Colette Kress Reveals Why 90% of Nvidia Customers Are Buying Beyond GPUs

At CES 2026, Nvidia’s Chief Financial Officer Colette Kress shared a striking insight into the company’s business evolution: nearly 90% of customers purchasing full AI systems are also buying networking products. This networking attachment rate illuminates a fundamental shift in how Nvidia is positioning itself—not just as a GPU supplier, but as an end-to-end AI infrastructure provider. The discovery underscores that networking has quietly become one of the company’s most powerful growth levers, even as competition in the GPU market intensifies.

The Networking Attachment Rate That’s Redefining Customer Relationships

Colette Kress’s disclosure about the 90% attachment rate signals a dramatic transformation in customer purchasing behavior. This metric reflects something deeper than simple product bundling: it reveals that organizations building large-scale AI data centers have recognized networking as inseparable from compute. Kress also noted that even companies deploying their own custom AI chips still frequently adopt Nvidia’s networking solutions, demonstrating the critical nature of these interconnected systems.

In the third quarter of fiscal 2026, this trend manifested in financial results: Nvidia generated $8.2 billion in networking revenue, representing a 162% year-over-year surge. This category encompasses NVLink (which interconnects GPUs), InfiniBand switches, and the Spectrum-X Ethernet networking platform. Major AI infrastructure builders including Meta, Microsoft, Oracle, and xAI have all committed to using Spectrum-X Ethernet switches in their next-generation data centers.

Why AI Infrastructure Demands Different Networking Approaches

The networking requirements for AI data centers differ fundamentally from traditional cloud data center architectures. During AI training workloads, data throughput between GPUs must remain exceptionally high—any GPU idling represents wasted computational capacity. Similarly, AI inference workloads depend on rapid data movement to deliver results with minimal latency. Standard enterprise networking, designed for different traffic patterns, cannot meet these demands. This technical necessity has positioned Nvidia’s specialized networking portfolio as irreplaceable within cutting-edge AI deployments.

The market is validating this specialization. According to industry analysts at IDC, revenues for ultra-high-speed 800GbE switches nearly doubled sequentially during the third quarter of 2025. Nvidia now commands an 11.6% share of the data center Ethernet switching market, positioning it behind only Arista Networks and Cisco Systems. This market standing reflects rapid market share gains in a segment that’s essential to the AI infrastructure buildout.

Rubin Platform: Bundling GPUs, Processors, and Networking at Scale

Nvidia’s latest strategic move extends this networking-first mindset. The company unveiled its Rubin platform at CES, which integrates GPUs, CPUs, and multiple networking technologies into rack-scale systems available in 8-GPU and 72-GPU configurations. The Vera Rubin NVL72 represents the most ambitious of these solutions, designed for deployment across all major cloud service providers in 2026.

The Rubin architecture features the new Spectrum-6 series of Ethernet switches, which support 800 GB/s per-port connectivity and deliver up to 102.4 Tb/s of total switching capacity. By shifting its commercial focus from selling individual GPUs to selling fully integrated rack-scale systems, Nvidia has positioned itself to capture networking revenue alongside compute revenue. This bundling approach effectively locks in the 90% attachment rate Colette Kress described, creating a structural advantage in customer relationships.

AI Networking Market Expansion: From $14.9B to $46.8B by 2029

The financial opportunity justifies Nvidia’s networking investments. Market researchers at MarketsandMarkets project the AI networking market will expand from $14.9 billion in 2025 to $46.8 billion by 2029, representing a compound annual growth rate of 33.8%. Even if competitive pressures erode Nvidia’s GPU market share over time, the networking business offers a parallel growth engine with comparable momentum.

This growth rate reflects the massive capital deployment underway in AI infrastructure. McKinsey research suggests that global data center capital spending must approach $7 trillion by 2030 to accommodate projected AI demand. A substantial portion of this spending will flow directly to vendors like Nvidia that can supply both compute and network infrastructure comprehensively.

Assessing the Investment Case: Opportunity Versus Uncertainty

The trajectory outlined by Colette Kress and reflected in market projections presents a compelling bull case for Nvidia’s long-term potential. The company controls essential infrastructure that AI developers cannot avoid purchasing, and its networking business provides revenue diversification beyond the GPU market.

However, significant uncertainties remain. Predicting AI demand four years into the future involves considerable speculation. Additionally, the pattern of previous capital-intensive industries suggests the possibility of overcapacity—scenarios where infrastructure investment outpaces actual demand. If data center overbuild occurs, demand for Nvidia’s AI systems could face meaningful headwinds, regardless of the company’s current market dominance.

The technology underlying AI will undoubtedly persist and drive productivity gains. Whether these gains translate into sustained profits for Nvidia shareholders requires navigating cyclical market dynamics and competitive erosion that remain unpredictable. The networking story Colette Kress emphasized is genuinely compelling, yet it doesn’t insulate the company from broader industry cycles or technological surprises.