Nvidia's networking sales topped $17.5 billion in one quarter, more than Cisco's whole business, and that may shape what AI costs you.
Nvidia, known for AI chips, now sells more networking gear than Cisco sells of everything. In the second quarter, Nvidia's networking revenue topped $17.5 billion. Cisco's revenue across all its products in the same period was $17.3 billion. The Register also reports that Nvidia now has bigger datacenter Ethernet sales than any other vendor.
It looks sudden, but it began in 2019. Nvidia paid $6.9 billion for Mellanox, a longtime partner. Mellanox made network cards, called NICs, that connect a computer to a network. It also made InfiniBand switches, a fast type of connection common in supercomputers. Outside of those, InfiniBand was not widely used in huge datacenters.
The deal let Nvidia sell whole clusters of linked machines, not just chips. Customers wanted clusters, and clusters need fast connections between the machines.
Nvidia's CEO Jensen Huang calls this approach co-design. Parts and software are built to work together for specific jobs. Analyst Sameh Boujelbene of Dell'Oro Group says customers pick Nvidia for its tightly integrated computing and networking. She says that makes deployments easier and speeds time to market.
Recursion, a drug discovery company, offers an example. Its CTO, Ben Mabey, says its data now totals more than 50 petabytes. Its first cluster used ordinary Ethernet and hit the limits of conventional networking. For its second, with around 300 accelerators, it bought Nvidia's InfiniBand switches and network cards. Mabey says buying from one vendor made tuning and troubleshooting far simpler, because Nvidia could treat it as a single system.
The Register describes two big jumps in Nvidia's networking revenue. The first came in 2023 after ChatGPT launched. The second spread across 2024 and 2025, as Nvidia shipped its Blackwell-generation chips in large numbers. Boujelbene ties both to new Nvidia switches and growing use of its Ethernet switches in AI clusters.
Elon Musk's Colossus supercomputer was an early user of Nvidia's Spectrum-X Ethernet switches. Meta also uses Nvidia's Spectrum switches, including at its Prometheus site in New Albany, Ohio.
Nvidia has not always reported networking revenue as its own line. The Register built its figures from comments by the company's CFO and some rough math, so treat the numbers as estimates. The article text provided also cuts off before it finishes explaining why Spectrum-X is selling so well.
When you buy AI tools or cloud services, ask whose chips and networking sit underneath, because one vendor increasingly supplies both.
World models, AI that learns physics from video, are being built into home robots and driverless cars, and your own videos may help train them.
Read more: AI that learns how the physical world works is coming to robots and carsA new Alibaba-led test of real coding requests found the best AI agent fully finished just 26.4 percent of tasks, so human review still matters.
Read more: AI coding helpers fully finish only about 1 in 4 real tasksA University of Chicago law professor asked Claude to fix page numbers. It deleted his syllabus and a year of notes instead. Dropbox saved him, not the AI.
Read more: An AI assistant deleted a law professor's year of lecture notes
