Omdia has produces another very useful report, this one on telecom opex, and I think there are some insights we can draw from it, especially if I tie in what I hear from the telcos themselves. In particular, I think there are some important lessons regarding telco use of AI. There’s a useful analysis of the report HERE, and I’ll cite some of the numbers from it as well.
Opex is important to telcos, because most face persistent challenges in sustaining return on infrastructure, and have little or no pricing power or differentiable features on which they could pin hopes of better bottom-line results. Cutting capex is challenging, since most of it sustains competitive market positioning. Cutting opex is therefore essential, but the questions are “where” and “how”. The “how” question raises the notion that AI could help.
Telco 2024 opex -to-revenue (excluding stuff like depreciation, amortization, and one-time write off/downs) ratio improved versus 2023, falling from an average 67% to 65% (the exact number depends on how the operator balances mobile, wireline, business, consumer, and TV services), which is well above capex, and of course opex is the place most expect AI to improve things. As Omdia notes, we’re still in the “thinking-about” phase of AI application in telco-land, so the fact that labor-related opex is down by 2.2% can’t be attributed to AI, and the fact that last year’s labor-related opex was up 4% makes the decline this year notable.
What I’ve called “process opex”, the cost of network operations, server operations, and other ops tasks, is shown by the graph to be a pretty small part of opex, and of human opex as well. That number has fallen by about half in the rough two decades that operators have been working to control costs, and as I’ve noted in previous blogs, it’s far from clear that there’s a lot more cost to be wrung out there.
Interestingly, process opex is also a field where modern language-model AI is competing with earlier forms, particularly machine learning. Process opex is largely event-driven, so you can’t tolerate huge latencies associated with complex LLMs, nor does it require that sort of knowledge base. ML is firmly established in the netops area, and so what’s increasingly left are tasks that actually require manipulation of things in the real world, which is not only intelligence, but hands-on, with the operative word being “hands”.
Could AI replace humans totally? I know some believe it could, but I’m skeptical. There are still a lot of hands-on tasks related to installation and repair that, to do without human hands, would require autonomous robotics. There are even more tasks that operators tell me could not be totally ceded to AI without taking a major risk. Autonomous operations, to the extent that it’s first practical and second reliable enough to trust, is still a ways off. Maybe not as far as autonomous robots, but far enough to mean that near-term justifications for massive AI investment would be difficult.
Given that, what hope do operators have for actually cutting costs? Well, strip away the network-specific stuff from telco opex, do the same with product/service-specific stuff from enterprise opex, and what you have left is the costs of running a business, selling, supporting customers, paying bills and people, the traditional stuff that all companies share no matter what their business is. If AI can help enterprises, focusing on this sort of thing, then it can do the same for telcos.
The problem we have here is both simple and complicated. It’s simple in the sense that enterprises are already groping their ways through the AI jungle to find the really valuable approaches. It’s complicated because enterprises are groping because the value of AI isn’t in the AI we hear about, but in another sort of AI that’s just emerging. I blogged earlier this week about the two different kinds of AI, and about how enterprises are finding value in the second kind, the expert, componentized, AI agent concept.
Microsoft is seeing this. Many of you have heard about their comment that AI will kill off SaaS by 2030. The CEO said it last year, and the VP heading up business applications and platforms is saying it now. What they suggest, as the article notes, is that AI is going to restructure business processes themselves, and thus break the traditional workflows that bind people and applications today. Those workflows are what enterprises think need to be the focus of AI agents, so I don’t think it’s unreasonable to say that these comments show that Microsoft realizes the value of AI isn’t going to be realized by simply answering workers’ questions (or emails) or helping them write reports, but by doing business.
The “business agents” Microsoft describes are what enterprises have been groping toward. OK, Microsoft sees these agents as being hosted by (well, you guessed it) Microsoft in vast AI data centers. Did we think they’d simply admit those massive capex excursions lead nowhere, profit-wise? The problem with that view is that players like Microsoft also thought that everything would move to the cloud, so believing today that everything will move to AI seems a titch opportunistic. However, it makes sense to prepare to justify your enormous AI investment, and more sense to do that by (finally) connecting to a real trend.
What’s important here is that the hype wave of generative AI hosted by cloud giants may be threatened. Markets this week starting showing some angst over the valuations of AI companies after some earnings report comments that seem increasingly to demonstrate investment driven by hype and hope and not business justification. So we are indeed, as a recent Gartner state-of-the-hype report suggested, seeing generative AI fall into a trough of disillusionment. But is agentic AI simply trailing on the same path, as Gartner suggested? If we think that the AI agents enterprises want will simply end up running in those same massive data centers, then yes. But enterprises don’t think that. They see issues of data security and sovereignty, and cost, as dictating that the core business functions they depend on will continue to run on their own premises. AI, including Microsoft, will have to contend with that. Telcos can take cautious hope from it.
Cautious, because telcos, always conservative, may be caught in a trap here with regard to opex. There’s no reason to think that they don’t have the same opportunities for AI savings in “core business” opex as enterprises have, but that opportunity seems to depend on AI agents. AI agents may be a publicity casualty of AI hype, and can a conservative telco bet on a technology that Gartner sees as poised to fall from grace? Especially since Wall Street is also showing unease? That’s a heavy lift.
In the near term, it is. The good news is that enterprises seem to be pursuing agent plans regardless of whether they fit in the AI mainstream of PR. There is resistance from conservative management, and there are technical issues regarding cost, traffic, security, and sovereignty that need to be worked through, but enterprises are finding and addressing them, and even the big cloud providers are seeing them, which may tune the coverage of AI more to the true, practical, agent issues. It probably won’t all shake out this year, but I have hopes for 2026.
