Predicting the long run is difficult.
As an analyst, that’s precisely what individuals hope I can do. After all, nobody can predict the long run, however we will mannequin varied outcomes given the perfect knowledge we’ve available.
So utilizing the info, let’s make a prediction about the long run impression of AI on enterprise networks.
There are a lot of methods to method this, however the important thing space I’ll tackle on this weblog sequence is the impression of AI on enterprise community configurations. The 2 key areas the place our knowledge are helpful for this are predicting the budgetary impression of:
- AI-driven bandwidth demand will increase.
- The continued transfer away from office-to-office connectivity to connecting customers to belongings within the cloud.
I’m not going to handle how automation will change community administration itself, however quite how these above modifications impression the whole value of possession (TCOs) of ensuing WAN configurations.
Deep studying and LLMs require massive transfers of information for coaching and inference. Palo Alto lately launched a report that they noticed an 890% enhance in bandwidth demand final yr. That’s huge! After all that’s largely between knowledge facilities, however the implications for any department workplace or website are nonetheless clear. Nokia predicts that AI-specific site visitors will develop at a CAGR of 24% a yr by way of 2033. A lot of the work of AI is occurring within the cloud, making conventional MPLS architectures with centralized breakouts inefficient and sluggish. To not point out costly
Whereas we will’t predict a lot in regards to the impression of AI on particular strains of enterprise, we can definitely have a look at how massive will increase in bandwidth demand and cloud-centric hybrid architectures impression community prices. This weblog sequence will tackle modeling a number of potential AI-age community configurations and the way they’ll impression budgets. I do that on a regular basis for patrons of our WAN Value Benchmark platform, which is the instrument we’re utilizing to do that modeling.
Modeling Community Configurations
We’ve been operating varied hypothetical eventualities since 2019, however in 2025, we would have liked a serious revision to maintain tempo with these new AI-driven configurations. This evaluation will have a look at an ordinary and high-capacity MPLS community and examine that to:
- a Conservative Tiered Websites WAN,
- an ordinary Tiered Websites WAN,
- a pared-down Distant Employee Hybrid WAN,
- an Web-First WAN,
- and probably the most radical revision, a Hub and Spoke WAN.
Aside from the Distant Hybrid WAN, all the hypothetical community eventualities have the identical bodily geography throughout 150 world websites. Our intention once we created this unique hypothetical community in 2019 was to seize a smaller enterprise/multinational community (and to maintain the comparability work simpler for our analysts!) and to ensure that it captures developments in key markets all around the world. Our community contains seven knowledge facilities among the many 150 websites in key regional knowledge heart hub cities:
- Tokyo
- Hong Kong
- Frankfurt
- London
- Ashburn
- San Jose
- São Paulo
We don’t specify a vertical for this hypothetical enterprise; nevertheless, it’s doubtless closest to a enterprise providers firm (assume consulting, accounting, legislation agency) as most websites are in a key metro’s downtown or suburbs. This doesn’t characterize an business with rural or distant websites which have very lengthy entry strains; most of our entry strains are within the 0-5 km and 6-15 km ranges. Whereas our websites fall principally within the U.S. & Canada and Western Europe, the community spreads throughout the globe.
Proportion of Websites in Every Subregion–All WAN Situations
Notice: Every part represents the whole contribution of the listed area to the worldwide WAN measurement.
Hypothetical Community Map
Twin MPLS WAN
Our baseline or management Twin MPLS community is supposed to characterize what enterprises are abandoning: an all-MPLS community with active-active twin ports and totally redundant entry strains. The geographic distribution of the common bandwidth obtainable at every website featured under displays the whole bandwidth availability of the community and the way websites are distributed inside every area. We’ll use this chart to check modifications to the community configurations and the way they impression obtainable bandwidth. This chart represents the whole bandwidth of all underlay merchandise at every particular person website, averaged throughout the complete subregion. For the beginning community that’s merely twice the MPLS port we chosen as it’s an active-active MPLS community.
