Nvidia's “most underrated” business is taking off like a “rocket ship”

When NVIDIA (NVDA) reports its second quarter revenue on August 27th, investors will focus head-on on the company's data center results. After all, that's where chip giants make money on selling high-power AI processors.

However, the data center segment includes more than just chip sales. It also explains some of the most important things about Nvidia, but is often overlooked.

Nvidia's networking products, consisting of NVLink, Infiniband and Ethernet Solutions, allow chips to communicate with each other, servers to discuss with each other within a large data center, and ultimately allow end users to connect to everything to run AI applications.

“The most important part of building a supercomputer is the infrastructure. The most important part is how these computing engines can be linked together to form a larger unit of computing.”

Paris, France - June 11: Jensen Huang, co-founder and CEO of Nvidia Corp., will be attending the 9th edition of the Vivedic trade show held at the Parc des Expositions de la Porte de Versailles in Paris on June 11, 2025. Vivedch, Europe's largest tech trade show, offers a unique digital format for four days of reconnection and recovery through innovation. The event brings together startups, CEOs, investors, technology leaders, and all digital transformation players shaping the future of the Internet. Founded in 2016 by Publicis Groupe and Groupe Les Echos, this annual technology conference, also known as Vivatech, is dedicated to innovation and promoting startups. — Nvidia CEO Jensen Huang will be attending the 9th edition of the Vivedic trade show held at the Parc des expositions de la Porte de Versailles in Paris on June 11, 2025. (Photo: Chesnot/Getty Images) ・ Chesnot via Getty Images

It also leads to some big sales. Nvidia's networking sales accounted for $12.9 billion of data center revenue in the previous fiscal year. While that may seem unimpressive, considering that chip sales have led to $1002.1 billion, it has overturned the $11.3 billion won this year by the game, the second-largest segment in Nvidia.

In the first quarter, Networking compensated for NVIDIA's $4.9 billion data center revenue. And it continues to grow as customers continue to build AI capabilities, whether they are research universities or large data centers.

“This is the most underrated part of Nvidia's business. Absolutely orders of magnitude,” Gene Munster, managing partner at Deepwater Asset Management, told Yahoo Finance. “Basically, networking is 11% of revenue, so it doesn't attract attention. But it's growing like a rocket ship.”

Speaking of AI explosions, Nvidia's Senior Vice President of Networking Kevin Deierling said the company will have to work across three different types of networks. The first is NVLink technology that allows GPUs to connect to each other within multiple servers in a server rack, such as a GPU in a server or cabinet, allowing them to communicate and improve overall performance.

Next is Infiniband. It connects multiple server nodes between data centers to form an essentially large AI computer. Next is the front-end network for storage and system management that uses Ethernet connectivity.

Nvidia CEO Jensen Huang will be introducing Grace Blackwell NVLink72 when he gave a keynote address at the Conuseent Electronics Show (CES) in Las Vegas, Nevada on January 6, 2025. Donald Trump. The annual Consumer Electronics Show (CES) will officially open in Las Vegas on January 7th, 2025, but the day before is packed with product announcements. (Photo by Patrick T. Fallon / AFP) (Photo by Patrick T. Fallon / AFP — Nvidia CEO Jensen Huang will introduce Grace Blackwell NVLink72 when he gave a keynote address at the Conusher Electronics Show (CES) in Las Vegas, Nevada on January 6, 2025 (Photos) ・ Patrick T. Fallon via Getty Images

“All these three networks are needed to build huge AI scales, or medium-sized enterprise-scale AI computers,” explained Deierling.

However, the purpose of all these different connections is not just to help chips and servers communicate. They are also intended to allow it to be done as soon as possible. If you are trying to run a set of servers as a single computing, you need to talk to each other in a flash.

Lack of data going to the GPU slows down the entire operation, slows down other processes, and affects the overall efficiency of the entire data center.

“[Nvidia is a] Very different businesses without networking,” Munster explained. [are] If it's not for their networking then what you want won't happen. “

It also ensures that these GPUs operate in lockstep with each other as companies continue to develop larger AI models and autonomous and semi-autonomous agent AI capabilities, allowing users to perform tasks.

This is especially true as an inference that requires a more powerful data center system when running AI models.

The AI industry is widely sorted around reasoning ideas. At the start of the AI explosion, the idea was that training AI models requires very powerful AI computers, and actually running them was somewhat less power intensive.

This sparked a horror on Wall Street earlier this year. This is when Deepseek claimed to have trained AI models below the top-level Nvidia chip. The idea at the time was that if companies could train and run AI models on inadequate chips, then Nvidia's expensive high power systems wouldn't be needed.

But the story quickly turned over as chip companies pointed out that the same AI model could benefit from running on powerful AI computers and could infer more quickly than when running on systems that were too unwise.

“I think there's still a misconception that reasoning is trivial and easy,” Deierling said.

“We see that as we get there it's starting to look more and more like training. [an] Agent workflow. So all of these networks are important. Have them together and tightly coupled to the CPU, GPU and DPU [data processing unit]it is all crucial to make reasoning a good experience. ”

However, Nvidia's rivals are turning. AMD is looking to gain more market share from the company, and Cloud Giants, such as Amazon, Google and Microsoft, continue to develop their own AI chips.

Industry groups also have their own competing networking technologies, including Ualink. This is intended to be a head-to-head match against NVLink, explained Forrester analyst Alvin Nguyen.

But for now, Nvidia continues to lead the pack. And as tech giants, researchers and companies continue their fight for Nvidia's chips, it ensures that the company's networking business will continue to grow too.