When Nvidia CEO Jensen Huang encouraged the company to begin experimenting with creating AIspecific processors in 2010more than ten years before the current hype around AIhe was years ahead of the market.With little fanfare, a similar move in 2020doubling down on data center networking through a strategic acquisition—led to the creation of one of the company’s most profitable and rapidly expanding divisions.
Nvidia’s networking division, which was created to link data centers, has become the company’s secondlargest revenue generator in a matter of years, after compute.According to Nvidia’s most recent reports, it raked in more than $31 billion for the entire year and reported $11 billion in revenue last quarter, a 267% rise from the previous year.
The division contains technologies such as NVLink, which facilitates communication between GPUs on a data center rack; Nvidia InfiniBand Switches, an innetwork computing platform; SpectrumX, the ethernet platform for AI networking; and copackaged optical switches, among others, driven by the growth in AI processing.
All of the technology required to construct a “AI factory,” a data center intended for AI model training, is included in Nvidia’s networking division.
Nvidia’s networking division is one of the company’s most outstanding new business sectors, according to Kevin Cook, senior equities strategist at Zacks Investment Research, who spoke with TechCrunch.Cook stated, [Nvidia’s networking business] reports $11 billion for the quarter; that amount is bigger than Cisco’s networking business, almost as big as the fullyear expectations.” He added that Nvidia’s networking company accomplishes what Cisco’s business does annually in just one quarter.
However, the company’s far larger chip division receives more attention than the business segment.Additionally, it receives less attention than the company’s gaming division, which is almost three times smaller.
Mellanox, a networking firm established in Israel in 1999 that Nvidia purchased for $7 billion in 2020, is where Nvidia’s networking business got its start.
Nvidia’s senior vice president of networking is Kevin Deierling.He became a member of the corporation after Mellanox was acquired.Deierling told TechCrunch that his poor marketing of Nvidia’s networking division may be the reason why many are unaware of it, but he does not appreciate that response.
According to Deierling, “people conceive of networking as just, ‘I have a printer, and I need to connect to it.Jensen stated that the data center is the new computer unit on the day he acquired us.Networking is a foundation, not only a means of transferring tiny amounts of data between computing nodes.
Deierling claimed he did not fully get Huang’s decision to purchase the business at the time, but he now understands.The company is able to offer its chips with the technology that works best for them because it has a networking business in addition to its GPU business.
Jensen realized it was the final component needed to complete GPUs when he acquired Mellanox in 2020,” stated Cook, the Zack’s analyst.
Deierling went on to say that he believes another factor contributing to Nvidia’s networking success is that it only offers the technology as a full-stack solution rather than separate parts, and it sells the technology through its partners rather than directly.
Deierling remarked, “I can not think of other organizations that have [the] full-stack skills that we have.” “We truly differ from one another. After developing the whole computing stack and completely integrated stack, we use all of our partners to launch the product.
During Huang’s keynote speech at the company’s annual Nvidia GTC technology conference on March 16, Nvidia recently revealed a plethora of upgrades to its networking architecture.In order to power a “AI supercomputer,” the company introduced the Nvidia Rubin platform, which consists of six new CPUs.Along with other items, Nvidia also revealed more effective Nvidia SpectrumX Ethernet Photonics switches and a new Nvidia Inference Context Memory Storage platform.
Regarding networking, Deierling stated, “It is no longer a peripheral to connect the printer or any other slow I/O device.” It is essential to the computer. Back then, the computer had what was known as the back lining. The network is now the backbone of the AI factory and is crucial.
Leave feedback about this