Deep sea to low orbit – telcos can’t build fast enough (Telco Diary)

From the newsletter (September 16; sign-up or follow if you want it sooner): A new traffic cycle is driving a major physical network build. The AI infrastructure story is becoming a telco infrastructure story: more fiber, more spectrum, more radios, more satellites, more compute, more capacity. The technology is changing, but the build cycle looks remarkably familiar. The question is who owns the infrastructure – and captures the value. Strip away all the tech talk and puffery – ‘AI-native’, ‘6G-ready’, ‘direct-to-device’, ‘multi-orbit’; whatever – and you have a fairly traditional yarn about the construction of physical networks to carry spiralling traffic loads. Maybe the pace and volume is different, and the load more speculative, but the story is the same. Start with AI-RAN. Nokia has turned up with another batch of operators – mostly in the Middle East, but extending and expanding a ready roster that also covers Europe, Asia Pacific, and North America – to test its Nvidia-backed platform. The names are new; the headline is not – that it thinks you can eke another 20% out of your spectrum holdings already, rising to 50 percent with software upgrades by the time the testing is done, at launch in 2027 – and to 100 percent by 2028. As written before. But this is Nokia’s big pitch to its old telco heartlands: AI-and-RAN, for RAN and on RAN, to make 5G/6G better, and to hoover up some edge workloads on the side. Sign up, pay up, and change your fortunes. Meanwhile, Samsung has demonstrated a version of the same AI-RAN shtick with Verizon – as reported here on Monday (September 14), via a catch-all Verizon notice about its new club of 6G innovators. But the Samsung release focuses just on its ISAC tests at a Dallas footie game, possibly a World Cup match, which consolidated network functions and accelerated compute on a single server on an edge-AI platform. So while Nokia’s biggest promise is about AI-on-RAN, even while it talks up its early AI-for-RAN wins, Samsung is testing something else. The Samsung/Verizon project presents a new 6G-native RAN-sensing workload. What is that? RAN on AI-for-RAN? Late-era 5G-based ISAC tests are everywhere, right now. But the terminology starts to eat itself when it is combined, and the news cycle falls like this. They both make use of the same infrastructure overhaul, but RAN efficiencies are an AI application for telcos (except when they’re not, such as NTT Docomo et al presenting a telco-engineered repeater strategy for mmWave networks), and ISAC is a telecoms innovation for enterprises. Take the fiber, as well. EXA Infrastructure is building a 6,552km cable between the US and the UK: 24 fibre pairs, 500Tbps of capacity, a 2029 launch (absolutely rapid, btw), and a line that customers are now buying fiber pairs, and not just wavelengths. Again, the industry can’t build networks fast enough. Now take the satellite market. Space42 and Viasat are putting up to $1bn into a shared satellite and ground infrastructure platform called Equatys, geared for D2D and advanced MSS. Their “tower company” model is interesting: multiple operators share the space infrastructure while retaining their own spectrum, customers, and commercial relationships. We talked yesterday, and over recent weeks, about this idea of a team sport, where telcos must learn to share the spoils, along with the work. Well, SpaceX, per yesterday’s write-up, doesn’t particularly want to play that way: its integrated vertical-infrastructure proposition – from model to data center to rocket to satellite – is about hyper-scaling the hyperscale model in the AI economy. But others won’t work like that; team-play is the way. Meanwhile, Elveo Mobile, created from the August 2026 merger of Lynk Global and Omnispace, is taking a different route, partnering with Apex to establish a spacecraft factory for its D2D constellation. Different model, same underlying problem: if satellite comms is to move from SOS messages to mainstream mobile, it is going to take a lot of satellites and a lot of rocket ships. SES’s completion of its 13-satellite O3b mPOWER MEO constellation fits the bill, too. It is less like a new strategic bet than these other announcements, but it represents another piece of high-capacity infrastructure being brought into service. But not all of these investments will pay off. This phase of the AI story may, for now, be defined less by clever apps and more by foundational infrastructure ownership. One last thing: fiber or FWA to build the right network? The answer depends on where you’re building, what infrastructure you have, and what the economics look like. Join Mimosa Networks on Wednesday 23 for a practical look at how operators including Prairie Hills Wireless, Viaero Wireless, and Reliance Jio are using different broadband technologies and spectrum strategies to solve their particular network challenges.