inc. LTM revenue: ~$1.5B ◐
  • EV/Sales: ~4.7x ◐
  • EV/EBITDA: ~22x ◐
  • Strategic logic: Lasers + optical communications + compound semiconductors. Created the leading laser/photonics conglomerate; competes broadly with Lumentum in datacom transceivers.
  • Confidence: ◐

Cisco — Acacia Communications (closed March 2021)

  • Headline price: $4.5B cash (revised up from original $2.6B 2019 deal that was renegotiated) ✓
  • Acacia LTM revenue: ~$465M (CY2020) ◐
  • EV/Sales: ~9.7x ✓
  • EV/EBITDA: ~30x ◐
  • Strategic logic: Coherent DSP, pluggable optics, 400ZR; key to Cisco’s data center optical strategy.
  • Premium: Final price was 71% premium to original 2019 deal price ✓
  • Confidence: ✓
  • Cross-relevance to MRVL: Acacia is the closest comp to Marvell’s PAM4 DSP business in optical DSP IP. The 9.7x EV/Sales suggests optical DSP IP commands a premium even relative to broader networking.

Broadcom — VMware (closed November 2023)

  • Headline price: $69B (cash + stock; ~$142/share equivalent) ✓
  • VMware LTM revenue: ~$13.4B ◐
  • EV/Sales: ~5.1x ◐
  • EV/EBITDA: ~21x ◐ (VMware adj EBITDA ~$3.3B)
  • Strategic logic: Software diversification beyond chips; recurring revenue platform.
  • Confidence: ✓
  • Cross-relevance to MRVL: Broadcom’s largest deal; reset comp expectations for what “diversified semi” buyers can pay. Less directly relevant for MRVL but informs Broadcom’s competitive bandwidth post-VMware.

AMD — Pensando (closed May 2022)

  • Headline price: $1.9B cash ✓
  • Pensando LTM revenue: Pre-revenue / early; ~$50M ⚠
  • EV/Sales: >35x (forward) ⚠
  • Strategic logic: DPU (data processing unit) for cloud / hyperscale; networking infrastructure.
  • Confidence: ✓ deal; ⚠ revenue
  • Cross-relevance to MRVL: Direct comp — Marvell OCTEON DPU is competitive with Pensando. The $1.9B price for a pre-revenue DPU implies significant strategic value attached to networking offload silicon.

AMD — ZT Systems (closed September 2024 announcement; closed early 2025)

  • Headline price: $4.9B (75% cash, 25% stock) ✓
  • ZT Systems LTM revenue: ~$10B (server systems builder) ◐
  • EV/Sales: ~0.5x (low — systems integration is low-margin) ◐
  • Strategic logic: AI server systems vertical integration; AMD plans to divest the manufacturing business and retain design/services.
  • Confidence: ✓
  • Cross-relevance to MRVL: Limited — different model (systems integration vs. silicon IP).

Intel — Habana Labs (closed December 2019)

  • Headline price: $2.0B cash ✓
  • Habana revenue: Pre-revenue / negligible ⚠
  • Strategic logic: AI training/inference accelerators (Gaudi line).
  • Confidence: ✓
  • Retrospective: Gaudi line ultimately under-performed expectations; Intel announced wind-down of Gaudi in late 2024 in favor of “Falcon Shores” GPU. Cautionary tale for AI accelerator M&A.
  • Cross-relevance to MRVL: Comp for “buy AI silicon team” deals; suggests revenue traction matters more than headline IP.

Nvidia — Mellanox (closed April 2020)

  • Headline price: $7.0B cash (originally $6.9B; revised up due to FX) ✓
  • Mellanox LTM revenue: ~$1.34B ◐
  • EV/Sales: ~5.2x ◐
  • EV/EBITDA: ~28x ◐
  • Strategic logic: InfiniBand / high-speed networking for HPC/AI; foundation of NVIDIA’s AI networking moat (NVLink, Quantum, Spectrum).
  • Premium: ~14% over pre-announcement Mellanox price ✓
  • Confidence: ✓
  • Cross-relevance to MRVL: Most strategically analogous deal in AI networking. The Mellanox-Nvidia integration created the InfiniBand / NVLink ecosystem that NVIDIA Fusion is now extending — and Marvell’s NVIDIA $2B investment (March 31, 2026) connects MRVL into that ecosystem. EV/Sales of ~5.2x for high-speed networking with strong AI tailwinds is a useful base-rate comp.

