Shakespeare said “There is a tide in the affairs of men which, taken at the flood, leads on to fortune; Omitted, all the voyage of their life Is bound in shallows and in miseries.” No offense to the Bard intended, but the problem with the tide/opportunity analogy is that tides happen more than once. You get another chance. So it may be with telcos and AI. While they surely missed the early AI wave, there are major changes underway in the space, and these could give telcos another shot at riding the flood.
I think that when OpenAI lead the generative AI charge, and when other giants jumped in to make massive AI data center investments, these companies all believed that there was substance to AI but had no real idea what the actual business case would be. They jumped in, taking a risk that they’d be able to adapt when the real AI value emerged. Enterprises were, from the first, dismissive of the AI-as-a-Service model. One said that they saw it as equivalent to an attempt to push cloud computing as the only kind of computing, as “time sharing” did in the 1960s, before massive changes in hardware made computing more personal. They liked what’s come to be known as the “agent” model, though their view of what an agent is and does is a bit different from the popular discussions would suggest.
AIaaS, as we might call it, is a sort-of-Hal-model, an AI expert that answers your questions and checks your work. It’s generative, but also general. Agent AI does things. How autonomous it is depends on how you define autonomy, but for sure it has a specialized ability to produce something based on instructions either given explicitly (as we do with AIaaS) or implicitly in a predefined response to an event. Agents are tools, not models.
Model hosting has already been done, by those giant data-center investors. Agent hosting is now emerging, and interestingly it’s OpenAI who may be doing the early signaling. Their deals with Oracle and AMD, for example, could well be reflecting a realization that agent hosting doesn’t look like an AIaaS data center, but more like a cloud computing framework. You don’t need super-GPUs, you need a lot of inexpensive ones. Agents are not experts from afar, but partners who lend a hand, a specialist hand, and take on tasks.
There’s another difference with the agent model, too. With AIaaS, you don’t expect Hal to give you an answer immediately. You do a pilgrimage to the oracle and burn some incense, then ask and wait while the oracle ponders and responds. Agents have to be able to do things in real time. In hosting terms, that means they have to be low-latency applications. Right now, enterprises achieve low latency computing by placing the hosting resources proximate to the need; “edge” computing is computing on premises. Not all computing today is done that way, of course, so we could expect agent hosting to develop a cloud component, but one with lower latency capability. Edge as a service, in short, and that’s where telcos could come in.
Telcos have roughly 25,000 remote facilities in which something could be hosted, of which roughly 12,000 are big enough to support racks of servers (if you want details, there are companies like GeoTel who can provide them). Even smaller ones could offer some hosting capability, though, and a single rack might be enough to host a very latency-sensitive agent element needed by a few applications proximate to one of those smaller locations. These centers are also hubs that concentrate access connections, which means connection latency to them is low. Telcos in the US, then, could offer tens of thousands of edge hosting points.
And not only potential points, because the facilities exist. Agent hosting is likely to develop based on a variety of user demand and AI technology initiatives, and just where they combine first and in the largest amount will vary. Who can rent space and prep it fast enough to fulfill a need? Nobody. Who has space available, space that can be prepped quickly and is ready to be connected? Telcos.
Who, historically speaking, is likely to get stranded on the sand with this tidal opportunity? Telcos. Of the 88 I have chats with, 72 can see this opportunity in at least some form, but of that number, 60 say that they’re looking for someone like OpenAI to come along and cut a deal for the hosting. Which means that they’re looking for a senior partner to disintermediate them. What they need to do is to take control, and there are two paths they could take to do that, one easier and less profitable, and one harder and more decisive.
The easy path is to focus on partnerships not with the model giants but with emerging players in AI agent tool software. That doesn’t demand telcos take much of a role in creating agents, or even defining what’s needed to host them. It does demand that they assess the opportunities and pick partners who can address the things they’re most likely to find concentrated where their facilities are available. Take health care as an example. Of roughly 6,300 hospitals in the US, over six thousand are within a mile of a potential telco hosting point, and over five thousand within a mile of one of the 12,000 larger facilities. There are other verticals similarly concentrated. This path lets telcos cut agent deals and deploy only when and where they have them.
The hard path is to determine what’s needed to host agents, deploy it in select locations, and then solicit or select partners to run their agents on the platform. With proper marketing, the telcos could expect AI agent providers to come to them, which makes them the senior partner in the deals. While this does require more market research, agent partner research, and upfront investment, it significantly reduces the risk of disintermediation.
But it doesn’t eliminate it, for two reasons. First, any telco initiative that aims to reduce disintermediation risk is going to generate offers from AI companies that pull the telco back to a subordinate role. Since this is what most telcos want in the first place, it’s very likely that they’ll backslide into their classic risk position. Second, telcos tend to move at a pace that makes glaciers seem speedy, so their early-stage efforts will surely tip off others, particularly cloud providers, web hosting companies, CDNs, and data-center-as-a-service players like Equinix, who could then simply move faster.
AI agent opportunity is another flood tide opportunity, but other tides may ebb and flow too. Will telcos decide to wait for one of them? If they do, isn’t that decision alone a strong reason to believe they’ll procrastinate through future cycles as well? Wait long enough, and the advantages telcos have may dissipate, just like the moon may wind away from earth and the tides cease. Agent opportunity may be the last real shot telcos have at the AI space, at a time when AI is trying to transform into a real value play. I think it would be smart for them to grab a boat and launch.
