Nvidia: The Rival That Bought Two Billion Dollars of the CompanyNarrow moat
Marvell Technology (MRVL) — moat facet
Nvidia funded the company designing the chips meant to replace Nvidia's — and got Marvell inside its own interconnect in return.
In March 2026 Nvidia invested $2 billion in Marvell and brought it into the NVLink Fusion ecosystem, with Marvell supplying custom XPUs and NVLink-compatible scale-up networking while Nvidia supplies the Vera CPUs, ConnectX NICs, BlueField DPUs, Spectrum-X switches and the interconnect itself1. The two also agreed to collaborate on silicon photonics and on AI-RAN for 5G and 6G.
Read the relationship without the press release and it is genuinely strange. Marvell's custom accelerator business exists because hyperscalers want an alternative to buying Nvidia GPUs — that is the entire premise of the product line. Nvidia has now funded the company designing those alternatives and invited it inside its own rack architecture2. Nvidia also competes directly with Marvell's optics franchise: its Spectrum-X and Quantum-X Photonics switches integrate co-packaged optics precisely to reduce the pluggable modules Marvell's DSPs sit inside3.
The logic on Nvidia's side is not mysterious. If custom silicon is going to exist regardless, better that it plug into NVLink than into an open standard Nvidia does not control — ecosystem membership as the softer form of lock-in. For Marvell it is real access to the rack interior, the market it could not otherwise reach, granted on the incumbent's fabric and on the incumbent's terms.
Watch whether Marvell's scale-up networking ships in volume inside NVLink Fusion systems, and watch the open-standard alternative in parallel. A supplier with only one route into a market is a supplier whose terms are set for it.
This one genuinely improved for Marvell. A year ago the rack interior was closed and the only route in was an open standard that has not yet arrived. Today Marvell is inside NVLink Fusion with $2 billion of Nvidia's money on its balance sheet and a joint silicon-photonics effort. Widening — with the qualification that the door was opened by the incumbent and can be narrowed by it, so this is access rather than a moat.
Nvidia's $2.0bn of convertible preferred is deep in the money; the gap shows what the partnership has been worth to the holder so far.
- ReportedNVIDIA invested $2B in Marvell; Marvell supplies custom XPUs and NVLink-compatible scale-up networking.Marvell Form 8-K, Exhibit 99.1 — NVIDIA and Marvell announce a strategic partnership connecting Marvell to the NVIDIA AI factory and AI-RAN ecosystem through NVLink Fusion; NVIDIA has invested $2 billion in Marvell; Marvell provides custom XPUs and NVLink Fusion compatible scale-up networking while NVIDIA provides Vera CPU, ConnectX NICs, BlueField DPUs, NVLink interconnect and Spectrum-X switches; the companies will also collaborate on silicon photonics and Aerial AI-RAN for 5G/6G — March 31, 2026 · publ. March 31, 2026 · source ↗
- Third-party estimateMarvell's largest customers use its custom silicon specifically to depend less on NVIDIA GPUs.Reporting on the NVIDIA-Marvell investment — NVLink Fusion characterised as soft ecosystem lock-in, given that Marvell's largest customers use its custom silicon specifically to reduce dependence on NVIDIA GPUs — 2026 · publ. 2026 · source ↗
- ReportedNVIDIA's Spectrum-X and Quantum-X Photonics switches integrate co-packaged optics.NVIDIA newsroom — Spectrum-X Photonics and Quantum-X Photonics co-packaged-optics networking switches, integrating silicon photonics into the switch to scale AI factories to millions of GPUs with 4x fewer lasers and 3.5x more power efficiency — 2025-2026 · publ. 2025 · source ↗
- Marvell Form 10-K, FY2026 — Business & Risk Factors (SEC EDGAR)
- Marvell Form 8-K, Exhibit 99.1 — NVIDIA AI Ecosystem Expands as Marvell Joins Forces Through NVLink Fusion
- NVIDIA — Spectrum-X Photonics co-packaged-optics networking switches