Nvidia has invested $2 billion in Marvell Technology, integrating the chipmaker into its NVLink Fusion ecosystem. This partnership covers custom AI accelerators, silicon photonics, and 5G/6G infrastructure, ensuring Marvell’s custom chip designs for companies like Amazon, Google, and Microsoft continue to generate Nvidia revenue through required platform components—a move that turns a competitive threat into an ecosystem tax.
Nvidia announced this strategic partnership with Marvell, focusing on NVLink Fusion, a platform allowing third-party silicon to connect to Nvidia’s interconnect fabric. Marvell’s stock rose nearly 13%, and Nvidia’s by 5.6% following the news. Rather than just a deal, it reflects an infrastructure policy embedded in silicon.
The partnership includes Marvell supplying custom XPUs and NVLink Fusion-compatible networking, while Nvidia contributes with CPUs, network interface cards, data processing units, interconnect, and switches. Nvidia and Marvell will also collaborate on silicon photonics, technology using light for faster data transfer between chips, crucial for AI clusters. Nvidia CEO Jensen Huang described it as an inflection point, with surging demand for AI infrastructure.
NVLink Fusion requires at least one Nvidia product in its platform, and Nvidia controls partner access to NVLink IP licenses. Thus, Marvell’s AI accelerators continue to funnel revenue to Nvidia. This mirrors Nvidia’s pattern of $2 billion investments in companies like CoreWeave and Synopsys, aimed at securing Nvidia’s dominance by making recipients reliant on its platform.
Marvell is a key target due to its booming custom AI accelerator business, which threatens Nvidia’s GPUs. Marvell’s custom AI business generated $1.5 billion in revenue for fiscal 2026, projected to double by fiscal 2028. It has 18 active projects, including collaborations with major tech companies like Amazon and Meta.
By investing and integrating Marvell into NVLink Fusion, Nvidia ensures Marvell, which builds custom solutions to rival Nvidia’s GPUs, still contributes to Nvidia. NVLink Fusion has expanded rapidly since debuting, with companies like Samsung, Arm, and SiFive joining. This breadth makes NVLink Fusion the default standard for AI silicon interconnects due to Nvidia’s software ecosystem, despite the alternative UALink’s struggles due to diverse member priorities and slower launch pace.
For Marvell, this partnership offers a way to meet customer demands for custom chips that work with existing Nvidia infrastructure. The silicon photonics aspect may prove essential soon as traditional copper interconnects hit bandwidth and energy limits, making optical solutions more viable despite manufacturing challenges. The 5G/6G aspect implies a broader ambition of influencing infrastructure beyond data centers.
Nvidia holds about 90% of the data center GPU market, with the semiconductor industry growing rapidly, which aligns with an accelerating AI market that demands immediate, functional hardware. This urgency benefits Nvidia and highlights NVLink Fusion’s sales pitch.
While $2 billion is minor on Nvidia’s accounts, it secures architectural dominance, ensuring even chips intended to replace Nvidia’s are made within its ecosystem. It’s a strategic toll on the path to the tech industry’s fastest-growing market.