But AI application require remote direct memory access, RDMA, not an extensible operating system, Genovese adds, flagging the potential risk to Arista 43%-44% operating margins.
“We think it could become more difficult to maintain these very high margins if EOS loses value in AI Data Centers and there is a mix shift to Enterprise driven by ongoing share,” the analyst said.
Nvidia not only has “access to the largest and the highest volume of Ethernet switching fabrics today”, Geneovese adds, but also “the most appropriate and functional networking operating system for AI, and a combined 2+ year lead on others in the industry.”
With just $750 million expected in new AI sales, Arista’s current AI deals in the pipeline cast doubt on its ability to ramp up sales from datacenter customers.
While the company is expected to continue to gain market share in enterprise, this “type of business typically does not support super-high,” the analyst said.