Intel — Tower Semiconductor (TERMINATED August 2023)

  • Headline price: $5.4B cash (announced Feb 2022; terminated Aug 2023 due to China regulatory inaction) ✓
  • Tower LTM revenue: ~$1.5B ◐
  • EV/Sales: ~3.6x ◐
  • Cross-relevance: Cautionary on regulatory risk — semi M&A faces increasingly hostile review (CHN MOFCOM specifically). MRVL’s Celestial AI and Polariton deals are sub-billion or low-billion and pure-IP; should pass review easily.
  • Confidence: ✓

AMD — Xilinx (closed February 2022)

  • Headline price: $49.0B all-stock ✓
  • Xilinx LTM revenue: ~$3.7B ◐
  • EV/Sales: ~13.2x ◐
  • EV/EBITDA: ~38x ◐
  • Strategic logic: Adaptive computing / FPGA platform; data center / AI accelerator complement to AMD CPUs/GPUs.
  • Premium: ~24% over pre-announcement Xilinx price ✓
  • Confidence: ✓
  • Cross-relevance: Adjacent — FPGA silicon at ~13x EV/Sales is a useful upper-bound comp for “specialty silicon with high strategic value.” Inphi at 14.6x is in the same range.

Peer M&A multiples summary

DealEVTarget RevEV/SalesYear
Lumentum-Oclaro$1.7B$580M2.9x2018
Cisco-Acacia$4.5B$465M9.7x2021
AMD-Xilinx$49B$3.7B13.2x2022
Coherent II-VI / Coherent$7.0B$1.5B4.7x2022
AMD-Pensando$1.9B~$50M>35x (fwd)2022
NVIDIA-Mellanox$7.0B$1.34B5.2x2020
Intel-Habana$2.0Bpre-revn/a2019
AMD-ZT Systems$4.9B$10B0.5x2024
Broadcom-VMware$69B$13.4B5.1x2023
Marvell-Inphi$10B$683M14.6x2021
Marvell-Celestial AI$3.25B<$50M ⚠>65x (fwd)2026

Median EV/Sales (silicon/optical only, excl. systems and software): ~5-7x for established revenue businesses; ~30x+ for pre-revenue strategic IP.

Section 3 — Custom AI Silicon M&A Comparables

The custom AI silicon (XPU/ASIC) value chain has very limited M&A precedent because the major design service / custom silicon partners are independent and remain so:

GUC (Global Unichip Corporation)

  • Listing: Taiwan listed (Ticker: 3443.TWO); ~58% owned by TSMC ✓
  • CY2024 revenue: ~$2.5B ◐ (rapid growth from AI ASIC services, esp. AWS Trainium)
  • CY2024 market cap: ~$8.5B-$10B (range over year) ◐
  • EV/Sales: ~3-4x ◐
  • Status: Independent. No active M&A signals. TSMC majority ownership de facto blocks change of control.
  • Strategic significance: GUC is the silicon design partner for Amazon Trainium (alleged to be moving more share to Alchip/GUC per Stifel/Benchmark theses; contested). If acquired by a Marvell competitor, would significantly shift custom silicon competitive landscape.

Alchip Technologies

  • Listing: Taiwan listed (Ticker: 3661.TWO); independent ✓
  • CY2024 revenue: ~$1.6B ◐
  • CY2024 market cap: ~$4-6B ◐ (highly volatile)
  • EV/Sales: ~2.5-4x ◐
  • Status: Independent. Founded 2003. No M&A signals.
  • Strategic significance: Alchip allegedly winning Amazon Trainium 3/4 share at Marvell’s expense (per Stifel Mar 10 2026 customer-cliff piece, contested by JPMorgan field research).

Faraday Technology

  • Listing: Taiwan listed (Ticker: 3035.TWO); independent ✓
  • CY2024 revenue: ~$0.9B ◐
  • EV/Sales: ~3-4x ◐
  • Strategic significance: Smaller ASIC design house; less directly competitive with Marvell.

Implications for Marvell valuation

  • Custom silicon M&A precedent is sparse. No major hyperscaler has acquired its custom silicon partner (Amazon has not bought Alchip/GUC; Google has its own internal team; Microsoft uses Marvell for Maia silicon design, not by acquisition).
  • Pure-play custom silicon design houses trade at 2.5-4x EV/Sales in public markets.
  • Marvell trades at a structurally higher multiple because the custom silicon business is one part of a diversified data center silicon platform (custom + DSP + switching + DPU).
  • Strategic premium analysis: If a hyperscaler (Amazon, Microsoft, Google) wanted to vertically integrate the custom silicon partner, the precedent multiples (Mellanox 5.2x, Inphi 14.6x, Acacia 9.7x) suggest a 7-12x EV/Sales range = $55-95B EV for MRVL standalone (vs. current ~$110-130B based on $135 stock price ⚠). This is well below current trading EV, implying the “strategic acquirer” floor is below market — which makes sense given Marvell’s already strong trading multiple.

Section 4 — Premium Analysis & Valuation Framework Application

What does Marvell’s M&A history tell us about its own intrinsic value framework?

  1. Inphi at 14.6x EV/Sales (2021) — Murphy paid this multiple for the optical DSP franchise that he is now using to anchor the optical interconnect thesis (1.6T DSP, expanding into Celestial photonic interconnect). Inphi’s pre-deal EV ~$10B in 2021 supports a high-single-digit / low-teens multiple for optical DSP-grade IP.

