Broadcom (AVGO) — Marvell's Primary Direct Competitor

Executive Description

Broadcom Inc. (NASDAQ: AVGO) is Marvell’s primary direct competitor across every meaningful AI-infrastructure segment: custom AI ASICs, Ethernet switching silicon, optical DSPs, co-packaged optics (CPO), PCIe/NVLink-class connectivity, and storage controllers. Broadcom is also a roughly 6x larger company by revenue (FY2025 AI semiconductor revenue alone of $8.4B; total semiconductor revenue >$30B; total enterprise revenue including VMware ~$50B+) compared to Marvell’s $8.2B FY2026 total revenue. The competitive picture is asymmetric: Broadcom is incumbent and dominant in networking (Tomahawk family at 75%+ Ethernet switching share) and in two of the four major hyperscaler custom-silicon programs (Google TPU training, Meta MTIA Iris). Marvell counters with strength in AWS Trainium, growing inference-side positioning at Google (TPU 8i with MediaTek), Meta MTIA Arke (inference variant), Microsoft Maia, and an optical-chiplet beachhead via the Celestial AI acquisition. Cross-link competitor disclosures re marvell for AVGO management commentary on Marvell.

Company Background

Founded: 1961 (as HP Semiconductor); current Broadcom Inc. structure formed via 2016 Avago-Broadcom merger; relisted on NASDAQ. ✓
CEO: Hock Tan (since 2006 at Avago; through Broadcom merger). ✓
Scale: FY2025 revenue ~$60B (semiconductor + VMware/software). Semiconductor solutions ~$30B; AI-specific semiconductor: $8.4B FY2025 (+106% YoY). ✓
Market cap: Multi-trillion-dollar enterprise value (early-2026 levels), making Broadcom the largest pure-play semiconductor company globally aside from Nvidia.
Business mix: Networking (Tomahawk, Jericho, Trident), custom AI ASIC, broadband, wireless (Apple FBAR / iPhone content), industrial, software (VMware, Symantec, CA).

Comparison: AVGO vs MRVL

Custom AI Silicon

Customer	Broadcom Position	Marvell Position	Edge
Google TPU (training)	TPU 1 through TPU 8t — exclusive 7+ year partner	None on training	AVGO ✓
Google TPU (inference) + MPU	Limited disclosed role	TPU 8i with MediaTek; MPU design partner	MRVL ✓
Meta MTIA Iris (training)	Lead partner; April 2026 multi-GW extension	None	AVGO ✓
Meta MTIA Arke (inference)	None disclosed	Lead partner	MRVL ✓
AWS Trainium 2/3	Limited / none disclosed	Exclusive or near-exclusive partner	MRVL ✓
Microsoft Maia	Networking SerDes role / selective IP	Architecture / DSP / physical implementation	MRVL ◐
OpenAI custom silicon	Reported design-partner role on rumored OpenAI ASIC	Not disclosed	AVGO ◐

Implication: Broadcom holds the larger and more entrenched custom-silicon book (Google + Meta training are the two largest single-customer AI ASIC programs by die-volume globally). Marvell holds AWS Trainium as its anchor and is expanding into inference adjacencies at Google and Meta. Broadcom’s FY2027-2028 custom-silicon ramp (Google TPU 8t/9t, Meta Iris multi-GW) is structurally larger than Marvell’s; however, Marvell has more upside leverage from a smaller base.

Networking & Switching Silicon

Broadcom Tomahawk 6 at 102.4 Tbps holds 75%+ Ethernet switching market share, supplemented by Jericho 3 (router/inter-DC), Trident 5 (campus/edge). ✓
Marvell Teralynx 10 matches at 102.4 Tbps but has materially smaller market share; Marvell’s strength is complementary DSP + DPU + Ethernet PHY rather than dominant standalone switching silicon.
Edge: AVGO clearly. Cross-link competitor disclosures re marvell Section 1 — Broadcom commentary.

Optical Interconnect & DSPs

Broadcom DSP family (Trident-pair, Bailly CPO) is a meaningful AI-fabric platform; Broadcom’s Tomahawk-Bailly co-packaged optics is positioned as the next-gen scale-up optical alternative to merchant pluggables.
Marvell DSP family (Ara 1.6T, Electra DSP, COLORZ pluggables) is competitive at the merchant-DSP level; Marvell’s 1.6T DSP is shipping in volume to module OEMs (Coherent, Innolight, Lumentum — cross-link partner supplier chain). Marvell’s Polariton acquisition extends the optical platform toward 3.2T modulation (cross-link polariton); Celestial AI extends into scale-up optical chiplets (cross-link celestial ai).
Edge: MRVL slight edge in merchant DSP volume; AVGO leads in CPO momentum. ◐

FY Revenue Scale

Metric	Broadcom (FY2025)	Marvell (FY2026)	Ratio
Total revenue	~$60B (incl. VMware/software)	~$8.2B	~7.3x
Semiconductor revenue	~$30B	~$8.2B	~3.7x
AI-specific semiconductor	~$8.4B	~$6.1B (data center)	~1.4x
Custom AI ASIC revenue (est.)	~$5-6B	~$2-3B	~2-3x
Optical/DSP revenue (est.)	~$1-2B	~$1-1.5B	~1-1.3x

Implication: While Broadcom is dramatically larger overall, the AI-semiconductor gap is much narrower (~1.4x) and the custom-AI-ASIC gap is on the order of 2-3x. This is the basis for the bull thesis that Marvell can compress the AI-revenue ratio with disproportionate FY2027-2028 growth.

Risks & Catalysts (For the Marvell investor)

Risks

Broadcom expands inference share: If Broadcom successfully bids into Meta Arke or Google TPU 8i contracts, Marvell’s incremental wallet share at hyperscalers contracts.
Tomahawk-Bailly CPO disrupts merchant DSP volumes in a way that pressures Marvell’s Ara/Electra ASP and unit volumes.
Broadcom NVLink-substitute fabric: If Broadcom’s scale-up fabric solutions gain hyperscaler traction faster than Marvell Photonic Fabric (Celestial), the scale-up optical TAM consolidates around AVGO.
Broadcom margin advantage: AVGO operates at >65% non-GAAP semiconductor gross margins; Marvell’s margin profile is structurally lower (~62-64% non-GAAP). Sustained margin gap limits Marvell’s reinvestment-flywheel velocity.

Catalysts

Marvell share gains in Google or Meta — visible inference-side ramp would compress AVGO’s AI lead.
Marvell custom-XPU customer-set expansion — incremental hyperscaler design wins (e.g., a tier-2 hyperscaler) would diversify Marvell’s risk concentration.
Photonic Fabric production rollout — credible scale-up optical chiplet would directly contest Tomahawk-Bailly CPO.