No one like the R word, but we don’t shy away from data and calling it like we see it. And from what we can see from the most recent market research coming out of IDC, the Ethernet switch market has been in recession for two quarters now and very close to three. And that is despite the enormous demand for back-end networks for AI infrastructure.
Or, maybe it is because of it. We think this was the case for server spending since the GenAI revolution began in earnest at the end of 2022, and once companies had to shell out money to buy or rent GPU-laden machines, they did so by stretching their existing server fleets and stalling spending on shiny new iron just as Intel was ramping up its “Sapphire Rapids” Xeon SPs and AMD was doing the same with its “Genoa” Epyc 9004s. Which was unfortunate indeed for them and their ODM and OEM partners, who were counting on a server bump after companies the world over were bleeding down their overcapacity spending from the coronavirus pandemic.
Having said all of that, with the AI crowd eager to shift away from InfiniBand networks commonly used in HPC simulation systems and towards Ethernet, we expect a step function spending for high bandwidth, low latency Ethernet switches and related DPUs for controlling congestion and doing adaptive routing for AI workloads. So this is a temporary slowdown in this part of the Ethernet space.