Ethernet switch sales in AI back-end networks more than tripled in 2025, capturing more than two-thirds of the entire data center switch market for AI clusters by the end of the year, as the infrastructure buildout shows no signs of slowing.

Dell’Oro Group data revealed that network switch spending was largely at the higher-end of the speed spectrum as hyperscalers and large AI firms look to significantly scale their infrastructure footprints.

The analyst firm suggests that 800 Gb/s (800G) switches accounted for the vast majority of the Ethernet switch shipments and revenues in AI back-end networks during the quarter, with cluster operators looking to push past the connectivity limitations of 400G.

Cisco’s Optics Group GM Bill Gartner told SDxCentral during that same quarter that “virtually all” hyperscalers are moving to 800G – but they won’t stay there for too long, with the likelihood of them jumping to the eventual next stage “when that's available.”

That next stage of interconnect speeds will jump from 800G to 1.6 Tb/s (1.6T), with Dell’Oro Group’s report suggesting switches capable of such speed will “drive significant revenue growth.”

The likes of Synopsys, Edgecore Networks, and Marvell Technology have all raced ahead to reveal 1.6T switch lines to meet that insatiable hyperscale demand, with Dell’Oro Group suggesting shipments are expected to begin shipping in the second half of 2026.

The analyst firm suggests the rise of even faster switch lines will also result in some significant share shifts in what is already a highly cut-throat market.

“In 2025, Celestica and Nvidia led the market with a combined 50% share. Arista ranked third, despite a significant portion of its AI-related product revenue being deferred,” Sameh Boujelbene, VP at Dell’Oro Group, detailed. “Cisco has also started to accelerate its switch shipments in AI back-end networks with large hyperscalers. HPE/Juniper scored new accounts.”

“Other new entrants are actively pursuing this opportunity,” Boujelbene added. “We expect the increasing need for vendor diversity at both the chip and system levels to drive continued share shifts and keep this market highly dynamic."

Spending in the back-end switch spaces looks to maintain momentum, with Dell’Oro Group recently forecasting that related investments will surpass $100 billion by 2030.

“The growing size of AI clusters, combined with ongoing supply chain constraints, is driving the need for vendor diversity and therefore for Ethernet,” Boujelbene said. “Amazon, Microsoft, Meta, Oracle, and xAI are all adopting Ethernet. Although InfiniBand continues to grow, Ethernet is expanding at a much faster pace.”

Boujelbene’s comments add to prior Dell’Oro findings that while InfiniBand held an 80% share of the AI back-end networks market as recently as 2023, the rise in demand for interoperability has resulted in a stunning comeback for Ethernet, with related switches expected to drive nearly $80 billion in sales over the next five years.

The resurgence of Ethernet has resulted in the industry moving to consolidate around it, taking what made InfiniBand work so well – like underlying lossless technology – and applying it to Ethernet.

Efforts like the Ultra Ethernet Consortium (UEC) have seen some of the firms Boujelbene’s report mentions coming together to ensure Ethernet can support high-performance workloads.

The Open Compute Project (OCP) this week debuted the first Ethernet for Scale-Up Networking (ESUN) specification. Version 1.0, spearheaded by engineers from Microsoft and Meta, looks to standardize Ethernet-based interconnects for large GPU domains with support for link-level congestion control mechanisms and improved link resiliency to reduce packet drops.