Bandwidth on demand: the in-demand model with no demand. Yet.

Utility pricing for networks and connectivity services is not a new concept. The idea is that customers should be charged based on what they use (e.g., per megabit of bandwidth) rather than being charged a fixed amount per month for a connection with a stated capacity (e.g., 1Gbps) that may not be used for much of the time (e.g., overnight) and that is not used at full capacity all the time.

There is also the problem of what happens when a customer needs more bandwidth, but only for a limited timespan (for example, periodic uploading of files to an off-site data centre). A flexible utility model should, in theory, also allow customers to increase bandwidth (at least within the limits of available local port speeds) at times of peak demand beyond normal usage.

The telecom industry’s answer has been various forms of bandwidth on demand (BoD) services for enterprise customers – usually available on layer 2 Ethernet connections. Such services have been available from multiple providers since the 2010s and offer the ability for customers to increase or decrease bandwidth (with perhaps an hour’s delay), with pricing adjusting accordingly. Many providers also offer the option for pre-programming increases/decreases in bandwidth on a given connection – e.g., moving from 1Gbps to 10Gbps overnight on Fridays. An alternative model is providing, say, a 1Gbps connection on a port that can support 10Gbps and then agreeing to a fixed price for 1Gbps and allowing overage to be charged at a higher rate.

However, GlobalData’s research has indicated that while enterprise customers say that they want bandwidth on demand, in reality, they rarely use it. Enterprise technology and services research director Gary Barton observed: “In our conversations with enterprises around 90% say they want bandwidth on demand services, but, when we talk with those who have BoD services available to them, usually less than 10% say that they use BoD with any regularity.”

What explains this apparent contradiction? Barton continues: “There are several key factors inhibiting the adoption of BoD services. It is very difficult for most businesses to fully understand when their peak demands will occur, so scheduling increases and decreases in bandwidth is rarely possible with an acceptable level of accuracy to guarantee services will always perform as expected. Furthermore, enterprises want predictable billing – an entirely on-demand model tends to increase the per-megabit cost while leaving enterprises open to potential bill shock. Hybrid models offering a guaranteed fixed bandwidth with the ability to go over that limit help when it comes to per-megabit pricing, but still do not prevent bill shock.”

The advent of Network-as-a-Service (NaaS) platforms and the growing impact of AI have, however, brought the bandwidth-on-demand discussion to the fore again. Is a changing landscape creating the market circumstances where BoD can flourish?

Barton notes: “Various providers have introduced NaaS platforms with access services that can be bought with no fixed-term contract and at hourly increments. There is no doubt that flexibility is valued by enterprises. In scenarios where cloud/AI workloads and (to a lesser extent) office locations may quickly change location, the ability to turn on and off connections, even high-capacity services such as wavelengths, will be more valuable.”

“It is too early yet to fully understand how agentic AI, in particular, will impact network demands. It is likely to change the density of traffic flows, with a greater need to increase upload capacity and greater flexibility. It is possible to see scenarios where agentic agents can automatically increase and decrease bandwidth across multiple connections on mesh networks. But this will need a lot of careful policy construction and enhanced monitoring capabilities, both to control network spending and for concerns regarding network security.”

Bandwidth on demand is likely to become more relevant in the AI era, but it is likely to need new billing models and platforms that can work with AI-oriented and AI-powered networks.