Nvidia Validates Ethernet w/ Spectrum-X Roadmap in 2025
1 min readEthernet vs InfiniBand
This has been the raging debate in the world of AI backend networks. Back in 2023, 80% of AI backend networks and supercomputers consisted of an Infiniband network. Fast-forward only 2 short years and Ethernet (800Gbps) is catching up at an alarming pace. Using the data available to me, it seems like for Inference workloads, 800Gbps Ethernet is now holding 70% market share. AI training is still dominated by Infiniband (70% vs 30%), but 1.6Tbps Ethernet will be in pilots/EFTs during 2026.
Here’s some great quotes from an article on networkworld.com.
“InfiniBand was designed from the ground up for synchronous, high-performance computing — with features like RDMA to bypass CPU jitter, adaptive routing, and congestion control,” Shainer said. “It’s the gold standard for AI training at scale, connecting more than 270 of the world’s top supercomputers. Ethernet is catching up, but traditional Ethernet designs — built for telco, enterprise, or hyperscale cloud — aren’t optimized for AI’s unique demands,” Shainer said.
In the coming year or two, “Ethernet will become the more prominent technology for networking, surpassing InfiniBand as 800G ramps and 1.6T take form,” predicts 650 Group in a recent report. “The 800G cycle for AI will set records for revenue and ports.”