Looking back on 2025, we got a few things wrong and a few things right. One thing we got right is that 2024 was indeed the year that the Router market resets itself to a new baseline for future market growth. The part we got wrong was the steepness of the decline in 2024 and the speed of the market’s rebound in 2025, especially for Core Routers.
We expected the customer inventory surplus to conclude by the first half of 2024, which it did, but we did not expect that it would take another two quarters for new orders to flow through into revenue for the system houses. So, the market dropped nearly 20% in 2024 (more than we were expecting). Both communication service providers (CSPs) and cloud providers pulled down purchases through 2024, adding to the pain. But then came AI.
Right at the bottom, at the reset point for the Router market, cloud providers were accelerating their investments in building AI data centers, and companies were beginning to push out Agentic AI, autonomous artificial intelligence systems that adapt and learn to take over tasks that require reasoning. The result was an accelerated investment cycle in all things AI. But, more importantly for the router market, the need to interconnect or transport data outside the confines of the data center wall was beginning. This was evident in the numbers we captured in the 3Q25 survey process: Core Router revenues grew over 40% year-over-year (Y/Y) in the first nine months of 2025. It is becoming a V-shaped recovery, driven by the need for more wide area network (WAN) capacity and data center interconnect (DCI). So, this brings us to 2026.
Our preliminary view for 2026 is that growth rates will remain high, and the market momentum from 2025 will continue.
- Cloud providers will continue building new AI data centers and connecting them to the WAN or neighboring data centers.
- Since electricity is a scarce resource, cloud providers will build data centers in new geographic regions. And in many cases, we believe that cloud providers will leverage CSPs to help build their new routes and networks, increasing future spending by network operators.
- Enterprises, in preparation for deploying AI Agents, will build AI-ready infrastructure to improve access speeds to cloud data centers, including adding more connections between their on-premises storage sites and cloud-based AI compute resources.
Building on this momentum, we believe there is a good chance that the 2026 results will surpass our expectations. But it depends on the answers to these questions:
- Scale-Across DCI: Cloud providers are interconnecting their AI GPU data centers with massive amounts of bandwidth to form large virtual data centers using data center switches and ZR+ optics. However, we are not sure whether all of these scale-across DCI builds will use data center switches or whether some will use routers. Hence, the question is, will routers with deep buffers and WAN protocols be chosen for scale-across DCI when span lengths are too long for data center switches?
- AI-Ready Infrastructure: We believe the early adopters are upgrading their infrastructure to be ready for Agentic AI. Will this become mainstream in 2026 or 2027?
- Fronthaul: Fronthaul with routers had many false starts in the past decade. Hence, we are still cautious about its prospects before 6G. However, recently, there has been increased activity around eCPRI (enhanced Common Public Radio Interface) for fronthaul. So, will router fronthaul build-outs occur before 6G?
