Nvidia Invests $2 Billion in Marvell to Bring Custom AI Chips Into Its NVLink Fusion Ecosystem
Nvidia takes a $2 billion stake in Marvell Technology, integrating one of the semiconductor industry's top custom ASIC designers into its NVLink Fusion interconnect platform.
Overview
Nvidia announced on March 31 that it has invested $2 billion in Marvell Technology and entered a strategic partnership to connect Marvell’s custom chip and networking capabilities to the Nvidia AI factory ecosystem through NVLink Fusion, according to an official press release from Nvidia. Marvell shares rose more than 13 percent on the news, while Nvidia gained roughly 5 percent, as reported by CNBC.
The deal pulls one of the semiconductor industry’s two dominant custom ASIC design houses — the other being Broadcom — directly into Nvidia’s proprietary interconnect orbit, a move that could reshape the competitive landscape for hyperscale AI infrastructure.
What We Know
Under the agreement, Marvell will provide custom XPUs and NVLink Fusion-compatible scale-up networking, while Nvidia will supply supporting technologies including its Vera CPU, ConnectX network interface cards, BlueField data processing units, NVLink interconnect, and Spectrum-X switches, according to the Nvidia press release. NVLink Fusion is a rack-scale platform that enables customers to build semi-custom AI infrastructure using the Nvidia NVLink ecosystem.
The two companies will also collaborate on silicon photonics technology — which uses light rather than copper wiring to transmit data at higher speeds and lower power consumption — and on transforming telecommunications networks into AI infrastructure through Nvidia Aerial AI-RAN for 5G and 6G, as detailed in the press release.
“The inference inflection has arrived. Token generation demand is surging, and the world is racing to build AI factories,” Nvidia CEO Jensen Huang said in the announcement. Marvell CEO Matt Murphy added that the partnership would enable customers to “build scalable, efficient AI infrastructure” by connecting Marvell’s leadership in analog, optical DSP, silicon photonics, and custom silicon to Nvidia’s ecosystem.
Marvell projects revenue growth of nearly 40 percent with a target of approximately $15 billion for fiscal 2028, as reported by CNBC.
Strategic Significance
The deal is notable because Marvell is simultaneously one of Nvidia’s competitors and one of its most important new partners, as Tom’s Hardware observed. Marvell ranks alongside Broadcom as one of only two major custom ASIC design houses serving hyperscalers. Its client list includes Amazon (for which it developed the Trainium series of AI accelerators), Microsoft, and Google — companies that have been investing heavily in custom silicon specifically to reduce their dependence on Nvidia’s premium GPUs.
By integrating Marvell into the NVLink Fusion ecosystem, Nvidia ensures that even customers who opt for custom XPUs rather than its flagship GPUs will still rely on Nvidia’s interconnect, networking, and rack-scale infrastructure. The partnership effectively converts what was a competitive threat into an ecosystem expansion, tying custom chip development to Nvidia’s proprietary fabric rather than to independent alternatives.
This continues a pattern of Nvidia using strategic investments to extend its reach beyond GPU sales. Big Tech firms including Alphabet and Meta are expected to spend at least $630 billion on AI infrastructure this year, according to CNBC, driving demand for the server and networking chips that both Nvidia and Marvell produce.
What We Don’t Know
The precise structure of Nvidia’s $2 billion stake — whether it represents newly issued shares, a secondary purchase, or a combination — has not been disclosed. It is also unclear how the partnership will affect Marvell’s existing custom silicon relationships with Amazon, Microsoft, and Google, which together account for a significant share of Marvell’s custom compute revenue.
Whether Broadcom, the other leading custom ASIC designer, will respond with its own interconnect strategy or partnership remains to be seen. The deal’s impact on hyperscalers’ long-term chip sovereignty ambitions is equally uncertain: bringing custom XPUs onto NVLink Fusion could either accelerate custom silicon adoption by making it easier to integrate, or deepen Nvidia’s infrastructure lock-in.
The timeline for commercial availability of Marvell XPUs running on NVLink Fusion racks has not been announced.