• United States
Senior Editor

Cisco’s supply chain shortages are history, AI networking gains momentum

Nov 16, 20234 mins
Generative AINetworking

Cisco says a backlog of purchased, uninstalled gear is causing a slowdown in new orders and impacting its bottom line.

Credit: Shutterstock

While the supply chain issues that dogged most of the networking vendors over the past couple of years have dwindled, Cisco says the current challenge is a backlog of purchased gear waiting to be installed by large enterprise and service providers.

"After three quarters of exceptionally strong product delivery, our customers are now focused on installing and implementing these unprecedented levels of products. The bottleneck that we previously saw in the supply chain has now shifted downstream to implementation by our customers and partners," Cisco CEO Chuck Robbins told analysts on its first-quarter financial call this week.  

Robbins cited, as an example, a large enterprise that may be doing an infrastructure refresh or doing a branch rollout and has ordered and received 400-500 switches. That gear may be sitting with the enterprise, waiting to be installed. Or the enterprise could have a partner that’s doing the staging, and that partner may be backed up. “So it varies, but it could be either one of those or a combination of both,” he said.

“Simply put, customers are now taking time to onboard and deploy these heightened product deliveries,” Robbins said. Cisco estimates “there is an additional one quarter to two quarters worth of shipped orders in customers’ hands, still waiting to be deployed," he added.

The result is less orders, until this backlog is installed, and an impact on company revenues, Robbins said.  "We believe this implementation phase is the primary reason for the slowdown in new orders."

That doesn't mean products are not selling. Robbins noted Q1 product revenue growth was 9% year over year. But if you compare that to Q4 product revenue growth of 20% and Q3 product revenue growth of 17%, it’s apparent where the product backlog potentially comes from.

As for overall supply chain constraints, Cisco said those are now behind the vendor. “… both ship and lead times and backlog have largely returned to normal levels,” according to Cisco CFO Scott Herren. Cisco said its product backlog was valued at nearly $14 billion in early 2022, and its software backlog was about $2 billion in February 2023.

Other vendors such as Arista Networks and Juniper Networks have reported similar normalization. Arista most recently said supply chain and lead times are improving steadily in 2023 and are expected to normalize in 2024.

Another hot topic in Cisco's first-quarter results is the growing impact of AI networking on the bottom line. 

"We see continued momentum in AI with three of the top four customers deploying our hyperscale Ethernet AI fabric. We also already have line of sight to over $1 billion in orders for AI infrastructure for major cloud providers in fiscal year 2025," Robbins said. 

The vendor said it is continuing to work with key GPU and storage partners to create AI networking gear around Ethernet technologies, GPU-enabled infrastructure, and jointly tested and validated reference architectures.

Robbins cited the 650 Group's recent finding that the AI switching market is forecasted to exceed $10 billion by 2027.

Arista, too, has reported an uptick in AI networking development. It expects to see AI networking trials start in 2024 and production deployments to become more widespread in 2025 resulting in $750 million in AI revenue.

Juniper has also reported AI networking growth in recent weeks.

"We expect AI adoption to drive a meaningful uptick in traffic growth that is likely to benefit our cloud wide-area footprint over time," Juniper CEO Rami Rahim said during the company's most recent financial update. "We also remain optimistic regarding our AI data center switching opportunity, but we are already seeing success with cloud major and enterprise accounts due to the performance and power efficiency of our custom silicon, the congestion management capabilities embedded within our Junos operating system and our support for technologies such as RDMA [remote direct memory access] networking."

"Our optimism is driven by our strong wide-area footprint, the rapid traffic growth that continues in many of these customers’ environments and the opportunity to capitalize on the adoption of large language model and the build-out of AI clusters, where we are seeing strong customer engagement that is driving optimism regarding our opportunity to benefit as the industry increasingly considers Ethernet as the right choice for a wide array of AI-ML use cases, including front-end, back-end, in-print and storage networks," Rahim said.