Nvidia has quietly built a networking division that now pulls in more quarterly revenue than Cisco’s entire networking business does in a year. According to TechCrunch AI, the segment reported $11 billion in revenue last quarter, a 267% year-over-year jump, and over $31 billion for the full fiscal year.
This isn’t a side project. It’s Nvidia’s second-largest revenue driver, right behind its compute (GPU) business. And almost nobody’s talking about it.
How It Started
The story begins with a $7 billion acquisition. In 2020, Nvidia bought Mellanox, an Israeli networking company founded in 1999. At the time, even Mellanox’s own people didn’t fully grasp CEO Jensen Huang’s vision.
“When Jensen bought Mellanox in 2020, he saw that was the missing piece to make GPUs a complete package,” Kevin Cook, senior equity strategist at Zacks Investment Research, told TechCrunch AI.
Huang’s bet mirrors the one he made in 2010 when he pushed Nvidia into AI-specific chips, more than a decade before the current AI wave. Both moves looked premature at the time. Both turned out to be prescient.
What the Division Actually Does
Nvidia’s networking business covers everything needed to wire up what Huang calls an “AI factory”, a data center purpose-built for training AI models. The product lineup includes:
- NVLink: handles GPU-to-GPU communication within a data center rack
- InfiniBand Switches: an in-network computing platform
- Spectrum-X: Nvidia’s ethernet platform designed specifically for AI workloads
- Co-packaged optics switches: next-gen interconnect technology
Kevin Deierling, Nvidia’s senior VP of networking (who joined through the Mellanox acquisition), explained the core idea: networking isn’t about connecting printers anymore. “The network is the back lining of the AI factory, and it’s super important,” Deierling said.
Why This Matters
The strategic advantage here is vertical integration. Nvidia doesn’t just sell GPUs; it sells the entire infrastructure stack those GPUs need to perform at scale. And it does it through partners, not direct sales.
“I can’t think of other companies that have full-stack capabilities that we have,” Deierling told TechCrunch AI. “We build the full compute stack, fully integrated stack, and then we go to market through all of our partners.”
This is significant for the broader AI industry. As model training becomes more compute-intensive, the bottleneck increasingly shifts from raw chip performance to how efficiently those chips communicate. Networking becomes the limiting factor, and Nvidia now owns that layer too.
For competitors like AMD and Intel who are trying to challenge Nvidia’s GPU dominance, this makes the hill even steeper. Beating Nvidia on chips alone isn’t enough when customers can get a fully integrated system from one vendor.
Fresh Announcements at GTC
Nvidia doubled down at its annual GTC conference on March 16, where Huang unveiled:
- The Nvidia Rubin platform with six new chips for building AI supercomputers
- A new Inference Context Memory Storage platform
- More efficient Spectrum-X Ethernet Photonics switches
Each announcement reinforces the same strategy: own more of the data center stack, make every piece work better together.
What Comes Next
Nvidia’s networking division grew 267% in a single year. That growth rate won’t last forever, but the structural advantage will. Every new AI data center built with Nvidia GPUs is a potential customer for Nvidia networking gear, and vice versa.
What stands out here is the timing. Huang made the Mellanox bet before most companies even understood they’d need AI-scale networking. Now that demand has arrived, Nvidia already has the product, the partnerships, and the integration story locked in.
For the full breakdown, check the original reporting from TechCrunch AI.