Nvidia's networking business overtook Cisco in quarterly revenue, with the company posting more than $17.5 billion in networking sales for the second quarter, exceeding Cisco's $17.3 billion across all products. The achievement marks Nvidia's rise as a dominant player in data centre networking, driven by its acquisition of Mellanox and deep integration of compute and networking for AI workloads.
Nvidia's networking revenue growth is built on its 2019 acquisition of Mellanox for $6.9 billion. At the time, Mellanox was primarily known for its ConnectX NICs and InfiniBand switches. InfiniBand was used in high-performance computing and supercomputing clusters, as well as in a big deployment at Microsoft. The purchase allowed Nvidia to position itself as a system builder, as customers wanted clusters rather than GPUs or even servers.
This co-design approach, designing components and software to work together for specific tasks, has been central to Nvidia's strategy. Dell'Oro Group analyst Sameh Boujelbene told The Register that customers choose Nvidia for its tightly integrated compute and networking architecture, optimised for AI workloads. The integration makes deployments easier and speeds time to market.
Ben Mabey, CTO of Recursion, a company applying machine learning to drug discovery, said the integration won him over. Recursion's first cluster used Ethernet and hit networking limits, so its second cluster used Nvidia's InfiniBand switches and ConnectX NICs. Mabey told The Register that buying compute and interconnects from the same vendor vastly simplified optimisation and troubleshooting. This level of integration had previously been available only from boutique supercomputer builders like Cray or Atos.
Networking revenue growth accelerated with the Blackwell platform and the introduction of NVSwitch trays for the GB200 NVL72 system, which enabled six dozen GPUs to function as a single unit. According to Boujelbene, NVSwitch now accounts for roughly one-third of Nvidia's networking revenue, up from virtually zero less than two years ago. Colossus was one of the earliest examples of a cluster using Nvidia's Spectrum-X Ethernet switches; Meta has also made extensive use of Nvidia's Spectrum switches in its training clusters.
Nvidia is now expanding its networking reach through NVLink Fusion, a licensing program that opens its interconnect technology to third-party chip designers. Companies including Marvell, Qualcomm, Arm, Fujitsu, MediaTek, and d-Matrix have announced plans to develop chips using NVLink Fusion. Boujelbene told The Register that this allows Nvidia to extend its networking footprint into the custom accelerator market, and described it as a strategic move to capture a larger share of AI networking spending even in clusters that do not use Nvidia GPUs.
Analysis
Why This Matters
- Nvidia's networking revenue surpassing Cisco's signals a shift in data centre infrastructure spending toward integrated AI systems.
- The company's tight coupling of compute and networking could set a new standard, affecting competitors like Broadcom, Cisco, and AMD.
- NVLink Fusion licensing may reshape the custom accelerator market, potentially locking in Nvidia's interconnect as a standard.
Background
Nvidia began as a graphics card manufacturer and later became a leader in AI hardware. In 2019, it acquired Mellanox for $6.9 billion, entering the networking market. Since then, its networking business has grown steadily, boosted by the AI boom sparked by ChatGPT in 2023 and the Blackwell platform launch in 2024-2025.
Key Perspectives
Dell'Oro Group analyst Sameh Boujelbene: Says Nvidia's tightly integrated compute and networking architecture optimised for AI workloads is why customers choose Nvidia, and notes that NVSwitch has become a major growth engine.
Ben Mabey, CTO of Recursion: Describes how buying compute and networking from the same vendor simplified optimisation and troubleshooting, a benefit previously limited to boutique supercomputer builders.
What to Watch
- The adoption rate of NVLink Fusion among third-party chip designers and whether it becomes a standard interconnect for AI accelerators.
- How competitors like Broadcom and Cisco respond to Nvidia's networking growth, particularly in Ethernet switching.
- The performance of Nvidia's networking business as the AI infrastructure buildout continues and new networking requirements for agentic AI emerge.
Checked
12 claims in this story checked against 1 source: 12 supported.
Show each claim
- SupportedNvidia's networking revenue topped $17.5 billion in the second quarter.www.theregister.com - Articles: “In Q2, Nvidia's networking revenues topped $17.5 billion.”
- SupportedCisco's revenue across all products was $17.3 billion in the same period.www.theregister.com - Articles: “Cisco's revenue during that same period, across all products, was $17.3 billion.”
- SupportedNvidia's networking revenue exceeded Cisco's total revenue for the quarter.www.theregister.com - Articles: “In Q2, Nvidia's networking revenues topped $17.5 billion. To put that in perspective, Cisco's revenue during that same period, across all products, was $17.3 billion.”
- SupportedNvidia's networking growth is associated with AI workloads and integration of compute and networking.www.theregister.com - Articles: “Customers choose Nvidia for its tightly integrated compute and networking architecture, optimized for AI workloads”
- SupportedNvidia's rise in data-centre networking makes it a dominant player.www.theregister.com - Articles: “It has also become a titan of the networking industry with bigger datacenter Ethernet sales than any other vendor.”
- SupportedNvidia acquired Mellanox in 2019 for $6.9 billion.www.theregister.com - Articles: “Jensen Huang started working towards this status in 2019 with the $6.9 billion acquisition of long-time networking partner Mellanox .”
- SupportedMellanox was primarily known for its ConnectX NICs and InfiniBand switches.www.theregister.com - Articles: “The company was mostly known for its ConnectX NICs and InfiniBand switches.”
- SupportedNvidia's customers wanted clusters rather than just GPUs or servers.www.theregister.com - Articles: “Nvidia understood its datacenter customers weren't interested in buying GPUs or even servers; they wanted clusters of systems.”
- SupportedBoujelbene said customers choose Nvidia for tightly integrated compute and networking architecture optimised for AI workloads.www.theregister.com - Articles: “"Customers choose Nvidia for its tightly integrated compute and networking architecture, optimized for AI workloads," she told The Register.”
- SupportedBoujelbene estimated NVSwitch accounts for roughly one-third of Nvidia's networking revenue, up from virtually zero less than two years earlier.www.theregister.com - Articles: “"We estimate that NVSwitch now accounts for roughly one-third of Nvidia’s networking revenue, up from virtually zero less than two years ago," Boujelbene told El Reg.”
- SupportedColossus was one of the earliest clusters to use Nvidia Spectrum-X Ethernet switches.www.theregister.com - Articles: “Elon Musk's Colossus supercomputer was one of the earliest examples. As we reported at the time, the first half of the 200,000-chip H100 and H200-based cluster came online in 2024 and, at the time, was among the largest ever clusters deployed using Nvidia's Spectrum-X Ethernet switches.”
- SupportedCompanies including Marvell, Qualcomm, Arm, Fujitsu, MediaTek and d-Matrix announced plans to develop chips using NVLink Fusion.www.theregister.com - Articles: “Many chip designers, including Marvell, Qualcomm, Arm, Fujitsu, MediaTek, and d-Matrix, have announced plans to develop CPUs and XPUs using NVLink Fusion.”