Nvidia is quietly building a multibillion-dollar behemoth to rival its chips business

Nvidia's CEO Jensen Huang took a significant step more than a decade ago by initiating the development of AI-specific chips, well before the current surge in artificial intelligence interest. In 2020, he made another strategic move by emphasizing data center networking through a major acquisition, which has since evolved into a vital revenue stream for the company, largely under the radar. In recent months, Nvidia's networking division has become the second-largest contributor to its revenue, following its core computing sector. Last quarter alone, this division generated a staggering $11 billion, marking a remarkable 267% increase year-over-year, and contributing over $31 billion to the annual revenue, as stated in Nvidia's latest earnings report. This growth is fueled by the escalating demand for AI processing technologies, which encompass NVLink for GPU communication, Nvidia InfiniBand Switches for in-network computing, Spectrum-X for AI networking, and co-packaged optics switches. Collectively, these innovations are integral to constructing what is termed an “AI factory”—a data center specifically designed for training AI models. Kevin Cook, a senior equity strategist at Zacks Investment Research, highlighted the significance of Nvidia's networking segment, noting its impressive quarterly earnings that surpass Cisco's entire networking business for the same period. Despite these accomplishments, Nvidia's networking division has not garnered the same level of attention as its more expansive chip and gaming sectors. The roots of Nvidia's networking capabilities trace back to its acquisition of Mellanox, a networking firm established in Israel in 1999, which Nvidia purchased for $7 billion in 2020. Kevin Deierling, who joined Nvidia through this acquisition, expressed that the lesser-known status of the networking business could be attributed to insufficient marketing efforts but emphasized that the scope of networking extends far beyond simple connectivity tasks. Deierling explained that Huang recognized the necessity of integrating networking with GPU technology to create a comprehensive solution. According to Cook, this acquisition was pivotal in transforming GPUs into a complete package, facilitating the sale of chips alongside the optimal networking technology. Nvidia recently unveiled a series of updates to its networking solutions at the Nvidia GTC technology conference, including the launch of the Nvidia Rubin platform, which features six new chips aimed at supporting an “AI supercomputer.” Other advancements include the Nvidia Inference Context Memory Storage platform and upgraded Spectrum-X Ethernet Photonics switches. Deierling articulated the evolving perception of networking within computing, asserting its foundational role in the AI landscape. He remarked, “Networking is no longer just a means to connect devices; it has become the backbone of the AI factory.”