Where does the network edge? Who will control the fringes? The FCC has given SpaceX the nod to deploy 15,000 new Starlink satellites, and also waived standard SCS leasing requirements, so it can use its AWS-3 and AWS-H spectrum from EchoStar for SCS without having to go via a terrestrial operator. So formally, SpaceX can now occupy both sides of the satellite-terrestrial divide. Of course, we might recall Vish’s mathematical takedown yesterday, which reduced the satellite question to a single (populated) square kilometer, and suggested viability hinges on capacity density. Which does not make Starlink a side-show, clearly. The question is how much an increasingly capable Starlink service encroaches on the terrestrial edge, and how much AT&T et al cede in terms of customer experience and ownership, and when Musk comes knocking for land deals of one sort or another. The lines are blurry.

Scale matters. In telecoms, some have it, and some are well placed to buy it. Vodafone has just bumped-up its annual savings target for its VodafoneThree merger from £700m to £1bn by the end of 2032. Scale is a defensive merger benefit, but also a means to fund a better network, and remove duplication etc. Vodafone’s £11bn investment plan to make its UK footprint into an all 5G-SA affair was part of the proposition to buy CK Hutchinson out of its stake. It is the standard response to a difficult mobile market: consolidate the old estate, monetize the new scale. AT&T is doing similar, in reverse – external capital to expand physical infrastructure, with an eye on extra-curricular AI endeavours, as opposed to corporate scale to rationalize it. Its fiber joint venture with GIP and CPP Investments puts Gigapower and Forged Fiber 37 into a wholesale vehicle to accelerate US fiber expansion.

As an aside, you might note that RCR’s strategy around this has shifted. The non-wireless stuff is moving to RCR Tech; as a last hurrah, RCR Wireless has a couple of subsea adventures for you, courtesy of EXA and AWS. EXA has been warning about the strain on international traffic, and has promptly commissioned a new transatlantic route – its ninth, and the first of its kind in a decade. More diverse routes are required for resiliency and redundancy, as well as new and upgraded routes. AWS is building its Sta’O’Nuk transpacific cable for these ends. EXA says its traffic quadrupled in 2025, versus the previous three years; it jumped another 20% in 2026. Seven of 21 Atlantic routes are being retired. Cable development takes five to seven years. There is a crunch, and fiber systems are feeling it first. When AT&T talks about fiber, it is not just a broadband story. Long-haul fiber and landing stations – plus wireless sites for mobility, for physical AI – determine where AI can run, and how quickly and resiliently.

And while AT&T watches Starlink at its back, and soldiers on with its digging and ducting, it has its eyes fixed on the front, and how to climb the higher-value layers on top. Its collaboration with OpenAI mixes AI agents and domain knowledge in an actual business process – called LegalEdge, for lawyers. Its three-layer network argument – connectivity, orchestration, tokens – tells you where it thinks the value goes: above the pipe, into software. DT, as we recall, is making a similar calculation, albeit more inward looking: €1.1bn in AI and automation savings in 2027, rising to €2.5bn by 2030. DT chief Timothy Höttges talks about inference data centers inside telco infrastructure, too. AI is not just for the hyperscaler cloud; some will sit inside the network itself. But that gets into the theoretical and wishful; right now, it is just an infrastructure build cycle for everyone – including Starlink.

The carriers should remember that, too.