From the newsletter (sign-up if you want it sooner): AI infrastructure is opening new opportunities for telcos, from data center connectivity to sovereign AI and defense networks. But the payoff will be gradual, shaped by geopolitics, long investment cycles, and the shift toward AI-driven critical infrastructure.
Three things from today about this AI infrastructure opportunity. One: it is very real for telcos, suddenly, but it also represents (another) long roll of the dice, loaded in favour of the AI top-feeders. Two: it is very serious for Europe, and all the also-ran regions looking to ride on US and Chinese coattails; the EU has just stumped up €10 billion of public funds as a stimulus to secure €20 billion of private money to build seven new ‘gigafactories’, even as the AI bubble is about to burst. Three: given the geopolitical climate, the military will drive the agenda, including around 5G-bound networking and sensing, which will mostly run on private networks.
How do we stitch all of this together? First, Matt (Kendall, introducing…) has taken a good look at all the things RCR has discussed recently – Verizon’s DCI dark-fiber deal with Google, AT&T’s longer-running network redesign, SK Telecom’s AI data center business – and, via a call with Omdia, concluded that the big telco pay-day for these varied AI adventures will wait a while longer. Throw in Christian’s profile of SK Telecom’s new frontier efforts, today, plus everything else over the past six months, and all the telco-jostling on the AI scene is clear. But Matt/Omdia say even headline 400-wavelength deals are likely structured as pay-as-you-go exercises to start – six wavelengths at a time.
A billion dollars, per the Verizon/Google deal, is the ceiling, probably. And as this industry knows very well, more data traffic rarely means more money – based just on the economics that costs go down as volumes go up. Which is why, despite the hyper-activity, telcos remain rather cautious. These as incremental investments – pre-buying optical equipment, expanding capacity in stages, avoiding speculative overbuilds. They are trying to capture AI demand without overcommitting capital. Which is the telco way. Dan Schulman might talk about meaningful revenue in 2027, but fiber infrastructure is designed to generate returns over decades. Not over quarters.
Second, we have this €30bn public/private AI money-matching scheme from the EU to build seven ‘gigafactories’ as part of its “major push” on tech sovereignty – and even its stated ambition to be “the AI Continent”, as a multi-state power to rival the US and China. It is as much a geopolitical scheme as an economic one. Europe wants to own and control AI, within its boundaries. The language is all here: European sovereignty, values, infrastructure. It is very like the domestic UK/French version of the same: the tender is for the whole stack, including the networks, and is geared to stimulate a whole local ecosystem of start-ups, SMEs, corporates, plus academic/research groups and governments.
It validates the urgent demand for telcos to be involved – high-speed connectivity, distributed facilities, cross-border deployments (as written by the EU) require data center interconnects, long-haul dark and lit fiber, public metro networks and industrial private networks, and everything between. The EU wants to build seven enormous facilities, and they all need to be connected, cloud to edge. Which also doesn’t mean telcos will book billions in revenue next year – the pipes have to be laid, the fibers have to be lit, and the edge nodes have to connect. And it is like the phase-three AI rollout, after the GPU spree on frontier models, and the geopolitical angst about owned infrastructure.
Thirdly, the geopolitics: RCR spoke with Ericsson Federal this week about this 6G-native concept of ISAC sensing networks, being fast-tracked as a commercial bolt-on for late-era 5G from next year because of cheap drone warfare in Europe and the Middle East. The tech effectively triangulates Doppler-style readouts between cell towers to detect and track drones across radio spectrum. Ericsson has a sophisticated pre-commercial multi-static version – “almost military-grade”, for 2027; “potentially military-grade” by 2028/29. Anti-drone sensing is the break-out case.
ISAC will be attached on private 5G for military usage first – at bases and outposts, typically on foreign soil – and then on private 5G for air/sea ports, energy facilities, and other critical infrastructure – as being bombed already by cheap drones, or else mindful of future threats. It will likely be deployed at stadiums for the 2028 Olympics in Los Angeles – again on private 5G. And then, soon enough, there is a (clearer telco) opportunity to sell sensing as-a-service for traffic management for the drone economy – to manage parcel and pizza deliveries in the US. Which will require a software upgrade on public 5G networks, and more than just Ericsson and AT&T.
And then there is the Industry 4.0 campus opportunity in factories and warehouses, which, judging by past cases, could be quite quick (IoT, AR/VR, AGVs/AMRs etc) or very slow (URLLC and TSN). We will write more about this next week.
But telecoms is a broad church, getting broader; and for telcos, which once owned it outright, the developing story makes very clear how quickly and differently it is growing with AI, and how much flex they require to capitalize.
