Nvidia CEO Jensen Huang demonstrated remarkable foresight by directing the company to begin developing specialized AI chips as early as 2010, long before the current wave of AI excitement. A comparable strategic decision in 2020—to aggressively expand its data center networking capabilities through a major acquisition—has since cultivated one of its most profitable and rapidly expanding divisions, albeit with far less public attention.

In a short span, Nvidia's networking unit, which focuses on interconnecting data centers, has risen to become the company's second-largest source of revenue, trailing only its core compute business. The division reported $11 billion in revenue for the last quarter, marking a staggering 267% increase from the previous year, and generated over $31 billion for the full fiscal year according to the latest earnings report.

EMBED_PLACEHOLDER_0

Fueled by the expansion of AI processing, this segment encompasses technologies such as NVLink for GPU communication within data center racks, Nvidia InfiniBand Switches, the in-network computing platform Spectrum-X for AI networking, and co-packaged optics switches. Collectively, these components provide the essential infrastructure for constructing what Nvidia terms an "AI factory"—a data center optimized for training AI models.

An analyst noted that the division's $11 billion quarterly revenue now surpasses the entire networking business of a major competitor like Cisco, achieving in three months what Cisco's segment does in a year. Despite this, the networking business operates somewhat in the shadow of Nvidia's significantly larger chip division and even its original gaming business, which is now nearly three times smaller.

The foundation of this success was laid with the $7 billion acquisition of Mellanox, an Israeli networking company founded in 1999. Kevin Deierling, Nvidia's Senior Vice President of Networking who joined via this acquisition, recalls CEO Jensen Huang's vision from day one: "the data center is the new unit of computing."

Deierling explained that networking is often misunderstood as merely connecting simple devices. In reality, it forms the critical backbone for modern AI infrastructure. He admitted that while the strategic rationale for acquiring Mellanox in 2020 wasn't immediately clear to him, its value is now evident. Integrating networking with its GPU business allows Nvidia to offer a complete, optimized package.

An industry analyst reinforced this, stating that Huang identified networking as the missing component to create a fully integrated GPU solution. Deierling attributes part of their success to a unique go-to-market strategy: selling only complete, full-stack solutions through partners rather than individual components. "We build the fully integrated stack and go to market through all our partners," he said, highlighting their distinctive position in the industry.

Nvidia recently unveiled a series of major updates to its networking ecosystem during Huang's keynote at the annual GTC conference on March 16. The announcements included the new Rubin platform, featuring six chips designed for an "AI supercomputer," a novel Inference Context Memory Storage platform, and more efficient Spectrum-X Ethernet Photonics switches.

Reflecting on the evolution, Deierling concluded that networking has transformed from a peripheral function into a fundamental component of computing. "In the old days, we had what was called the backplane inside the computer. Today, the network is the backplane of the AI factory, and it's super important."