Sometimes, there’s an important grain of truth buried in…well…crap. So it is, I think, with the story of the great opportunity AI offers to the telcos. To be clear, AI is not a great opportunity for telcos. It’s not even clear if it’s a great opportunity for anyone, particularly the three cloud giants that lead the current charge. But what it might be is a signpost to what a great opportunity would look like, and for telcos these days, that’s enough hope to be important.
Why would everyone be hyping up telco AI opportunity? OK, yes, part at least could be that everything AI gets hyped up, and everything is looked at a candidate for AI hype. Why is the telco vertical such a candidate, though? The only possible answer is that telcos need to find something to make money on besides routine connection services. That, in itself, is at least making the AI telco wave a validation of a real need. One question it raises is whether AI is more than that, and perhaps another question is whether AI is the only “more than that” out there. If it is, then might there be steps to consider that would open more than one potential new service opportunity. Is there an infrastructure of the future that we don’t have to assign to a specific future?
To try to get at these questions, we need to look at just what a telco role in AI might be. Well, one thing we keep hearing about is that telco use of AI for internal missions would then build an infrastructure base telcos could leverage in other AI service missions. AI-RAN, we’re told, is a stepping-stone, but stepping stones have to lead your steps to somewhere you want to be. The problem with AI is that it offers only three destination options, all of which have major problems, and so does that first stepping stone.
The stepping-stone problem is the familiar confusion of what AI can do with what justifies AI. Operators have told nearly everyone they don’t want another 3GPP-driven fork-lift upgrade to infrastructure. We don’t have AI-RAN now, so clearly it would require an upgrade, and a fork-lift? Maybe. The point is that there’s resistance to spending on the next generation of mobile infrastructure when the last one didn’t pay back. So it’s more likely that vendors would push AI hype to improve the prospective business case than because AI could actually do the improving. Then there’s those three AI destination options and their issues.
First, the GPUaaS story. This comes up all the time, and in theory there’s some financial sanity behind it since telcos have a very low IRR, and thus can afford a low-margin play. The problem is that it’s far from clear that GPUaaS is a viable opportunity. Not only is it not clear whether enterprises would see it as a low-cost-high-data-sovereignty solution to cloud AI problems, it’s not clear whether the current AI giants would address that opportunity if it did emerge, leaving telcos in a race to the bottom.
Second, the telcos could get into the AI cloud business. This would have the advantage of higher margins, and enterprises trust telcos more with core data than cloud providers, by almost a 3:1 margin. But could telcos frame a set of AI services that would be competitive? There are only three of the Big Three to concentrate development in, and over a hundred telcos who largely can’t cooperate for regulatory reasons.
Finally, there’s the elusive edge. What makes it elusive is that the talk about “edge computing” takes on the classic “Field of Dreams” look, where you propose a technology advance with the theory that if you can deploy it, someone will figure out what to do with it, and then pay to do it.
Fact: You can’t justify edge computing without validating real-time applications. We already have “edge computing” to support real-time process control missions in factories, warehouses, etc. To presume that we can make a telco service opportunity from edge computing demands we find real-time missions where this self-hosted “local edge” doesn’t provide what we need. That would mean one of two things. First, you have a real-time process that spreads geographically, so a local edge is impossible. Second, you have a real-time application that involves a combination of compute complexity and frequency of use that makes owning your own edge financially unreasonable.
We have some limited and ineffective market validation of the first point, with self-driving vehicles. Here, the problem is that the vehicle itself can host much of the process control mission itself, and in fact the demands on latency and availability are too high to cede this edge mission to some shared and remote resource pool. Still, I believe there are applications in the production and transportation areas, as well as in utilities, where it would be possible to eke out a value for hosted edge services.
The second of our two hosted-edge opportunity areas is likely to be something involving people, consumers, rather than companies and their workers. I’ve often blogged about a future point where a consumer and phone moved through a series of “information fields” that represented available products and services, interacted with these fields in some way, and then took some action. The action might be simply telling the consumer that something they were looking for was available at the best target price just ahead on the right, and it might be as complex as ordering the “something” and directing the consumer to pick it up. The problem with this is that it would likely take a pretty broad community of investors to assemble something valuable enough to attract users, and of course this means more risk that a single non-player would derail the whole thing.
There is an interesting take-away here, though. All the credible opportunities for telcos involve “edge computing” of some sort. While AI hosting is almost certainly something telcos can’t exploit to improve profits, it is an example of an edge application, and it’s at least credible to assume that some of the future edge needs would include AI. All that means that we could in theory validate at least some of an AI investment with something real build on one of these edge opportunities.
I have always believed that the extreme edge is not a good target. Yes, you could stick AI-RAN in a cell site; you’d almost have to, in fact. However, you don’t get much economy of scale with that, and you don’t really have a great location to converge traffic from a wider area. Metro placement, meaning placement in a centralized location where access network technology would naturally concentrate traffic, makes more sense. There, you could deploy hosting, and so there you’d need a combination of a data center and a WAN aggregation point.
How unified this needs to be is, so far as operators are now concerned, an open question. Certainly there would be a value to having operational unity, the ability to manage the equipment on a common platform, but could even deeper integration be helpful? A half-dozen operators have, at times, commented to me on having a “metro platform” built on a single standard technology base. AT&T has, perhaps, talked most in public about the value of this sort of thing, and DriveNets is a vendor who actually offers it, though their current marketing is (no surprise) more focused on AI missions than on generalized metro.
I think that if there’s a new future service set to boost telco profits, it’s one that’s hosted in metro locations. I think that a metro-centric infrastructure plan is what telcos need to be looking at, and AI is simply a potential technology to further that plan. AI that preps for something real is better than AI that depends on the realization of the hype, and even on that realization creating a very broad opportunity. Might that happen? Sure, but is it a good strategy to hype up an AI story and hope something comes along that redeems it?
I think you can see some of the signals I’ve noted here in Meta’s “manifesto” on AI. The simple truth is that for nearly all vendors, the revenue booms of the past are almost ancient history. They’d love to have another, and this yearning feeds their participation in virtually any hype wave that presents itself. So what if AI is hype? Believing it buys us some time, and maybe during that time something real will come up. If not, well, we’ve never had a shortage of hype.
Of course, the other half of the message of this post is that while there’s always hype, there’s often some reality available to exploit, if hype starts to look a bit thin. In telco-land, it looks pretty clear that hype-thinning is happening now. In other areas of tech, there are signs as well, and the “elusive edge” is likely the converging point for any real opportunity, for any player in tech.
