Nokia does a lot of interesting reports, and their 2025 traffic report and forecast is surely no exception. It documents the way that network traffic, and thus perhaps even network access bandwidth, has shifted from business-dominated to consumer content delivery. It has similar insights about future traffic, of course, and so we’ll look at the report and how it compares with what I hear from both operators and enterprises. I like the report for what it covers, but I’d like to see it cover more, and a bit differently in one specific way.
Let’s start by addressing a question, which is “Why do we care?” For telcos, for their vendors, there is nothing, and I mean nothing, as important as the way that different types of traffic combine to make up the overall load. If the future is a linear expansion of the present, same everything but more of it, then there is no real need for “new” services, only more of old ones. In particular, new features that a customer might be willing to pay for are less likely in this linear growth scenario. If, on the other hand, future traffic includes things that current traffic does not, then whatever is different about the new things might justify new features.
Let’s take an example. Everyone knows that the largest source of traffic today is consumer streaming video. If there was little or no growth in traffic over the next decade, other than growth in consumer video, then there’s no reason to think that future networks would need different service features. On the other hand, if we assumed that some sort of real-time stuff were needed, then we could assume that things like latency control and even packet loss control would be more important, so this new traffic might spawn new revenue to provide those features.
The report predicts global WAN traffic growth ranging from three to seven times that of today, reflecting a conservative/moderate/aggressive range. The moderate level, roughly five times today, seems reasonable to me, and if I were to bet on the others, I’d give the conservative 3x growth the lowest chance of coming about. The two figures presented on traffic first are useful in an overall growth sense, but they raise what I think is the critical point in traffic forecasting, which is whether the future is predictable by projecting the past, or whether it depends on the introduction of things that have never been, and whose impact is therefore very difficult to assess.
The first topic in the report is arguably an attempt to introduce the new, in this case, AI traffic. Consumer video is the main source of traffic throughout, but the report predicts AI will be the largest incremental driver. Whether this is likely true depends on a factor that tends to point out the limitations of this sort of traffic forecast. What is “AI traffic”? Is it the exchange of prompt and response, is it traffic relating to the delivery of AI-generated content, is it traffic to and from an AI agent handling some industrial/enterprise process or what? The report forecasts enterprise AI traffic and enterprise industrial traffic to grow significantly, but it’s not clear to me whether AI or industrial traffic enterprises generate will directly impact WAN traffic. So far, enterprises don’t see significant AI impact on their WAN.
The report does classify AI traffic as “direct” and “indirect”. However, I am uncomfortable with the “indirect” category; it essentially says that AI traffic is any traffic that would result from the use of AI beyond direct AI exchanges, the traffic difference in an application with or without AI deployment is “indirect” AI traffic. That seems too broad and imprecise to be useful, IMHO, and I think it results in overstating AI’s impact on WAN traffic. Even with this definition, though, the WAN AI Traffic (Figure 1) shows that AI traffic even in 2034 won’t be as large as non-AI traffic for either consumers or enterprises, which I think is true.
But to get back to our opening point, so what? The fact is that the great majority of things that could likely drive AI traffic growth won’t generate traffic that looks much different from what we see today. Chatting with a bot is like chatting with a human, except insofar as the bot might generate in-cluster traffic to look like a human. That doesn’t impact the WAN, or create opportunity for the operators. The fact is that any shift in the nature of traffic has to come not from a technology like AI, but from applications whose QoS demands are dramatically different. AI, in the end, can be either a new solution to traditional IT challenges, or a means of addressing new ones. The former would inherit the QoS needs of the traditional challenges, and thus wouldn’t generate anything new for operators to charge for.
Let’s look at another section of the report to see whether we can find some revolutionary new traffic. “Consumer mobile traffic growth” is cited as ranging from 12% to 18% CAGR, with the segment showing the most dramatic growth being conversational AI. Well, consumers won’t pay for that, nor will they pay for a premium service to deliver it. If this is it, new-traffic-wise, then consumers won’t play in, and since consumers dominate traffic today, that to me means nothing will work, nothing will help. Forget 6G; there’s no revenue coming from it.
But the second-best growth source, while contributing little, means a lot. “Metaverse” contributes nothing visible to traffic today, but contributes about 4% in 2034. Why focus on something so trivial? Because a metaverse is like an iceberg; most of it is below the surface. A metaverse is a game or a video unless you assume that it’s a way to visualize an augmented real world activity. To make that work, you need a world model, but before we get to what that means, let’s look at the “Enterprise and Industry” picture. Once we do, you’ll understand why I want to come back to world models, a blog topic yesterday.
Enterprise traffic today is largely related to traditional applications, to doing and running a business. Today, based on what enterprises tell me, it’s roughly 75% planning and reporting or promotional and pre/post-sale support and 25% transactional, meaning actually doing something to alter the state of business data. There is no consequential IoT traffic, no XR stuff. The report doesn’t give the breakdown of current enterprise traffic, but by 2024 it posits that IoT and VR/AR will make up 45% of traffic, a staggering growth level. However, I think that most of the roughly 30% IoT is really transactional information from retail terminals, my 25% in today-traffic. The 27% XR is all new, but from what?
From real-time applications, IMHO, so it’s now time to get back to the consumer world model topic. Any real-world, real-time augmented reality application has to be tightly coupled to the real world. By “tightly” I means that is has to have an up-to-the-minute (actually up-to-the-milliseconds) view of a lot of stuff, so you need real IoT on a massive scale. You also need data to be sent to your XR elements, and you need something mighty like AI to process all of this. The report predicts significant growth in “agentic AI” traffic, which enterprises (and I) agree with, but it’s not clear just how much of this actually reaches the WAN unless you presume it feeds a world model set with wide geographic scope.
I think the report does a great job of projecting trends, but where there is so far no trend to really project, that approach is too speculative. We won’t realize a shift in traffic balance, source, or QoS requirements through the orderly evolution of what’s done today, we’ll realize it through the creation of applications, support of missions, that we don’t have and can’t do today. I think that’s the one issue the report doesn’t address. What new thing drives all consumer, all enterprise, change? Real-world, real-time. There is no other possibility, and so if you want to look out a decade, you need to look at how, when, and whether that RW/RT stuff will evolve.
