This podcast conversation covers:
Carlie Carson, Research Analyst on the enterprise team at TeleGeography, joins podcast host Greg Bryan to discuss:
▪️ How has AI inferring impacted the enterprise network?
▪️ Is SD-WAN still relevant?
▪️ Has MPLS disappeared from the enterprise WAN?
The pair leverage their own Enterprise Network Trends and Strategy Report to bring actual data to the trends we often talk about on the pod.
Carlie and Greg discuss the shifting underlay landscape as MPLS usage drops and Dedicated Internet Access (DIA) takes center stage, why enterprises choose DIA over business broadband, and the real impact (or lack thereof) that AI is having on current enterprise bandwidth demands.
They also unpack the reality of satellite connectivity adoption, the evolution of SD-WAN, SASE, and Zero Trust, and why Network as a Service (NAS) is still struggling to gain widespread traction.
Key takeaways:
Since 2018, we've been tracking the evolving nature of corporate enterprise networks through our WAN Manager Survey. In this seventh edition, we collected 52 survey responses and conducted 13 interviews with WAN managers for additional insight into their network-related decisions and strategies.
The Underlay Shift: DIA takes the lead over MPLS
MPLS decline
Usage dropped sharply from 82% of respondent sites in 2018 to just 22%. While MPLS remains necessary for specific geographies (such as China and Africa) and legacy data centers, it is no longer the central option around which networks are built.
DIA dominance
Dedicated Internet Access (DIA) expanded from 27% in 2018 to 54% of sites, solidifying its place as the primary underlay option. Nearly 95% to 96% of surveyed enterprises run DIA somewhere in their network.
Broadband limitations
Despite business broadband offering cost savings, enterprises still lean heavily on DIA due to strict carrier-grade SLAs, high reliability, and uncontended bandwidth requirements.
The AI reality check on bandwidth
Minimal current impact
Contrary to market hype, AI and Generative AI are not yet significant drivers of enterprise bandwidth growth. Between a quarter and a third of surveyed managers stated AI was not impactful to their bandwidth needs.
Current usage patterns
Most enterprises currently utilize AI for simple text prompts rather than automated, bandwidth-heavy application loops or localized model training.
True bandwidth drivers
Cloud resources and Software as a Service (SaaS) applications remain the primary engines driving bandwidth expansion and port upgrades up to the 10-Gig range.
SD-WAN, SASE, and the Rise of Zero Trust
SD-WAN maturity
Approximately 63% of enterprises have deployed SD-WAN. While deployment times still average about a year, a growing minority (~12%) are choosing to bypass SD-WAN entirely as network topologies shift toward edge security.
Rapid SASE adoption
SASE implementation has reached 53%. Within security architectures, Zero Trust Network Access (ZTNA) reached 79% adoption, effectively replacing traditional IP connections.
Management preferences
Managed SD-WAN accounts for 56% of deployments. Smaller IT teams lean heavily on managed or co-managed solutions, while larger teams tend to insource network management.
Niche technologies: satellite and NaaS
LEO satellite growth
About 44% of respondents run satellite connectivity at more than 5% of their sites. Low Earth Orbit (LEO) options like Starlink are primarily leveraged for hard-to-reach locations (such as agriculture or remote industrial sites) or as secondary backup links.
NaaS stagnation
Network as a Service (NaaS) adoption remains flat between 15% and 17%. While interest exists for seasonal bandwidth scaling and unified performance monitoring, broad enterprise implementation remains limited.
Enterprise network engineering is moving away from rigid, single-vendor setups toward a hybrid model prioritizing flexibility, direct internet access, and cloud-first security. As bandwidth costs decrease and automated tools like AI Ops mature, WAN managers will continue balancing connectivity performance with integrated security models.