  2. Celestial AI at >65x forward EV/Sales (2026) — Marvell paid an extreme forward multiple for pre-revenue photonic interconnect IP. Implication: Marvell’s discount rate for photonic/optical interconnect IP is very low (i.e., management believes the technology is structurally enabling for FY28+ AI infrastructure).

  3. Polariton (2026) — Most likely <$500M deal value for plasmonic modulator IP team. Marvell continues to pay for forward-looking modulator IP because LWLG, Polariton, and others are perceived as critical to >3.2T port speed transitions.

  4. The acquisition cadence is the thesis. Murphy’s M&A pattern is: 1 transformational deal every 2-3 years (Cavium → Inphi → Celestial), supplemented by IP/team bolt-ons (Aquantia, Innovium, Tanzanite, Polariton). Each transformational deal extends Marvell’s “platform” (compute → optical → photonic). A 5th transformational deal is plausible by ~2028-2029 — likely targeting either silicon photonics (Lightmatter, LWLG modulator, or similar) or memory fabric (CXL extension beyond Tanzanite).

Implied Marvell strategic-acquirer ceiling

If Marvell were itself a target, peer comps suggest:

  • Base case (median peer optical / networking M&A): ~5-7x EV/Sales applied to FY27 ~$11B revenue = $55B-$77B EV (well below current trading EV).
  • Aggressive case (strategic premium for AI infrastructure incumbent): ~10-14x EV/Sales (Acacia/Inphi multiples) applied to FY27 ~$11B revenue = $110B-$154B EV (in line with current trading).
  • No-deal case: Marvell trades at approximately the upper bound of comparable multiples already, suggesting strategic acquirers (NVIDIA, Broadcom, Intel) face a “make-vs-buy” tradeoff where Marvell’s market price already prices in much of the strategic value.

NVIDIA $2B partnership investment (March 31, 2026)

The NVIDIA $2B investment — announced four weeks after Q4 FY26 earnings — is not a buyout but a strategic investment that signals NVIDIA’s preference for partnership over acquisition. Implication: a full NVIDIA acquisition of Marvell is unlikely in the near term (NVIDIA can secure the strategic value via NVLink Fusion ecosystem participation rather than a $150B+ takeout).

Section 5 — LWLG-Relevant Cross-References

For the LWLG investor frame, this comp set highlights:

  1. Polariton (Apr 2026) deal multiple is unknown but likely small — this is a useful anchor for what an EO-modulator IP company is worth in the current market. LWLG should not expect a strategic offer that materially exceeds Polariton terms, unless LWLG can demonstrate revenue traction or qualified-customer milestones.

  2. Celestial AI’s $3.25B price tag at <$50M revenue is the upper-bound for what a major silicon photonics platform can fetch from Marvell. Celestial had: Photonic Fabric architecture, DARPA/government funding, hyperscaler engagement, and an established engineering team (~150+). LWLG would need to match that profile to justify a similar valuation.

  3. Marvell’s own Inphi acquisition at $10B / 14.6x EV/Sales is the far upper bound — that price was justified by Inphi’s already-leading PAM4 DSP revenue franchise and clear AI optical TAM line of sight. LWLG-stage assets would not hit those multiples.

  4. No comparable company-of-LWLG-profile (pre-revenue polymer EO modulator) has been acquired in recent semi M&A history — the closest is Polariton (plasmonic modulator IP). Pre-revenue modulator IP appears to trade in the $200-500M range as a strategic IP/team acquisition.

Sources

  • Marvell IR press releases: https://investor.marvell.com/news-events/press-releases/
  • Marvell Cavium acquisition 8-K (2018-07-06)
  • Marvell Inphi acquisition press release (2021-04-20)
  • Marvell Celestial AI announcement (2025-12-02): https://investor.marvell.com/news-events/press-releases/
  • Marvell Polariton announcement (2026-04-22): see Yahoo Finance referenced in sell side coverage pull
  • Lumentum/Oclaro merger details: SEC filings 2018
  • Cisco Acacia revised deal: 2021 SEC 8-K
  • AMD Xilinx all-stock merger: 2022 SEC 8-K
  • Broadcom VMware closing: 2023-11-22 8-K
  • NVIDIA Mellanox closing: 2020-04-27 8-K
  • AMD Pensando: 2022 SEC 8-K
  • Intel Habana announcement: 2019-12-16
  • Intel Tower termination: 2023-08-16
  • AMD ZT Systems: 2024-08-19 announcement
  • GUC, Alchip public filings (TWSE)
  • comps valuation — peer trading multiples (vs. M&A multiples here)
  • earnings calls — Celestial AI announcement language Q3 FY26
  • valuation framework — primary use site for these comps
  • dsp cannibalization model — Inphi DSP franchise underwrites optical thesis
  • 03_ecosystem/... (if present) — Polariton / Celestial competitive positioning vs. LWLG
  • competitive landscape — LWLG competitive frame for Polariton / Celestial

Last updated: 2026-04-29

Cross-references