en.Wedoany.com Reported - When it comes to NVIDIA, many people think of their GPUs. Indeed, with its leading CUDA and hardware, NVIDIA's GPU has become the default choice in the AI market. Through the acquisition of Groq's intellectual property and technology, as well as investments in CPU technology, NVIDIA has gradually built a solid moat in computing power. However, beyond this, another type of chip from NVIDIA has quietly risen to dominance.
According to IDC data, NVIDIA became the top revenue-generating supplier of data center Ethernet switches in the first quarter of 2026. The chip strategy behind this achievement has played a crucial role.
The Quiet Rise of a "New Giant"
Due to the booming development of artificial intelligence and the construction of various data centers worldwide, the market has experienced significant growth. According to reports, the Ethernet switch market revenue reached $15.4 billion, a year-over-year increase of 39.8%. Revenue from hyperscale data centers and enterprise data centers alone reached $10 billion, a 61% increase compared to the same period in 2025.
From a market growth perspective, the Americas region led with a year-over-year growth rate of 49.7%, followed by the EMEA region with 32.2%, and the Asia-Pacific region ranked third with a 25.9% year-over-year growth rate.
NVIDIA, with $2.1 billion in revenue and a year-over-year growth of 192.7%, currently holds a 21.5% share of the data center Ethernet market. The key factor driving this revenue growth is NVIDIA's Spectrum-X platform, which provides end-to-end networking solutions, including BlueField DPUs and NVIDIA LinkX cables. These products are specifically designed for large-scale GPU clusters, which is the direction of today's AI development.
According to reports, through the integrated co-design of GPUs and networks, NVIDIA has met the demands of hyperscale data centers and enterprises for AI factory network infrastructure. This structural shift is reshaping the vendor landscape of the entire data center networking industry.
IDC further noted in the market that demand for 400G and 800G deployments remains strong and will continue to grow in the coming years. In the first quarter of 2026, 800G switches accounted for 35.8% of data center revenue share, while 200G and 400G switches held a 34.1% market share. These switches together accounted for 70% of the global data center Ethernet revenue share.
With its AI-optimized Spectrum-X platform, this success highlights NVIDIA's growing dominance across the entire AI infrastructure landscape, from GPUs and CPUs to high-performance networks. As data centers race to expand for the AI era, NVIDIA is proving that it is not only a leader in accelerators but also a trusted end-to-end partner for next-generation smart factories.
Thanks to this performance, NVIDIA has surpassed competitors like Arista Networks, which held a 20.7% share of the data center Ethernet switch market in the first quarter of this year. Other major players include Cisco, Huawei, and HPE.
From related statements, it is evident that this chip giant is increasingly focusing on networking technology, viewing it as a major growth driver. At a recent shareholder meeting, NVIDIA CEO Jensen Huang stated that Spectrum-X "is now larger than all other Ethernet networking products combined."
During the earnings call in May, NVIDIA CFO Colette Kress said that the company's broader data center networking revenue had tripled year-over-year to $15 billion. In comparison, in an earlier quarter, NVIDIA disclosed that its quarterly networking business revenue had approached $11 billion, a year-over-year increase of 263%.
This highlights the importance of the networking business in the company's revenue.
An Acquisition That Changed the Landscape
NVIDIA's success in the switch market is undoubtedly due to a key acquisition completed in 2019—acquiring Ethernet and InfiniBand interconnect giant Mellanox Technologies for approximately $6.9 billion.
At the time, this deal did not attract as much attention as later AI GPUs. Many believed NVIDIA was simply filling a gap in its networking product line. However, looking back today, it appears more like a strategic investment that would define NVIDIA's competitive landscape for the next decade.
Back then, in data centers, the GPU was still just an accelerator within a server, and the network primarily played the role of "connecting devices." However, as the scale of large model training continues to expand, with GPU counts growing from hundreds to thousands and tens of thousands, the true bottleneck limiting AI cluster efficiency is no longer just the GPU itself, but the data exchange capability between GPUs.
A widely circulated industry saying aptly captures this change: "In the AI era, GPUs determine the upper limit of computing power, while networks determine the utilization rate of computing power."
For example, in a training cluster with tens of thousands of GPUs, if network latency increases by a few microseconds, or congestion causes some GPUs to wait for data, the ultimate loss is not just a few microseconds, but the cost of tens of thousands of GPUs idling simultaneously. For AI factories costing billions of dollars to build, such losses are unacceptable.
It is against this backdrop that the value of Mellanox truly began to be realized.
As one of the most important players in the global high-speed interconnect field, Mellanox has long been deeply involved in InfiniBand and high-performance Ethernet switching chips. Its switching ASICs, network interface cards (NICs), smart NICs, and networking software have accumulated deep advantages in the HPC field. After the acquisition, NVIDIA not only gained a complete networking product line but, more importantly, the ability to unify the design of GPUs, CPUs, DPUs, switches, and even optical interconnects.
The widely discussed Spectrum-X today is a representative example of this integration capability.
It is not simply about selling a switch; it combines Spectrum switching chips, BlueField DPUs, ConnectX NICs, LinkX high-speed interconnects, and a software stack into a complete AI networking platform. Compared to the traditional data center model of "servers from one vendor, switches from another, and NICs from yet another," NVIDIA is now delivering a complete AI infrastructure to customers.
For hyperscale cloud providers, the greatest value of this model is not reducing the number of procurement sources, but enabling the co-optimization of GPUs, networks, and software, thereby improving the training efficiency of the entire AI cluster.
This is why, in recent years, more and more AI factories have adopted a "GPU + network" integrated procurement model, rather than purchasing servers and switches separately.
