So I’m making something out of thin air, here, but given AT&T’s good second-quarter showing – where the US firm is using fiber to sell 5G and 5G to sell fiber, and to raise margins, free-up cashflow, and back itself to invest in its own rising stock; and where it is engaged in a mega five-year fiber build-project, to expand its footprint by about 50%; and where its chief, John Stankey, has just declared fiber as the king-tech for connectivity – then it is probably worth discussing what comes next, after the groundwork is laid. Because Lumen – which, ironically, has just sold its consumer fiber-access business to AT&T – is on a tear-up in the AI economy already, and it has got some kind of a nod from Gartner as the “company to beat” for “connectivity services for AI”. However you take these analyst reports – and there is a Gartner disclaimer that its grading is not an “endorsement” – Lumen is clearly doing something right.

In telecoms – actually, just in tech, lest we forget (again) – the future starts with physical plumbing. And fiber is not the forgotten utility beneath the AI economy; it is one of its most important foundations. So forget your 5G access network for now; the AI-era discipline, currently, is to connect global data centers over land and under sea, and run AI inputs-and-outputs back and forth into homes and offices – to indoor gateways, mostly connected on fiber lines. AT&T is making good gains, already, from selling these fiber access lines – and also selling a good deal of 5G on the side, to carry nominal AI workloads into the street and the mall, and back again. The question is: what happens when the fiber is built? Or more precisely: what happens on top of these access systems, and backwards of them, even as they are being multiplied and diversified? Lumen is good at this, says Gartner, and getting better, says Lumen itself.

The Lumen argument goes that the real value of connectivity in the AI era will not come simply from owning fiber assets, but from turning those assets into programmable, automated, enterprise-grade network services. It is the same discussion with mobile networks, except that the deadlines are looser. The two positions are not contradictory; they just describe different stages of the same evolution. AT&T’s second-quarter performance shows the helpful economics of a fiber-first access strategy. It added a record 646,000 combined fiber and fixed wireless connections – and more than a million, including 4G/5G subs. Its profit is growing faster than its revenue – and even its revenue, up 2.5 percent in the period, is decent versus most trad-telcos with straight mobile plays. So fiber improves the economics of the access network, first: lower churn, higher value, opportunities for bundling, a platform for services.

It also gives AT&T a foundation from which to compete in wireless, rather than treating mobile as the whole story, and a foothold in the AI economy when it finally goes to work in homes, offices, and metros. As above, Stankey calls fiber “the best connectivity technology available”. Which is plain, and also quite-the-statement for a 5G aristocrat like AT&T. But Lumen would argue there is work to do on top, with SD-WAN, and dark/lit fiber on longhaul routes, beyond the metro edge and indoor access – which is actually something Verizon has articulated very well. There’s less (we’ve heard) of this from AT&T. But it has similar assets: long-haul routes, wavelength services, private internet, Ethernet and wholesale. It has been expanding 400G across US metros, ready for the AI rush. But yeah, it seems quieter – versus the likes of Verizon, Colt, Zayo, Lumen especially. So have its European counterparts, as it goes.

Maybe Stankey et al discussed this on the Q2 results call this morning; RCR will listen back to the recording. It should be talking about on the next one, and certainly soon. Because AI workloads are changing the requirements placed on networks – of all sorts – as they move between AI clouds, factories, and edge locations. So the network must provide dynamic and predictable latency, automation, and security – on-demand, like in the cloud. And more than capacity and latency, enterprises want to know about these other things, and choice and control over them. So AT&T’s results tell us a lot about the telco economics of AI-era fiber infrastructure, but less about how a telco like AT&T will monetize AI traffic once the infrastructure is in place.