Why Jensen Huang and Matt Murphy Shared the Stage at Computex 2026

The event signals more than a friendly business relationship. It reflects a fundamental truth about the next phase of artificial intelligence: the ability to move data efficiently across accelerators, racks, and entire AI factories is now as strategically important as the computing power of the accelerators themselves.

The $2 Billion Partnership That Set the Stage

On March 31, 2026, Nvidia and Marvell announced a strategic partnership centered on Nvidia's NVLink Fusion platform, a rack-scale architecture designed to let customers build semi-custom AI infrastructure using both Nvidia and third-party components . The deal was described in multiple reports as including a $2 billion investment by Nvidia in Marvell .

Under the partnership, Marvell will contribute custom XPUs and NVLink Fusion-compatible scale-up networking, as well as advanced optical interconnect and silicon photonics technology . The collaboration also extends to Nvidia's AI factory and AI-RAN ecosystem, signaling ambitions beyond traditional hyperscale data centers and into telecommunications infrastructure .

The arrangement is a major departure from the traditional semiconductor model of transactional supplier relationships. It is a bet on deep co-architecture, where two major chip companies integrate their roadmaps in pursuit of a shared networking vision.

Why Connectivity Became the Center of Gravity

For years, AI infrastructure conversations centered on GPUs, floating-point performance, and memory bandwidth. But as AI models and training clusters have scaled to hundreds of thousands of accelerators, the bottleneck has shifted. Marvell itself stated the case bluntly in early 2026: "The primary bottleneck in AI data center infrastructure has shifted from compute to connectivity" .

Next-generation AI workloads demand ultra-high-bandwidth, low-latency interconnects that can handle massive data flows across an increasingly heterogeneous mix of processors, accelerators, and memory pools . Traditional copper-based interconnects face physical limits at these scales, making optical and silicon photonics solutions essential for the rack-scale and cluster-scale fabrics that large AI deployments require .

This is where Marvell's portfolio fits into Nvidia's strategy. Marvell brings custom silicon design capabilities, application-specific integrated circuits, and a deep portfolio of data center connectivity products that address the exact gaps in Nvidia's own ecosystem .

What the Computex Keynote Signals About Nvidia's Strategy

Jensen Huang's decision to appear as a guest during a partner's keynote—rather than simply highlighting Marvell during his own separate Computex address on June 1—is a deliberate signal . It emphasizes that Nvidia is choosing a partner-heavy approach to solve the networking layer of AI infrastructure, rather than treating it as a purely internal problem .

The event's official title, "The Future of AI Depends on Connectivity," is itself a statement of strategic intent . It frames the partnership not as a niche technical collaboration, but as an answer to the defining infrastructure challenge of the next AI era.

Nvidia is positioning NVLink Fusion as the integration point for an expanding ecosystem of custom and semi-custom silicon that will connect to its AI factories. The Marvell partnership is the most visible example, but the platform is designed to accommodate additional partners over time .

The Bigger Picture: A Fabric-Centric AI Infrastructure Era

The Nvidia-Marvell partnership is not an isolated deal. It is a leading indicator of several converging trends reshaping how the semiconductor industry builds for AI:

• Infrastructure design is pivoting from GPU-centric to fabric-centric. The network connecting accelerators—its bandwidth, latency, and scalability—is becoming a primary design constraint, not an afterthought .
• Deep capital and technical integration between chip companies is replacing traditional arms-length vendor relationships. A $2 billion equity investment paired with co-development signals a long-term bet on shared architecture, not a short-term procurement agreement .
• Custom silicon for connectivity is emerging as a first-class investment area alongside compute, driven by the need for optical interconnects, custom XPUs, and NVLink Fusion-compatible infrastructure .
• The AI infrastructure stack is opening to semi-custom designs. NVLink Fusion is designed to let hyperscale customers and cloud providers build differentiated infrastructure without leaving Nvidia's software and interconnect ecosystem entirely .

For AI practitioners and industry watchers, the Computex 2026 joint appearance is more than a keynote scheduling detail. It is the moment when the networking layer of the AI data center was publicly elevated to the same strategic tier as compute—and when Nvidia made clear that it would secure that layer through deep partnerships, not solo engineering.

"The network is the computer" has rarely been more literal for AI—and the Huang-Murphy stage share at Computex is as clear a signal as any that connectivity has entered the strategic spotlight .