Nvidia Built a Networking Giant While You Watched the GPU Business
Everyone knows Nvidia for its GPUs. Fewer people know that Nvidia’s networking business generated $11 billion in revenue last quarter, a 267% year-over-year increase. That is more than Cisco’s networking business does in a year, compressed into a single quarter. The division includes NVLink, InfiniBand switches, the Spectrum-X ethernet platform, and co-packaged optics, everything needed to build what Nvidia calls an “AI factory.” This is the company’s second-largest revenue driver, and it gets a fraction of the attention.
Nvidia Networking Division at a Glance
- $11 billion in Q4 revenue, up 267% year over year
- Full-year networking revenue exceeded $31 billion
- Products include NVLink, InfiniBand, Spectrum-X ethernet, and co-packaged optics
- Originated from Nvidia’s $7 billion acquisition of Mellanox in 2020
- Nearly three times larger than Nvidia’s original gaming business
How Mellanox Became a Multibillion-Dollar Division
The origin of Nvidia’s networking business traces back to Mellanox, an Israeli networking company founded in 1999. Nvidia acquired it in 2020 for $7 billion. At the time, analysts saw it as a smart infrastructure play. Few predicted it would grow into a business that rivals the GPU division in growth rate.
Jensen Huang’s bet on networking mirrors his earlier bet on AI chips in 2010, more than a decade before the current AI wave. In both cases, he moved years ahead of the market.
“Nvidia’s networking business reports $11 billion for the quarter. That number is greater than Cisco’s networking business, almost as big as the full-year estimates,” said Kevin Cook, senior equity strategist at Zacks Investment Research.
Why Networking Matters for AI Infrastructure
GPUs get the attention because they do the computation. But in a data center with thousands of GPUs, the network that connects them determines overall performance. Latency between chips, bandwidth for data transfer, and the ability for every GPU to communicate with every other GPU directly affect training speed and inference cost.
NVLink allows GPU-to-GPU communication within a server rack. InfiniBand provides high-bandwidth, low-latency connections between racks. Spectrum-X handles ethernet networking at AI scale. Together, these products form the connective tissue of modern AI infrastructure.
The Quiet Advantage
Networking is less glamorous than chips or gaming, which explains why it gets less press coverage. But for Nvidia, it creates a powerful lock-in effect. Customers who buy Nvidia GPUs also need Nvidia networking to connect them. That bundling makes it harder for competitors like AMD or Intel to compete on chips alone, because they cannot match the full-stack offering.
With $31 billion in annual networking revenue and growth accelerating, this is no longer a side business. It is a core pillar of Nvidia’s AI strategy, and one that investors and competitors should be watching as closely as the GPU roadmap.
