How to Make Your Hybrid Network AI-Ready Without Increasing Costs

How will AI impact enterprise networks?

That's the huge topic I'm tackling with data from TeleGeography’s WAN Cost Benchmark—a customized platform that makes it easy to model and track your WAN network costs.

This is the second post in this series. Check out the first post here to learn which factors we're considering and how we're evaluating various scenarios that model possible AI-driven WAN configurations. 

The point of the series is to compare the total cost of ownership (TCO) of a legacy MPLS network—one that represents many multinationals’ WANs still in 2025—to various hybrid or internet-first alternatives designed to accommodate the changes AI and cloud are bringing to the modern enterprise IT team.

While our argument for SD-WAN-enabled hybrids at the start of the decade was mostly about cost, it is now much more about getting the network right-sized for the coming AI-driven bandwidth crunch, resiliency, app-specific policy and performance, and full utilization of cloud. 

Your Budget + Your Legacy Network + AI

In this post we're looking at the budgetary impact of making your legacy network capable of handing the AI boom by evaluating two versions of the ‘tiered network’ strategy that breaks the network into site types. This strategy makes it easier for network teams to change products and bandwidths and right-size to the needs of that location. For our hypothetical network scenarios, we broke each of our hypothetical network’s 150 global sites into four tiers: 

• Data Centers (DCs) = 7 sites.
• Tier 1, larger sites that need more bandwidth and the best resiliency, often downtown or in close-in suburbs = 18 sites.
• Tier 2, medium sites that might need less bandwidth but still require resiliency = 49 sites.
• Tier 3, the smallest sites with the least bandwidth and less need for resiliency = 76 sites.

A special note for DIA service: Increasingly, we are finding that carriers offer DIA for the port-only price when the customer building is on-net or even near-net, meaning there is no additional local access charge. This is most common in urban areas in competitive well-developed metros, so in those locations, we have dropped additional access charges when switching MPLS service to DIA across all scenarios.

Conservative Tiered Hybrid WAN

Our Conservative Tiered Hybrid scenario imagines an enterprise that might be early in their AI journey and generally a bit cautious. Their network team wants to make the move to SD-WAN and local internet breakouts, but has security or compliance concerns, proprietary apps or on-prem data centers, or just a conservative outlook on IT that keeps it from going internet-first. This company doesn’t yet need big bandwidth increases, but is looking to the SD-WAN hybrid as a way to ensure resiliency, enact application specific-policies, and facilitate cloud utilization with local breakouts. These are the changes at each tier:

• DCs—Keep dual MPLS.
• Tier 1—Keep dual MPLS, add a single DIA port as a tertiary connection with separate access lines.
• Tier 2—One MPLS port and one DIA port with separate access lines where applicable. 
• Tier 3—Dual DIA ports (active-active) with redundant access lines where applicable. 
• All tiers—Basic Managed SD-WAN for the total encrypted throughput of all underlay services at each site.

Average Total Bandwidth per Site in Each Subregion—Dual MPLS WAN vs. Conservative Tiered Hybrid WAN

Note: Each bar represents the average total site bandwidth, including multiple ports or underlay services, across all sites in the listed subregion.

• The resulting network increases the average available bandwidth at each site by about 8%, mostly due to the tertiary links added at Tier 1 sites. For the most part, the bandwidth at any given site is similar to the original Dual MPLS.  

Distribution of Total Site Speeds—Conservative Tiered Hybrid WAN

Note: Each section represents the percentage of total site bandwidths that fall within each bandwidth range.

• A majority of the sites fall within a total site bandwidth range of 200 Mbps or below, a fairly cost-conscious amount of bandwidth. 
• The most typical sites in the network are between 100 and 200 Mbps across all underlay services. 
• Roughly 14% have a total site bandwidth of more than GigE.  

Now, before we look at prices, let’s take a look at a similar network set up from a more aggressive IT team who is already feeling the AI bandwidth crunch and is more willing to add internet into the mix to take advantage of lower cost per bit.

Tiered Hybrid WAN

The Tiered Hybrid WAN represents an enterprise that is a bit more ready than the conservative scenario to push towards internet, but not ready to leave MPLS behind all together. They are also willing to try broadband, but only in markets where they are certain the ISPs are ready to provide an enterprise level of service. Most importantly, they have started to utilize AI and have seen cloud workloads and traffic increase. They need to boost bandwidth without breaking the bank. While they are keeping some MPLS, they are adding DIA at larger ports due to the lower cost per bit. These are the changes at each tier: 

• DCs—Dual DIA ports.
• Tier 1—One MPLS port and one DIA port with separate access lines.
• Tier 2—Dual active-active DIA ports with redundant access where applicable.
• Tier 3—Dual active-active DIA ports in most global markets, but one DIA port and one business broadband connection in the U.S., Western Europe, and a few other competitive ISP markets.
• China is an exception to the rules, as we have kept dual MPLS at those sites.
• All tiers—Basic Managed SD-WAN for the total encrypted throughput of all underlay services at each site.

Average Total Bandwidth per Site in Each Subregion—Dual MPLS WAN vs. Tiered Hybrid WAN

Note: Each bar represents the average total site bandwidth, including multiple ports or underlay services, across all sites in the listed subregion.

• We increased the bandwidth available at sites around the world to about the same degree as the high capacity Dual MPLS network, with average bandwidth increasing about 62% per site.

Distribution of Total Site Speeds—Tiered Hybrid WAN

Note: Each section represents the percentage of total site bandwidths that fall within each bandwidth range.

• A slim majority of sites fall in the more than GigE to 2,000 Mbps total site bandwidth range.
• Only about one-third of sites have a total bandwidth of 500 Mbps or less. 

Total Cost of Ownership Comparisons

The tiered network scenarios give us a picture of two strategies for the modern network starting to accommodate cloud and AI. The Conservative Hybrid scenario represents an enterprise that wants to modernize their network to facilitate cloud without changing the budget. The Tiered Hybrid scenario shows us an example of an enterprise that has already felt the AI-bandwidth crunch and wants to boost speeds without boosting budgets. 

This TCO comparison includes the high-capacity Dual MPLS WAN to illustrate the budgetary benefits of going hybrid, beyond the resiliency, application performance, and cloud-first  drivers. 

Original Dual MPLS, High Capacity Dual MPLS, Conservative Tiered Hybrid, and Tiered Hybrid WAN Scenario TCOs

Note: Each column represents the total annual cost of ownership for that WAN scenario, broken out by product in each color section.

• The annual TCO for the Conservative Tiered Hybrid increased capacity over the Dual MPLS slightly–8%–but the cost went down by about 9% from $8.1m to $7.4m. 
• The Tiered Hybrid and High Capacity Dual MPLS both increased the average bandwidth available per office by more than 60%, however, the Tiered Hybrid TCO only rose by 18% going from $8.1m to $9.6.
• Compare the above point to the high capacity Dual MPLS which went up 43%. This is also with keeping a rigid, application unaware, and centralized breakout legacy network.

It should be no surprise that a hybrid network is a better cost option at this point, but it is worth fully understanding the budgetary balance. DIA and especially broadband are obviously more cost effective than MPLS, but again, the enterprise that is running headlong into AI needs to have a faster and more flexible network, not just a cheaper one. Multinationals with sites around the world need to understand how product choice, bandwidth, and geography impact their mission to find a network that can handle the future demands of the rapidly changing IT team. Next time we will take a look at the enterprise that has gone fully into cloud and realized this can help them cut costs on real estate by foregoing a return to office policy and thus cutting the network footprint.