Common Whole Bandwidth per Web site in Every Subregion—Twin MPLS WAN
Notice: Every bar represents the common whole website bandwidth, together with a number of ports or underlay providers, throughout all websites within the listed subregion.
- East Asia has the very best bandwidth, with a mean of two,868 Mbps, as a result of it has two knowledge facilities (DCs) and comparatively fewer websites than Western Europe or the U.S., which additionally has two DCs.
- Equally, the common in Latin America is introduced as much as a mean bandwidth of 907 Mbps by the Brazilian DC.
- Different areas present a lot smaller common bandwidths, particularly these which can be costly. Within the community design, we stored value in thoughts to mirror the fact that bandwidth could also be restricted in high-cost areas.
The common bandwidth per website globally is 1,298 Mbps; nevertheless, that is introduced up considerably by the massive capacities on the knowledge facilities and some massive headquarters or campuses. Typical websites are a lot decrease, within the 10-200 Mbps vary.
Distribution of Whole Web site Speeds—Twin MPLS WAN
Notice: Every part represents the share of whole website bandwidths that fall inside every bandwidth vary.
- Because the WAN has active-active MPLS ports, the whole website pace is definitely twice that of the one port, which is the quantity we concentrate on to find out the obtainable bandwidth at any given website or throughout the community.
- Solely about one-quarter of websites have a complete bandwidth of 100 Mbps or much less.
- About 15 websites, or 11%, are above GigE. These characterize the seven knowledge facilities and some massive regional hubs.
Excessive Capability Twin MPLS WAN
Our beginning management community would show strained underneath AI-drive bandwidth calls for, so we added an extra management community. Within the MPLS setting, many enterprises would renew on a three- and even five-year foundation and maybe solely then tackle rising bandwidth necessities. This extra management is in place to display the situation the place maybe a WAN supervisor in 2025 was renewing a community that they had bought in 2022 and now wanted to spice up bandwidth by at the very least 50% throughout the board. If this enterprise was not but in a position to undertake SD-WAN/SASE and nonetheless wanted MPLS at each website, we needed to see the price impression of a bigger MPLS community.
Common Whole Bandwidth per Web site in Every Subregion—Twin MPLS WAN vs. Excessive Capability Twin MPLS WAN
Notice: Every bar represents the common whole website bandwidth, together with a number of ports or underlay providers, throughout all websites within the listed subregion.
- The common bandwidth per website comes out to 2,133 Mbps, which is a 64% enhance over the unique MPLS community.
Distribution of Whole Web site Speeds—Excessive Capability Twin MPLS WAN
Notice: Every part represents the share of whole website bandwidths that fall inside every bandwidth vary.
- We now not have any bandwidths within the 11-20 Mbps vary.
- The smallest bandwidth vary, 21-50 Mbps, has dropped from about 13% to three% of the worldwide WAN.
- The share of websites falling within the bandwidth vary 501-1,000 Mbps has decreased by about 1%, now accounting for 13% of websites.
- Round 25% of the websites now have a bandwidth bigger than GigE.
So, how do these bigger bandwidths impression value? Nicely, the price per bit goes down with greater capacities, so whereas we anticipate prices to go up considerably, we didn’t anticipate prices to go up by as a lot because the bandwidth enhance.
Unique Twin MPLS and Excessive-Capability Twin MPLS Situation TCOs
Notice: Every column represents the whole annual value of possession for that WAN situation, damaged out by product in every colour part.
- Whereas the common bandwidth obtainable per workplace went up by 64%, the TCO solely rose by 43%, going from $8.1m to $11.6m.
- Whereas the unit value benefit is obvious, this may nonetheless characterize a drastic change in finances for the community to accommodate the rising bandwidth calls for of the fashionable enterprise.
Whereas nobody would need to enhance their community finances by 43%, that isn’t the top of the story. An all-MPLS structure is inefficient for internet-driven cloud site visitors, and previous class of service know-how can’t benefit from software particular coverage and efficiency administration when enterprises undertake SD-WAN.
Within the subsequent posts we’re going to have a look at varied hybrid eventualities that benefit from SD-WAN, native breakouts for higher cloud connectivity, and elevated bandwidth.
