What many network leaders manage today is neither an MPLS-only estate nor a fully internet-first one. It is the accumulated result of infrastructure decisions made over the past decade, layered with cloud migration, SD-WAN adoption, acquisitions, contract renewals, and changing application traffic.
Some MPLS circuits remained because they continued to deliver measurable value: for specific applications, for locations with limited options, for regulatory obligations that mandate private connectivity, or for service levels nothing else could contractually match. Others remained because early exit was uneconomical, the replacement case was unclear, or decommissioning became a project nobody had the headcount to complete.
The real question is no longer whether MPLS stays or goes. It is which sites and workloads still justify the premium. In many organizations, the answer depends less on a single corporate policy and more on local realities such as access availability, local-loop economics, application dependencies, and SLA requirements.
MPLS Survived Because of Application Requirements, Not Technology Preferences
For cloud and SaaS applications, direct internet access has become the standard approach. But applications no longer fall neatly into “cloud” and “internal” categories. ERP moved to the cloud. Voice became SaaS. Manufacturing systems started communicating with cloud platforms. The old rule of sending cloud traffic over the internet and internal traffic over MPLS has become much harder to apply.
If application location no longer tells you which network to use, the decision has to come back to the performance parameters the workload requires. Voice, virtual desktops, ERP platforms, manufacturing systems, large data transfers, and AI workflows all respond differently to latency, jitter, packet loss, congestion, and interruption.
Those requirements continue to change. Cisco research found that AI inference flows can last around twice as long as ordinary web transactions, produce more sustained traffic and place different demands on upstream capacity. As AI agents begin operating continuously, the need may not simply be for more bandwidth. It may also include greater path visibility, traffic prioritization, stronger resilience, and a clearer understanding of which traffic has become operationally critical.
Choosing the right access model therefore involves more than comparing speed and latency. It also requires considering how much control and service assurance the workload needs. DIA can deliver excellent throughput and low latency, particularly where applications are internet-delivered. The difference is the boundary of control. MPLS provides a larger controlled domain, and stronger service guarantees for certain traffic patterns. A DIA provider may offer an SLA for its own service and network, but it cannot contractually guarantee the performance of every third-party network that traffic crosses on the public internet.
Move to internet-based connectivity, and that accountability does not disappear. Someone still has to determine whether a performance issue sits with the access circuit, the SD-WAN overlay, the internet path, or the application itself. As more of the end-to-end path falls outside a single provider’s control, visibility across those layers becomes increasingly important.
| Control domain | What it covers |
|---|---|
| Enterprise-controlled | LAN, routing policy, SD-WAN edge and overlay |
| Provider-controlled | Access circuit, carrier backbone and peering |
| Beyond either party’s direct control | Internet transit, third-party networks and SaaS provider infrastructure |
Security requirements can also influence the access decision. MPLS does not inherently encrypt traffic, but some organizations continue to value the logical isolation of a provider-managed private network for particular workloads or risk-management requirements. That isolation still needs to sit alongside the appropriate encryption, segmentation, access controls, and monitoring.
Whether those characteristics justify the additional cost depends on the workload and the site.
That is where many renewal decisions become difficult. Enterprises may know the cost, carrier, and contract term for every circuit, but not which applications depend on it, what service requirement it was intended to meet, or what would happen if it were removed.
A circuit inventory is not the same as an application dependency map. For most organizations I work with, that map does not exist. The circuit inventory is sitting in a spreadsheet somewhere, but the rationale behind each circuit often leaves with the engineer who designed it.
Every Renewal Is a Chance to Reassess the Access Strategy
If the original reason for a circuit is no longer documented, renewal is the point to rebuild the case.
The question is not simply, “Should we retire MPLS?” It is, “What does this location need now, and which access model best meets that requirement?”
What applications depend on the site? What performance and availability levels do they require? What new workloads or traffic patterns are likely to emerge during the next contract term? What fixed, wireless, or satellite options are available at the address? Does the site need a primary circuit, a resilient secondary path, or a different mix of access altogether?
That sounds straightforward. In most organizations, the information needed to answer those questions is fragmented. Application dependencies may sit with architecture teams, contract dates with procurement, carrier options with sourcing, and site details with local IT or facilities. The organization may know what the circuit costs and when the contract ends, but not whether the current access model is still the right one.
The Cost Conversation Is Often About Local Access, Not Ports
The MPLS-versus-DIA debate is often framed around port pricing and architecture. In practice, the economics are frequently decided by the local access circuit.
A branch in a well-served metro area may have several low-cost DIA or broadband options. A warehouse, industrial site, or secondary-market location may face expensive local-loop charges, long installation times, or limited carrier choice. In those cases, replacing MPLS may deliver little cost savings, while wireless or satellite may provide a more practical primary or resilient secondary path.
The right combination of MPLS, DIA, broadband, wireless, or satellite therefore depends on the economics, service requirements, and connectivity options available at each address.
A global WAN strategy can define the standards, performance requirements, and preferred access models. But the business case still has to be tested at each address.
Carrier availability, construction requirements, local competition, installation lead times, and the quality of fixed, wireless, or satellite services can vary significantly from one location to another. Two sites built to the same global standard can therefore have completely different cost profiles and deployment options.
The same dynamic plays out between markets: In parts of Europe, regulated wholesale access keeps loop costs low enough that replacing MPLS saves little, while in North America, cheap on-net DIA makes the same circuit look indefensible.
The challenge is not choosing between centralized strategy and local decision-making. It is applying one consistent network strategy across hundreds of local access markets, without forcing every site into the same connectivity model.
From Inherited Network to Intentional Network
The organizations getting WAN modernization right are not the ones that have removed the most MPLS. They are the ones that understand why every remaining circuit exists, what business outcome it supports, and what would happen if it disappeared tomorrow.
The real risk is not keeping MPLS or replacing it. It is making either decision without understanding the role each circuit plays. When that visibility is missing, organizations either continue paying for services they no longer need or remove connectivity that still supports critical applications and operations.
In 2026, the challenge is no longer choosing between MPLS and DIA. It is building an access strategy in which every circuit, service, and resilient path exists for a clear reason, rather than simply because it was inherited.
For multinational organizations, that challenge is harder still. Few providers can stitch regional networks into a single global network, wrap them in managed services, and support them under one SLA. That is why WAN modernization is not just an access decision. It is also a question of how the network will be operated, governed, and supported across regions
Author
Bal Athwal - Vice President, Strategic Global Enterprise Solutions
Bal Athwal works with global enterprises on complex connectivity challenges where standard solutions fall short. His focus is on designing network architectures across large, distributed environments, balancing performance, resilience, and real-world constraints.
His work includes building multi-region connectivity solutions across complex, multi-carrier environments, navigating both the technical and commercial realities of global network infrastructure, and engineering solutions that go beyond standard carrier offerings.
He holds an MBA in Technology Management from Stevens Institute of Technology.