Here’s an important truth for you; for telcos, AI is dangerous. And no, it’s not the security issues or hallucinations, or the fact that the technology is expensive, it’s that the “promise of AI” is, for telcos, largely a trap. Few, if any, of the initiatives that are promoted for the telco community have any real chance of making a substantial difference to telco profits, and that’s what they need to be focusing on. AI is distracting them.
There are three broad things that telcos themselves say AI could do. First, it could improve operations to lower costs and raise service availability. Second, it could introduce autonomous behavior to things like RAN, and so reduce capex and perhaps even enhance service at the same time. Third, it could open up new services to offer, including perhaps AI hosting or edge computing. The questions telcos, and that we, have to address are whether each of these is true, and whether the impact could be meaningful.
All of this has to be considered in light of another basic telco truth. Telcos are not motivated by opportunity, but by competition. They value what they have already, they’re focused on preserving it, and they don’t see themselves stretching out to grab for new stuff, even if that doesn’t risk their preserve-the-present goal. They see evolution of infrastructure driving service adoption, not new adoption justifying a change to infrastructure. We’ll come back to that regularly.
OK, lowering opex. AI can do that, but the problem is that the majority of opex cost for telcos has always been headcount, and efforts to reduce that have been going on for decades. Operators tell me that they think that, on the average, they’ve already reduced headcount by “ninety percent” of what can be realistically cut. There’s not a lot left for AI to do; perhaps enough to cut a bit less than two percent of total network/process opex, which could boost profits by “a bit more than half a percent”, if it could be accomplished at zero cost, and by less than a tenth of a percent if you assumed they had to build out AI infrastructure or buy AI services from a cloud provider.
The most credible impact of AI in the area of costs is, according to telcos themselves, the customer support chatbot. The problem is that support portals have been in play for a long time, and while it’s likely that AI could improve them, it’s far from certain that they’d cut support call center costs significantly, because the current portals seem to have hit the wall in terms of handling things that can be handled without human interaction. There’s also a question of whether developing an AI support chatbot could be done, and run, with up-to-date information. It’s a big project, a very big project.
The second, capex-efficiency, benefit is interesting. Could AI design a better network, could it manage spectrum and cell assignment/placement in a RAN? Again, almost all telcos think it could, but well over two-thirds are still unsure on the “how?” the “how much?” and the “how costly?” pieces of the puzzle. Almost half also say they’re worried about AI risks, given that there is no industry consensus on any of these points and that AI horror stories abound.
A big part of the better-network problem is the declining cost of bandwidth, and the very limited willingness of consumers to pay for more of it. One classic question is evocative; if the unit cost of bandwidth were to fall to near-zero, what value is there in route optimization? OK, it won’t fall that far, but could it fall far enough to render optimization benefits less than compelling? Particularly given AI cost and risk?
Operators are reluctant to openly rain on the AI RAN parade, but privately are doubtful that the benefits are significant. “There are a lot of things that can save money or improve services in theory, but haven’t been proven out in practice”, one telco told me. As THIS article shows, they’re not exactly jumping to join the AI-RAN Alliance, and many think that the whole initiative is little more than the latest attempt to resurrect network slicing. Most operators tell me that if anything is to be done here, it would have to come as part of the 6G evolution, while at the same time expressing doubts that the 6G evolution itself offers much to justify an investment.
Which leaves us with the old “new-revenue” story, or stories. Network slicing is one such story, and IoT, real-time services, edge computing, and even AI hosting are others. As I’ve already noted, nothing puts a telco exec to sleep faster than a presentation on any OTT-like service, and many doze off with any mention of something more radical than a VPN.
This fear of newness ties in with 6G fears, which in turn tie into 5G disappointment. Operators did a big upgrade cycle to get to 5G, and were promised all kinds of new opportunities, arguably none of which were delivered. Why would any of these new services be any different, they wonder, and that’s a fair question. In fact, the big problem with all the supposed new service opportunities is that telcos themselves have little or no chance of bringing them to realization on their own. All require a complex ecosystemic web of players, all of whom have to take risks and see a proportional reward in order to move, and any failure of a piece fails the puzzle as a whole.
Does all this justify my opening point, that AI is a danger to telcos? I think it does. Among telco CxOs, it’s rare for me to hear from one who doesn’t feel a sense of crisis, a fear that the current service model isn’t sustainable and that something new is required. Might these telcos jump on AI for its undisputed novelty, and it’s obvious popularity among media and financial types? Sure. But if none of the AI directions being reviewed show much chance of leading to a positive outcome in terms of service-model transformation, then novelty and popularity aren’t going to be enough.
