NVIDIA Invests $2 Billion in Marvell to Expand AI Infrastructure Ecosystem

Nvidia said on Tuesday it will invest $2 billion in Marvell Technology and deepen ties with the chipmaker through its NVLink Fusion platform, expanding the range of custom AI infrastructure that can plug into Nvidia’s broader ecosystem.

The partnership will connect Marvell’s custom XPUs and NVLink-compatible scale-up networking to Nvidia’s AI factory and AI-RAN stack, while Nvidia will provide components including its Vera CPU, ConnectX network interface cards, BlueField DPUs, NVLink interconnect and Spectrum-X switches, the companies said. They also plan to work together on silicon photonics and optical interconnects, two technologies seen as increasingly important as AI systems grow larger and more power-hungry.

The deal further underscores Nvidia’s effort to keep its hardware and software platform at the center of the AI buildout even as large customers seek more customized systems. By bringing Marvell into the NVLink Fusion ecosystem, Nvidia is giving customers a way to build semi-custom AI infrastructure that remains compatible with Nvidia’s rack-scale systems and networking stack.

For Marvell, the tie-up adds momentum to a business that has been leaning harder into AI data center silicon, networking and optical connectivity. Earlier this month, the company forecast fiscal 2028 revenue of nearly $15 billion, above Wall Street estimates, driven by demand for custom chips and high-speed interconnects for AI infrastructure. It has also been bulking up in networking and photonics through deals for XConn Technologies and Celestial AI. Marvell shares rose nearly 12% in premarket trading on Tuesday, while Nvidia gained about 1.5%.