What network models are enterprises looking at for the future? How might the network of 2028 differ from that of 2025? I got some information from 294 enterprises that offer some answers to these questions, but they also point out that there are many different drivers operating on networks over the next three years, and these drivers impact enterprises based on things like the vertical market they’re part of, the size of the company and their office sites, and the unit value of labor and nature of work in each office. So one size here won’t fit all, but we can only deal with broad market questions based on averages.
Enterprises don’t see any specific shifts in the missions that drive networking to change things in the next three years. If that’s true, then it’s likely that the same cost-driven network modernization policies that have been in force over the last decade (except during the COVID lockdowns) will continue to drive things. More for less, then, is the goal.
One area where there’s a lot of interest is in branch or secondary site access. The traditional approach to this has been the MPLS VPN, but enterprises are very concerned about site access costs, and VPN access will, they say, cost as much as ten times the price of the same bandwidth delivered as business Internet. Yes, they admit that QoS and reliability might be better with the MPLS VPN, but they point out that companies depend on the Internet for direct prospect/customer connectivity already, and that mobile workers or those in home offices or mobile also depend on an Internet path back to enterprise applications.
Of our 294 enterprises, all say they already depend on the Internet to reach customers and prospects, and all say that mobile workers use the Internet to reach the company VPN. About a third say that some branch locations are connected via the Internet already, including those who use mobile-worker VPN access techniques for very small office locations.
Formal SD-WAN or SASE is harder to pin down; roughly a fifth of enterprises say they use it, but I’m not confident that all those who don’t say are in the “don’t use” category. For example, 147 of the 294 commented on what site conditions would to them justify VPN connectivity, which is exactly half the total, and why would they have a view here if they didn’t have to decide between VPNs and an Internet-based VPN connection somewhere?
With regard to those comments on site conditions, the primary metric enterprises say they look at is the number of workers. On the average, companies suggest that an IP VPN isn’t cost-effective in sites with fewer than 17 workers, but the nature of the workers and the work seems to influence this. Companies whose remote sites are largely staffed with high-unit-value-of-labor professionals (healthcare, finance, etc.) set the decision point as low as 8 workers, and those whose office locations are staffed with people whose labor value is lower set it as high as 28. The presence of any custom network-connected devices (again, common in healthcare and finance) argues in favor of MPLS, too.
Of the 294 companies 208 say they could do more with SD-WAN and related technologies, meaning that there are sites where they believe it could be applied. The “why-wasn’t-it” comments offered center on simple inertia; enterprises tend not to change things unless there’s something specific to drive them to consider change. Here, there seems to be an indication that a driver is emerging. Among the 208, 78 say that they are “considering” initiating or expanding the displacement of MPLS VPNs, and all this group cite a company drive to reduce network service costs. Most interestingly, of the 28 comments I’ve gotten since the latest tariff/market flap, 22 say that reconsideration of MPLS VPN usage is likely to come along. This suggests that financial uncertainty could generate an interest in a preemptive cost-reduction push, even before any specific need to do that has emerged.
AI is another complex topic, even beyond network implications but especially within them. Going back to the first Andover Intel comments on AI back in early 2023, enterprises weren’t really considering the issue at all. Within a year, they divided into two camps, the AI-traffic optimists and pessimists, with roughly two thirds in the first group and the remainder in the second. Even at this point, there wasn’t much specificity regarding what kind of traffic is generated and where it’s going.
In late 2024, most AI-traffic optimists said that the primary impact would be from self-hosting AI, and likely in the data center LAN. The optimist group had grown, at this point, to account for three-quarters of enterprises. The pessimist group didn’t see self-hosting at all (they believed in cloud AI) or they didn’t believe in large language models, generative AI, or extensive AI training. This group, while small, included a larger percentage of companies who’d actually adopted AI for significant business missions.
I think the pattern of viewpoints is due to the evolution in the level of actual AI experience, in the planning phase and then in deployment. Everyone comments on things that they know are new and think are interesting, but only when businesses start a real exploration can we expect the comments to mean much in a predictive sense. This is why surveys get things wrong so often; if you don’t validate actual familiarity in some way, you can bet that everyone will comment on any exciting new technology, but their comments will rarely matter more than what you’d get from a random encounter on the street.
The broadest comment from enterprises, made by 221 of 294, was that increasing the performance and reliability of the data center network was important. This group is larger than the number who explicitly cite AI as a driver to the data center LAN (133), and things like increased use of virtualization and greater traffic loads in general were cited, but equally important was the view that technology improvements should make it possible to increase capacity well beyond any (minimal) increases in cost. Modernization, then, is still an explicit driver of network change.
Enterprises didn’t have any significant number of comments on their broad networking goals, another indication that they aren’t seeing these goals changing. I think that even AI and SD-WAN are simply examples of things they hope to accommodate in the orderly more-for-less modernization. What might drive a more proactive, business-driven, model shift?
A business case, obviously. If we look only at the question of model-generative forces, we could speculate that the business case would have to create new network-connected workflows, which means a combination of new data and new data consumers. “New” here would mean previously unexploited, which likely means that the new data consumers would have to be drawn from the 40% of non-office-centric workers. New data would then mean information either not previously collected at all, or collected but not fully utilized, that associates with these workers. That would almost surely mean IoT data, or “telemetry” as some would call it.
OK, you’ve heard this from me before, but truth is truth. If we want to see a big jump in network and IT spending, we have to unlock a big new benefits pool. But there is perhaps a new wrinkle here. Of a separate group of 188 companies who reported use of IoT, 173 said they had a significant edge computing initiative dedicated to the control of real-time processes. Of that group, 73 said they collected all of this for analysis, 42 said they collected some, and the remaining 58 said they didn’t collect any of it except as a backup. Is all that data not being collected and analyzed truly useless? Could we examine it and look for productive uses? Might we then identify some new network missions? Worth thinking about.
