“Backlog normalization, inventory digestion, and spending optimization caused a decline in Ethernet data center switch sales in the first quarter of the year,” said Sameh Boujelbene, Vice President at Dell’Oro Group.Sales of Ethernet data center switches decreased in Q1 2024 for the first time since Q4 2020, according to the most recent Dell’Oro Group study. Spending optimization in the enterprise and cloud service provider (SP) segments, as well as inventory digestion and backlog normalization, are the reasons for this decline.
Though Ethernet is becoming more prevalent in AI back-end networks, this hasn’t been sufficient to offset the decline in front-end network spending overall.
“In line with our expectations, backlog normalization, inventory digestion, and spending optimization caused a decline in Ethernet data center switch sales in the first quarter of the year,” said Sameh Boujelbene, Vice President at Dell’Oro Group. ‘The decline was broad-based across Cloud Service Providers (SPs) as well as Enterprise segments, though certain accounts performed better than others depending on where they are in the digestion cycle. Despite overall spending softness in the front-end network, spending in AI back-end networks continues to grow exponentially. While Ethernet was able to capture an increasing share of AI infrastructure spending, this growth couldn’t offset the decline in the front-end network.”
Exceptions: Arista Networks and Huawei
The report also noted that most major branded vendors, except Arista Networks and Huawei, experienced revenue drops. White-Box vendors saw a revenue uptick, driven by spending recovery from major Cloud SPs like Google, Amazon, Alibaba, Tencent, and ByteDance.
Shipments of 200 Gbps, 400 Gbps, and 800 Gbps data center switches made up nearly 25% of total port shipments in Q1 2024, with these speeds showing growth during the quarter.
The adoption of 400 Gbps and 800 Gbps data center switches is expected to accelerate in 2024 and 2025, as AI workloads increasingly favor Ethernet, as discussed in the Dell’Oro Group’s ‘AI Network for AI Workload’ report.