In a sense, NVIDIA is no longer just selling GPUs; it is selling the entire AI data center.
However, according to an analysis provided by Semianalysis, in addition to winning customers through product strength, NVIDIA also seems to be leveraging its influence. Semianalysis stated that many new cloud executives they spoke with believe that if their clusters contain non-NVIDIA networking equipment, or if their cloud services offer AMD GPUs or TPUs, NVIDIA will retaliate. This is another reason driving the exceptional performance of NVIDIA's networking business.
Despite rapid growth, NVIDIA has not stopped pursuing new technologies.
The Future of NVIDIA's Networking Technology
In an interview with networkworld last September, Gilad Shainer, Senior Vice President of Networking at NVIDIA, stated that data centers are evolving into a new type of computing unit, where the primary computing unit shifts from CPUs to GPUs, and functions are distributed across different components to support AI workloads. This infrastructure evolution requires synchronized data transmission and involves at least four networks: the compute network, the scale-up network, the scale-out network, and the access network.
"Today, the scale of data centers has changed. In the AI era, the data center itself has become the computing unit. We no longer ask 'How many CPUs can I buy?' but rather 'How do I design a data center that can run my workloads with maximum efficiency?'" Shainer said. "This shift fundamentally changes how we design, connect, and optimize infrastructure. The mega data center has become the new computing unit, and existing network architectures can no longer cope," he added.
Shainer previously explained in an NVIDIA blog: "What we need is a layered design using cutting-edge technologies—like co-packaged optics, which once seemed like science fiction."
NVIDIA has also integrated a set of algorithms into its Spectrum-X Ethernet platform that can implement various network protocols, enabling Spectrum-X switches, ConnectX-8 SuperNICs, and systems equipped with Blackwell GPUs to connect over long distances without hardware changes. These Spectrum-XGS algorithms dynamically adjust control parameters using real-time telemetry data (tracking traffic patterns, latency, congestion levels, and inter-site distances).
Developing and building Ethernet technology is a key component of NVIDIA's roadmap. Since the initial launch of Spectrum-X in 2023, NVIDIA has rapidly developed Ethernet into a core R&D direction. At the same time, NVIDIA is still actively developing InfiniBand technology, which remains its core connectivity solution.
"InfiniBand was designed from the start for synchronous high-performance computing, featuring capabilities like RDMA to bypass CPU jitter, and supporting adaptive routing and congestion control," Shainer said. "It is the gold standard for large-scale AI training, connecting over 270 of the world's top supercomputers. Ethernet is catching up, but traditional Ethernet designs—built for telecom, enterprise, or hyperscale cloud—were not optimized for the unique demands of AI," Shainer said.
"When we first started focusing on AI backend networks in late 2023, InfiniBand dominated the market with over 80% share," wrote Sameh Boujelbene, Vice President at Dell'Oro Group, in a previous report. "Despite InfiniBand's dominance, we have always predicted that Ethernet would eventually prevail in scale-out applications. However, it is noteworthy how quickly Ethernet has been adopted in AI backend networks. As the industry moves toward 800Gbps and beyond, we believe Ethernet has now firmly established an advantage and is poised to surpass InfiniBand in these high-performance deployments."
650 Group also predicted in a report: "In the next year or two, with the proliferation of 800G and the formation of 1.6T networks, Ethernet will become the more mainstream networking technology, surpassing InfiniBand. The 800G cycle in AI will set new records for revenue and port shipments."
For scalable networks like NVIDIA's NVLink, optical components are a crucial element in the connection, as a large amount of bandwidth needs to be transmitted between connected GPU silicon devices.
"We are working to increase compute density within a single rack so that the rack can use copper cabling. Copper has zero power consumption, is reliable, and very cost-effective. Whenever copper can be used, it should be used. But when customers need to extend the network over longer distances, copper cannot be used because it cannot transmit that far, and then fiber optics are needed," Shainer said.
Currently, NVLink offers up to 1.8 TB/s of bidirectional bandwidth per GPU, supporting up to 72 GPUs per rack. It is expected that faster and higher-capacity NVLink technology will develop rapidly in the coming years to meet the demands of higher speeds and increased inter-GPU communication.
In the field of optical communications, NVIDIA provides pluggable optical modules for its Ethernet and InfiniBand networking equipment. However, NVIDIA is also aggressively moving into the co-packaged optics (CPO) networking domain. CPO integrates the network optical module directly into the switch ASIC chip. CPO technology is expected to develop rapidly in the coming months and years to handle AI traffic and, eventually, other high-performance network traffic.
Final Thoughts
An IDC switch market report, seemingly just a change in vendor rankings, actually reflects a fundamental shift in the competitive logic of the AI era.
In the past, GPUs, CPUs, switches, and NICs belonged to different markets; today, they are being redefined as parts of the same AI infrastructure. What determines the outcome of competition is no longer which chip has the leading performance, but who can truly integrate computing, networking, interconnects, software, and systems.
From acquiring Mellanox, to building Spectrum-X, to deploying NVLink, CPO, BlueField, and a complete networking platform, NVIDIA has essentially been answering the same question—how to make increasingly large AI clusters run with maximum efficiency.
This is why switching chips have become NVIDIA's new growth engine.
In the coming years, as AI factories move from tens of thousands of cards to hundreds of thousands or even millions, the importance of networks will only continue to increase. For NVIDIA, what is truly worth watching is perhaps not how many more GPUs it sells, but whether it is quietly becoming the arbiter of the entire AI infrastructure.
This is a topic worthy of deep consideration for all industry participants.
