USA

Nvidia’s ‘most underappreciated’ business is taking off like a ‘rocket ship’

When NVIDIA (NVDA) declares their second quarter earnings on August 27, investors will focus on the company’s data center results. As a result, Chip Giant earns revenue from the sale of high -power AI processors.

However, the data center segment includes more than chip sales. In addition, NVIDIA’s most important, most of the time, although it is overlooked, offers: Network technologies explain.

It consists of it Nvlink– InfinibandAnd Ethernet Solutions, NVIDIA’s network products, their chips communicate with each other, allows the servers to talk to each other in the large data centers, and ultimately allows end users to connect to all of their AI applications.

Gilad Shainer, Senior Vice President of Networking in NVIDIA, said, “The most important part of creating a super computer is the infrastructure. The most important part is how you connect this information process to create a larger information process unit.”

Jensen Huang, NVIDIA CEO, participates in the 9th edition of Vivatech Trade Fair in Parc Des Expositions in Paris on 11 June 2025. (Photo: Chesnot/Getty Images) · Chesnot through Getty Images

This turns into some major sales. NVIDIA’s network sales constituted $ 12.9 billion of $ 115.1 billion data center income previous financial year. When you think that the CHIP sales bring $ 102.1 billion, it may not look impressive, but NVIDIA’s second largest segment of $ 11.3 billion of $ 11.3 billion received throughout the year.

In the first quarter, Networking created $ 4.9 billion of NVIDIA’s $ 39.1 billion data center revenue. In research universities or large data centers, customers will continue to grow as they continue to create their AI capacity.

“Nvidia’s work is the least unspecified part of Nvidia, with the orders of magnitude,” said Deepwater Asset Management Partner Gene Munter Yahoo Finance. “Basically, networking does not attract attention because 11% of the income is growing like a rocket ship.”

In the case of AI explosion, NVIDIA Network Senior Vice President Kevin Deierling said that the company should work on three different networks. The first is the NVLink technology that connects GPUs in a server, allows multiple servers in a tall, cabinet -like server shelf to transmit and increase overall performance.

Later, there is Indian Infiniband, which connects multiple server nodes among the data centers to create a large AI computer. Then, there is a front end network for storage and system management that uses the Ethernet connection.

NVIDIA CEO Jensen Huang offers a Grace Blackwell Nvlink72 on January 6, 2025, while offering an opening address at the consumer electronics show (CES) in Nevada, Nevada. By Donald Trump elected US President. Annual Consumer Electronics Fair (CES) officially opened in Las Vegas on January 7, 2025, but previous days are full of product announcements. (Photo: Patrick T. Fallon / AFP) (Photo: Patrick T. Fallon / AFP through Getty Images) — NVIDIA CEO, Jensen Huang, presents a Grace Blackwell Nvlink72 on January 6, 2025 at the Opening Speech of the Consumer Electronics Show (CES) in Nevada, Nevada. · Through Patrick T. Fallon Getty Images

“It is necessary to build AI computer with a giant AI -scale, even medium -sized operating scale of these three networks, Deierling said Deierling.

However, the aim of these various connections is not only to help communicate with chips and servers. They also aim to allow them to do as fast as possible. If you are trying to run a series of servers as a single information process, they need to talk to each other in the blink of an eye.

The lack of data to the GPUs slows down all operations, delays other operations and affects the overall efficiency of a data center.

“[Nvidia is a] Very different work without a network, Mun Munster explained. [are] Without the networks, it wouldn’t have been desire. “

And as companies continue to develop larger AI models and autonomous and semi -autonomous AI capabilities that can perform tasks for users, these GPUs work to lock each other.

This is especially valid because it requires more powerful data center systems – especially to run AI models.

The AI industry is in the midst of a wide range of sequence around the idea of inference. At the beginning of the AI explosion, thinking, educating AI models would require extremely powerful AI computers, and in fact, it would be slightly less intense to run them.

This led to some fears in Wall Street, in which Deepseek claimed to have trained AI models on the top Nvidia chips. If the thought at that time could train and operate AI models in insufficient weak chips, Nvidia’s expensive high -power systems did not need.

However, this narrative said that the same AI models benefit from working on strong AI computers and allow them to have faster information than working on less developed systems.

“I think there is still a misunderstanding that my inference is insignificant and easy, de said Deierling.

“Apparently because we started to look more [an] Agent workflow. So all these networks are important. Bring them together, CPU, GPU and DPU tightly dependent [data processing unit]All this is very important to make the inference a good experience. “

But Nvidia’s rivals are drawing apartments. AMD wants to get more market share from the company, and cloud giants such as Amazon, Google and Microsoft continue to improve their own AI chips.

Forrester analyst Alvin Nguyen announced that they have their own competitor network technologies, including Ualink, which aims to go head -to -head with industrial groups.

But for now, the Nvidia continues to manage the package. As technology giants, researchers and businesses continue to fight on the chips of Nvidia, the company’s network business is guaranteed that all the network business will continue to grow.