The world breathed a sigh of relief this week after Nvidia’s earnings came very much in the green, exceeding analyst expectations with $57.01 billion in total revenues.

Nvidia quashed some fears that the AI bubble was slowing, a concern given that two of its biggest earlier backers in Peter Thiel and SoftBank, opted to sell shares in the weeks prior. But the chip giant came out swinging with a Q3 report that, in the words of Futurum CEO Daniel Newman, saw Nvidia “throw the fear out the window.”

While all eyes were on its compute segment – which grew 56% year over year (YoY) despite “geopolitical issues” impacting some sizable purchase orders that ended up never materializing – networking yet again helped to drive growth for the company.

Nvidia’s networking revenue rose to $8.2 billion, representing a 162% YoY increase driven by demand for NVLink, InfiniBand, and Spectrum X solutions.

CFO Colette Kress told investors that its networking segment “more than doubled,” with the vendor seeing robust double-digit growth across its Ethernet and InfiniBand fabric products.

According to Kress, demand for Nvidia’s Ethernet networking technologies is now “roughly on par with InfiniBand.”

Nvidia only really embraced Ethernet through the launch of its Spectrum-X fabric in 2024, having heavily leaned on its rival protocol before and following its 2019 acquisition of Mellanox.

In just a year then, demand for its Ethernet solutions for AI workloads is so intense that it’s now at the same level as a technology it’s worked on for years.

“Our networking business purpose-built for AI, [is] now the largest in the world,” Kress said. “With NVLink, InfiniBind, and Spectrum X Ethernet, all contributing to growth, we are winning in data center networking as the majority of AI deployments now include our switches.”

Inference helping drive inference demand

With the eyes of the tech world drawn to shiny GPUs, networking has quietly and quickly become a growth driver for Nvidia.

Its Q2 report in August saw the segment post revenues of $7.3 billion, the same month it brought its networking technologies to data center scale-out with Spectrum-XGS, an Ethernet-based platform capable of interconnecting workloads across multiple data centers.

Nvidia’s latest earnings saw the tech giant push past concerns, and in some analysts' view, its rise shows little to no signs of stopping.

“We haven’t even seen the beginning of the enterprise AI wave and the sovereign clouds,” Futurum’s Newman said. “We expect the forecasts for AI infrastructure continuing to tick up. Likely the 2030 number of $1 trillion will see upward revisions.”

According to Nvidia CEO Jensen Huang, networking is helping to drive demand for its tech from companies wanting to run inference workloads.

The firm’s rack-scale Blackwell systems topped a recent AI inference benchmark from SemiAnalysis, with networking technologies including its NVLink, NVLink Switch, and the NVL72’s rack-scale networking fabric helping to achieve performance levels that outpaced rival systems from the likes of AMD.

“Our hope is that inference is a very large part of the market,” Huang told investors. “Because if inference is large, then what it suggests is that people are using it in more applications, and they're using it more frequently. And this is where Grace Blackwell is just an order of magnitude better, more advanced than anything in the world.

“The scale-up network that we have achieved … it's going to take a long time before somebody is able to take that on. And our leadership there is surely multiyear.”