Nvidia Pours $2 Billion Into Marvell, Binding Custom Chip Designer to NVLink Ecosystem

Nvidia has committed $2 billion to acquiring a stake in Marvell Technology while simultaneously establishing a strategic partnership that integrates Marvell into Nvidia's NVLink Fusion platform. This move extends Nvidia's proprietary interconnect ecosystem to encompass one of the sector's most prominent custom silicon designers, alongside Broadcom as a peer in that space.

Marvell's client roster spans major cloud operators including Amazon Web Services, for which the company engineered the Trainium line of AI accelerators, as well as Microsoft and Google. These bespoke processors have been developed largely to offer hyperscalers a path away from dependence on Nvidia's GPU offerings, making Nvidia's financial stake in Marvell a strategically significant move.

Under the partnership framework, Marvell will deliver custom XPUs alongside NVLink Fusion-compatible networking infrastructure for scaling operations. Nvidia's contribution encompasses Vera CPUs, ConnectX network interface cards, Bluefield data processing units, NVLink interconnect technology, and Spectrum-X switching solutions. The companies have also committed to joint development efforts in silicon photonics and AI-RAN systems designed for 5G and 6G deployments.

The inference inflection has arrived. Token generation demand is surging, and the world is racing to build AI factories. Together with Marvell, we are enabling customers to leverage Nvidia's AI infrastructure ecosystem and scale to build specialized AI compute.

Jensen Huang, founder and CEO of Nvidia

Introduced in May of the previous year, NVLink Fusion represents a framework for heterogeneous AI systems in which accelerators from non-Nvidia sources can interact with Nvidia's GPUs, processors, and network components through a unified high-bandwidth, low-latency interconnect. Any platform utilizing this technology must incorporate a minimum of one Nvidia element—whether processor, accelerator, or switching hardware.

Marvell's role centers on supplying custom XPUs and optical interconnect solutions operating at high speeds. During its fiscal 2026 period, which concluded in January 2026, Marvell generated $8.2 billion in total revenue, with data center operations representing over 74 percent of that figure. The company's late-year acquisition of Celestial AI brought photonic fabric capabilities into its technology portfolio, and this partnership now positions that expertise within Nvidia's broader infrastructure framework.

By connecting Marvell's leadership in high-performance analog, optical DSP, silicon photonics and custom silicon to Nvidia's expanding AI ecosystem through NVLink Fusion, we are enabling customers to build scalable, efficient AI infrastructure.

Matt Murphy, chairman and CEO of Marvell

By incorporating Marvell into the NVLink Fusion framework, Nvidia establishes a structural dependency whereby custom processors developed by Marvell must maintain compatibility with and reliance upon Nvidia's infrastructure layer. Since every NVLink Fusion implementation mandates inclusion of at least one Nvidia component, Marvell's ASIC designs utilizing this fabric continue to generate revenue for Nvidia.

Marvell enters an NVLink Fusion ecosystem that has expanded progressively since its inception. Samsung Foundry became a participant in October, providing design-to-manufacturing capabilities for NVLink-compatible custom silicon. Arm joined the program in November, permitting its licensees to develop processors with integrated NVLink support. Nvidia's primary competitors—AMD, Intel, and Broadcom—remain outside this ecosystem and have instead aligned behind UALink, an open standard for rack-level interconnect infrastructure.

Source: Tom's Hardware