One of our primary holdings is Marvell Technology. We wrote about the company in January, explaining what it does and how we believe it is well-positioned to be the backbone of artificial intelligence (AI) infrastructure. As it turned out, Nvidia was thinking the same thing and announced on March 31 that it would invest $2 billion into Marvell as part of a strategic partnership. The Nvidia Marvell deal sent Marvell’s stock price up approximately 7%, while Nvidia’s increased by roughly 2.7%.

The current rate of development of AI infrastructure is massive. Both Alphabet and Meta are expected to spend at least $630 Billion on AI infrastructure this year. The companies are rushing to develop “AI Factories” — essentially, massive data centers that run AI workloads. However, developing these systems requires more than just AI processors; each system needs a vast amount of networking equipment, custom chips, and high-speed wiring to connect all the components. That is the gap this partnership intends to address.

Nvidia is currently the largest supplier of AI processors, and Marvell supplies the most customized infrastructure surrounding those processors. This agreement centers on a new Nvidia product called NVLink Fusion, which allows customers (such as a large cloud provider) to create their own customized processors and connect them to Nvidia’s GPUs and other hardware using common interfaces. All connected components remain compatible with Nvidia’s software framework, eliminating the need to rebuild software every time they customize.

Any cloud provider knows that rebuilding software is extremely costly, which is where the real value of compatibility lies.

There are two aspects to Marvell’s role in this partnership:

1. Custom Chips: Through NVLink Fusion, Marvell will create and sell XPUs (customized processing units designed for particular applications or workloads). A large cloud provider may purchase a generic processor off-the-shelf that performs many functions fairly well. Alternatively, through collaboration with Marvell, they could obtain a processor specifically designed to perform one task exceptionally well. This has always been Marvell’s area of specialization, and this partnership provides Marvell access to both Nvidia’s platform and its customer base to expand upon such activities.
2. Optical Interconnection/Connectivity: Marvell and Nvidia intend to collaborate on silicon photonics/optical interconnect technologies. Inside a data center, processors, servers, and storage devices are constantly communicating. Traditional electrically-connected cables begin to lose signal strength at higher speeds. Using optical technology that communicates through light rather than electrical signals means faster speeds and lower energy consumption. Given the increasing size and complexity of AI systems, moving information quickly throughout the system will become even more critical. Marvell has considerable experience in optical communications, and, through this partnership, will apply that expertise on a greater scale.

Nvidia’s contribution also breaks down into two categories:

1. Hardware: Nvidia is providing its suite of data center products, including Vera CPUs, ConnectX Network Cards, BlueField Data Processing Units & Spectrum-X Switches. They handle processing, route network traffic, and manage workloads. These are the building blocks that Marvell’s custom silicon will plug into.
2. Ecosystem access: Nvidia lets customers integrate Marvell’s chips into NVLink Fusion, keeping them inside their software and hardware world while still giving them room to customize. That’s a strategic win for Nvidia because it means customers don’t have to leave to get the flexibility they want. Jensen Huang, Nvidia’s CEO, framed it around the urgency: “The inference inflection has arrived. Token generation demand is surging, and the world is racing to build AI factories.”

For us, this is a strong signal. Marvell has previously said that its data center revenue could reach $15 billion by fiscal 2028, representing a growth rate of nearly 40%. A $2 billion commitment from the most important company in AI hardware makes that target feel more realistic, not less. Matt Murphy, Marvell’s Chairman and CEO, said the Nvidia Marvell deal connects Marvell’s leadership in analog, optical DSP, silicon photonics, and custom silicon to Nvidia’s expanding AI ecosystem. That’s exactly what we want to see in a company we already believe in.