Broadcom vs Marvell
Broadcom and Marvell are both placed on 9 of the same AI data center parts in this catalog, spanning layers 1, 2, 3. Broadcom appears on 12 parts in total and Marvell on 13.
At a glance
| Broadcom | Marvell | |
|---|---|---|
| Ownership | public | public |
| Headquarters | Palo Alto, United States | Santa Clara, United States |
| Primary listing | AVGO | MRVL |
| Market cap | 1.54T USD | 205.4B USD |
| Parts placed | 12 | 13 |
| Layers covered | 3 | 3 |
| Units / subsidiaries | 1 | 0 |
| Seen in machines | 4 | 1 |
Market caps are a nightly snapshot in each listing's own currency, never converted.
What the shared parts are
- Hyperscaler custom ASICs — AI accelerator chips (often called XPUs) designed in-house by a hyperscaler for its own fleet, typically co-developed with an outside ASIC design partner and fabbed by a merchant foundry.
- ASIC design services, IP & EDA — Outside firms and licensors — design-services houses, IP blocks, and EDA software — that hyperscalers and accelerator startups use to turn a chip specification into a manufacturable design.
- NICs / SuperNICs — Network interface cards, including Nvidia's higher-bandwidth "SuperNIC" variants, that connect a server's CPU and GPUs to the cluster network.
- PCIe/CXL switches & retimers — Fabric silicon — switches that fan a single PCIe or CXL root out to many downstream devices, and retimers that clean up the signal over longer runs — connecting accelerators, SSDs, and pooled memory back to the host.
- 5 more in the table below.
The 9 parts both make
| Part | Broadcom | Marvell |
|---|---|---|
| Hyperscaler custom ASICs Layer 1 — Compute silicon & packaging | design partner inferred Google TPU; Meta MTIA | design partner inferred AWS Trainium/Inferentia; Microsoft Maia |
| ASIC design services, IP & EDA Layer 1 — Compute silicon & packaging | service inferred | service inferred |
| NICs / SuperNICs Layer 1 — Compute silicon & packaging | manufacturer inferred | manufacturer inferred |
| PCIe/CXL switches & retimers Layer 2 — Memory & storage | manufacturer inferred | manufacturer inferred |
| Scale-out switch ASICs Layer 3 — Networking & interconnect | manufacturer disclosed Tomahawk/Jericho | manufacturer inferred Teralynx |
| Optical DSPs, drivers, TIAs Layer 3 — Networking & interconnect | manufacturer inferred | manufacturer inferred DSPs |
| Silicon photonics & co-packaged optics Layer 3 — Networking & interconnect | manufacturer inferred CPO switches | manufacturer inferred |
| Retimers inside AECs Layer 3 — Networking & interconnect | manufacturer inferred | manufacturer inferred |
| Timing & telemetry silicon Layer 3 — Networking & interconnect | manufacturer inferred | manufacturer inferred |
Only Broadcom
- Scale-up fabric layer 3
- Optical transceivers layer 3
- Lasers layer 3
Only Marvell
- DPUs layer 1
- CXL controllers & memory expanders layer 2
- DCI / coherent optics layer 3
- Controllers layer 2
Named in the same machines
- Nvidia GB300 NVL72 rack teardown
Questions
- Do Broadcom and Marvell compete?
- In this catalog they are both placed on 9 of the same parts, across 3 layers: Hyperscaler custom ASICs, ASIC design services, IP & EDA, NICs / SuperNICs, PCIe/CXL switches & retimers and 5 more. Sharing a part does not mean selling the same product to the same buyer, so read each row's product note.
- Which has more placements, Broadcom or Marvell?
- Broadcom has 12 placements and Marvell has 13. Placement counts reflect how thoroughly each company has been catalogued as well as its real breadth, so they are not a ranking.
Generated from the catalog's part-to-company table. Each placement is tagged disclosed, reported or inferred; see sources & method. Neither company is endorsed or ranked here.