The Tipping Point: Managing the Cost of Data Center Interconnect in the AI Era

In my previous blog post, I talked about the growing financial burden that AI-driven data growth will place on enterprises. IDC predicts that data generation will grow explosively, at a compound annual growth rate (CAGR) of 40.5% through 2027, which will lead to unprecedented growth in the need for more robust data center interconnect (DCI) solutions by enterprises.

Traditionally, enterprises have leased high-capacity circuits from service providers for DCI. These leased circuits are often 10G links using Carrier Ethernet or wavelength services—which are perfectly adequate for meeting enterprises’ current requirements. However, AI is starting to push DCI capacity towards high-bandwidth 100G, 400G, and 800G connections. Since the cost structure of DCI is usually based on bandwidth usage, costs will quickly multiply, as enterprises will need more circuits to support increased capacity demands. A less costly alternative for enterprises is to lease dark fiber from service providers and create their own private network.

Calculating the cost of DCI for AI

The big question for enterprise IT, specifically for network architects, is when to make the switch from leasing circuits to leasing dark fiber. Organizations will likely reach a tipping point when the cost of adding more capacity to leased circuits to support AI workloads will become too high. This is when a move to a different DCI model with a more sustainable cost structure is fully justified.

ACG Research investigated the three-year total cost of ownership (TCO) of Carrier Ethernet, wavelength services, and dark fiber in different use cases across metro (50 km), short-haul (200 km), and long-haul (500 km) DCI networks. This analysis revealed that dark fiber is the most strategic long-term investment for enterprises to manage exponential AI-driven bandwidth growth. Here are the results of this TCO research for the different network types:

Metro networks TCO advantage

The tipping point for moving to dark fiber in metro networks occurs when bandwidth needs exceed 100G, with TCO savings at 400G of 48% compared to Carrier Ethernet and 55% compared to wavelength services. Unlike Carrier Ethernet or wavelength services, the flat transport costs of dark fiber allow enterprises to scale bandwidth without incurring incremental fees.

Short-haul networks TCO advantage

In regional networks, dark fiber shows significant TCO benefits over Carrier Ethernet beginning at 100G, with 61% savings at 400G. For wavelength services, dark fiber has a tipping point of 30% TCO savings beyond 400G and 48% TCO savings at 800G. While distance-based costs are higher, the flat pricing model of dark fiber ensures long-term savings.

Long-haul networks TCO advantage

For long-haul connections, dark fiber is 46% more cost-effective than Carrier Ethernet at bandwidths greater than 400G. It is 14% more cost-effective than wavelength services at bandwidths greater than 800G. However, when comparing dark fiber and wavelength services, it’s important to note that dark fiber scales linearly with distance whereas wavelength costs rise in step increments that are not always proportional to bandwidth. So, in long-haul deployments, wavelength services may be more cost-effective because of the higher cost of long-distance fiber construction or limited availability. That is why it is important to evaluate costs for each service, taking into consideration the location of data centers.

How to make the switch

Given the rapid pace of AI innovation and the inevitable surge in AI data, forward-thinking enterprises should start planning a switch to dark fiber now to avoid future balance sheet hits. The simplest and most cost-effective way forward is to combine dark fiber with Cisco Routed Optical Networking, which enables optical wavelengths to be delivered directly from high-capacity ports in routers or switches in the data center. It simplifies the network by replacing dedicated transponders with industry-standard coherent pluggable optics and simpler optical line systems. For point-to-point DCI solutions, enterprises can deploy Cisco ZR/ZR+ coherent pluggable optics with switches to collapse the switching and optical layers at Layer 2. For more complex services, they can converge routing and optical layers onto a single IP/MPLS network where all switching happens at Layer 3.

This innovative approach brings many benefits for enterprises. The combination of dark fiber and Routed Optical Networking offers cost-effective scaling to multi-terabit capacities, while streamlining operations and reducing OpEx. The standardized capabilities of coherent pluggable optics eliminate the need for complex traffic engineering, while delivering the resilient, low-latency connectivity that AI applications need. Furthermore, enterprises can now gain end-to-end observability and control over their DCI network, ensure SLA compliance with requirements for bandwidth, latency, and availability, and enhance DCI performance across dark fiber links. Cisco Provider Connectivity Assurance, formerly Accedian Skylight, provides enterprises with these capabilities, with continuous and real-time telemetry across the physical and logical layers of their DCI connections.

While leasing 10G circuits might meet the current DCI needs of enterprises, it’s not a matter of if but when the cost will become too high. It is certain that AI workloads will contribute to unprecedented growth in data traffic, which makes switching to dark fiber now a strategic move. Since operating DCI over dark fiber is more complex than a managed service, enterprises can partner with service providers that lease dark fiber and benefit from their operational expertise. By investing in scalable DCI infrastructure using dark fiber today, enterprises can position themselves to capably support the increased data volumes from AI and other data-intensive workloads tomorrow.

Learn more about Cisco DCI solutions with Routed Optical Networking.

Related blogs:

Share: