Nvidia–Marvell deal reshapes AI infrastructure

Nvidia is investing US$2 billion in Marvell Technology through an equity stake aimed at deepening collaboration on artificial intelligence infrastructure. The deal links Marvell’s custom XPUs and networking products with Nvidia’s NVLink Fusion AI factory platform and AI-RAN offerings for telecom networks.

Under the arrangement, Marvell’s networking, silicon photonics, and telecom hardware will connect directly into Nvidia’s broader compute stack. Nvidia, known for GPUs used in AI and accelerated computing, is using the investment to tie its core compute business more closely to data center and telecom hardware.

The collaboration is intended to extend Nvidia’s AI factory concept into more semi-custom data center and telecom builds. Rather than only selling fixed rack systems, Nvidia is bringing Marvell into its ecosystem so customers that want their own accelerators can still use Nvidia’s interconnects, networking, and software.

This approach also addresses custom-chip efforts by large buyers while wiring Nvidia’s platform deeper into 5G and prospective 6G radio access networks. Competitors such as Broadcom (AVGO), Intel (INTC), and AMD are active in this area, making ecosystem depth an important strategic consideration for AI infrastructure.

Marvell and Nvidia are key semiconductor players in data infrastructure and AI. Marvell focuses on semiconductors for networking, storage, and connectivity, while Nvidia leads in GPUs and AI computing across data centers and other applications.

Bank of America (BAC) cited a newly announced partnership with Nvidia when it raised its price target on Marvell stock to $125 and maintained a Buy rating. The firms will work together in Nvidia’s NVLink Fusion ecosystem, which is designed to support more flexible, high performance computing systems.

Within this ecosystem, Marvell contributes custom XPUs and networking switches, while Nvidia brings CPUs, networking chips, and AI interconnect technology. The goal is to build heterogeneous AI systems that can scale quickly and handle increasingly complex workloads.

The companies are also expanding collaboration into silicon photonics and co-packaged optics, targeting deployments through 2028 and beyond. Bank of America (BAC) noted potential adoption by major cloud providers, including Amazon (AMZN), whose Trn4 chip is expected in late 2026 or early 2027, and Microsoft (MSFT) with its Maia-3 chip on a similar timeline.

Marvell’s stock has reacted strongly to the developments. Shares closed at $99.05 on March 31 after jumping 12.8% in a single day, extending an already strong run. Over the past year, the stock has gained more than 60%, outpacing the S&P 500 (SPX).

The Nvidia partnership positions Marvell near the center of the next phase of AI infrastructure growth, spanning cloud data centers and telecom networks. As AI workloads expand into 5G and future wireless networks, integrating compute with high speed networking and silicon photonics has become a key focus for large chipmakers.

At Nvidia’s 2026 GTC event, CEO Jensen Huang said management expects at least US$1 trillion in cumulative lifetime sales from the company’s current generation Blackwell chips and next generation Vera Rubin chips by the end of 2027. He had previously projected US$500 billion in lifetime sales for these units by the end of 2026.

Blackwell chips began primary sales in the fourth quarter of fiscal 2025, covering late 2024 to early 2025. Nvidia reported US$216 billion in revenue in fiscal 2026, which ended in January 2026, with an unspecified portion coming from Blackwell sales.

Wall Street analysts expect Nvidia to generate US$369 billion in revenue for fiscal 2027, a 71% year over year increase, and US$480 billion for fiscal 2028. Combined, these estimates total just over US$1 trillion in revenue across the period considered.

Nvidia’s AI strategy now spans Vera CPUs, Rubin GPUs, Bluefield DPUs, ConnectX NICs, Spectrum X switches, and third party XPUs such as Marvell’s that connect natively to NVLink Fusion. The US$2 billion Marvell stake fits into this push to offer a more complete AI infrastructure platform for large scale deployments.