There was a moment at FutureNet World where you could almost hear the industry collectively nodding. Not because anything particularly shocking had been said. Quite the opposite. Everything sounded inevitable. AI will run the networks. OSS will disappear into something elegant and frictionless. Engineers will be replaced by “intent”. You will no longer fix problems because, apparently, problems simply won’t happen. Marvellous.
Well, it is, on paper, the telecom equivalent of a self-driving car that never crashes, never breaks down, and politely refuels itself while you sleep. The sort of thing that makes for a very strong keynote and a slightly weaker balance sheet.
The central idea is simple enough. Networks are getting too big, too fast, and too complicated for humans to manage in the traditional way. So, we hand the whole thing over to automation, layer in some AI, and let the system quietly optimise itself into perfection.
No tickets. No outages. No engineers frantically rebooting things at 3am (been there, done that). Just outcomes. You can see why it is appealing. The difficulty is that telecoms has been here before. Not with AI, admittedly, but certainly with the promise that the next wave of technology will finally line up cost, complexity, and revenue into something vaguely resembling a business model.
Unfortunately, it rarely does. What actually happens is rather more predictable. You remove one set of problems and replace them with another, slightly more expensive set that is harder to explain to the finance director. The engineers disappear, and in their place arrive platforms, vendors, integrations, and a growing collection of things that “just work” until they very much don’t.
One must remember automation doesn’t remove cost. It moves it somewhere less visible. And invisible things, as a rule, are where trouble likes to hide.
There was also a lot of talk about making the network “programmable”. This is the bit where telecom operators finally become platforms, rather than pipes. APIs everywhere. Developers building on top. New services appearing as if by magic.
If this sounds familiar, it should. The industry has been trying to do this since at least the early 2000s, with all the commercial success of a man trying to sell umbrellas in the Sahara. This time, we are told, is different. AI will make it real. Networks will respond dynamically, services will be spun up in real time, and enterprises will happily pay for capabilities they didn’t know they needed five minutes earlier.
Possibly. But there is a slightly awkward question that hovered in the background. Who, exactly, is going to pay for all this? Because while the technical direction now feels broadly agreed, the commercial side still looks like it has been sketched on the back of a napkin during the coffee break.
Enterprises say they want flexibility. They say they want performance. They say they want resilience. All true. What they don’t consistently say is that they are willing to pay a meaningful premium for a programmable network when a perfectly adequate one already exists at a lower cost.
That gap matters. From our Research perspective, this is where things get interesting, and slightly uncomfortable. The industry is investing heavily in solving an operational problem, but the demand-side validation is still catching up. Which use cases justify autonomous networks at scale? Which sectors will pay for dynamic, API-driven connectivity in a way that moves revenue, not just PowerPoint slides? And where does AI shift from being a cost-control mechanism to something that genuinely drives top-line growth?
At the moment the answers tend to involve a lot of confidence and not nearly enough evidence. There is also a second, quieter risk. If every operator builds roughly the same “invisible” network, powered by similar AI models, orchestrated in similar ways, then differentiation starts to evaporate. The cleverness moves under the bonnet, where customers can’t see it, and the buying decision drifts back to the usual suspects: price, coverage, and whoever answers the phone first (and yes answering the phone quickly is important).
Telecoms have spent years trying to escape that particular trap. This does not obviously fix it. In fact, there is a decent argument that it makes it worse. When infrastructure becomes invisible, value tends to migrate upwards. To applications. To platforms. To whoever owns the customer interface and history is not especially kind to infrastructure players in that scenario.
None of this is to say the industry is wrong. It isn’t. The current model is creaking. Networks are too complex, margins are under pressure, and manual operations at this scale are about as sensible as maintaining a modern jet engine with a spanner, gaffer tape and a hopeful expression.
Automation is necessary. But necessity and profitability are not the same thing. FutureNet World made it very clear that telecoms knows where it is going technically. Autonomous networks, AI-driven operations, seamless orchestration. All that feels inevitable.
What remains less clear is whether the industry is building something customers will actually pay more for, or simply something that is more efficient to run. Those are not interchangeable outcomes. Because if this all works exactly as planned, we may end up with networks that are faster, smarter, and beautifully invisible.
Which is excellent. Right up until someone asks how they make money.
By Peter Zanatta
