A quick overview of news from today – echoing the hub-and-spoke discussion on Tuesday. There are new numbers about the minutiae of data center construction, which, separate of all the headline builds (covered everywhere), show the ripple effect on supplier industries: there is a bonanza in the busways market (to grow at 15% CAGR from $3.7bn in 2026 to $8.6bn by 2032), for example; the sale of ‘physical infrastructure’ for data centers, the general nuts-and-bolts of them, grew 28% to $12 billion in the first quarter, says Dell’Oro (as reported here Tuesday). Which just confirms how the supplier ecosystem is buzzing on the AI bubble, right now – as everyone knows.

But we also have reports of longhaul fiber works by London-headquartered RETN and euNetworks, for example, building the backbone spoke-channels for the AI revolution (or whatever it may be). RETN says 50% of its IP/MPLS traffic now goes over 400G coherent pluggable (ZR/ZR+) infrastructure, following a mass 400G upgrade across its pan-Eurasian backbone last year. Meanwhile, euNetworks has a new route, running 1057km via the Alps to offer the most direct connectivity between Paris and Milan, it says – shorter than the standard path via Lyon and Marseille along the coast. Which is about capacity expansion and route diversity – for the workloads, and for their security.

This is what AWS is talking about, in conversation with RCR Tech (should be RCR Wireless) – just on a grander scale. Stephen Callaghan, senior technologist for its network-core, says AI workloads create synchronized flows between clusters, which require dedicated network ‘fabrics’, expanded DCI capacity, and tighter integration of compete, power, and connectivity. AWS is scaling metro, long-haul, and subsea infrastructure, deploying 800G optics while still investing in new fiber routes – because tech upgrades cannot keep pace with AI demand. The challenge is not one thing; it is about coordinating all these elements to prevent the network from being the bottleneck.

Meanwhile, there’s stuff about the access network, skirting the enterprise edge. More cable operators are deploying low-latency DOCSIS on Aurora’s platforms, it seems – for faster ‘fiber-like’ broadband speeds, without breaking the bank. Lord knows, the carrier industry is desperate for an easier go of it. AI-ready? Nope! No 5G networks anywhere support the sub-10ms target for AR and multimodal AI vision, says Ookla – even if networks are doing okay for AI text and voice latency, with most achieving under 50ms and 40ms respectively. Vodafone, meanwhile, has a new report calling for greater investment, warning connectivity is critical for climate resilience as well as AI cats-and-dogs.

Which brings the focus down to services and applications, and what gets used, what makes a difference, and what makes money – amid all the bombast and chicanery about future AI pyrotechnics and space comms. Vodafone lists the ways, as justification for its pitch about (kinder regulation and sharper cooperation) for bigger investments: telecoms for disaster warning and emergency response, for example. Which are worthy applications, clearly, albeit a little worthy-sounding. What about the stuff that makes money? IoT? Hmmm. Well, the global AI-in-IoT market (at least) is to grow from $4bn in 2025 to $6.45 billion by 2035, apparently. And Telefónica is taking it seriously.

Telefónica Tech is launching an eSIM solution based on the SGP.32 standard with Thales, for a single SIM with profiles from multiple operators. It integrates with Kite, Telefónica’s IoT platform; the firm said it is strengthening its position as an IoT ‘orchestrator’. As well, Verizon has a deal with BMW to connect new BMWs in North America, using KDDI’s connectivity platform. Which is an example of IoT moving beyond sensors and industrial assets into software-defined vehicles for streaming, navigation, data services – and back again to where it all started: M2M. And then there is the edge itself, this amorphous critical connectivity zone which might be connected in various ways.

So Nokia is pitching ‘deployable’ 5G for defense comms in contested environments, and has a deal with NestAI, in which it co-invested €100m in late 2025, to develop AI-enabled capabilities for military and public safety cases where comms can’t be assumed. It’s another confusing Nokia release, actually – ‘deployable’ essentially means those 5G-in-a-box solutions, to go in backpacks and helicopters, to be deployed in 10 minutes for crisis comms. Which was a mainstay of its doomed ECE unit, a shrunk-down core with small-cell radios; there’s no mention of ‘private wireless’ in the press note, although that is also what Nokia has tended to pitch for defense comms. I mean, what else – right?

Anyway, it is confusing (and RCR should follow up, time permitting). The idea is about critical comms and compute for AI-style decision-making at the edge, when networks are disrupted or denied. Which is the same story, again: AI is only as useful as the infrastructure underneath it – which sometimes needs to survive where the network is the target, rather than just the enabler. On a cheerier note, on a manicured edge, Cisco is taking the same AI-ready network reliability argument to The R&A, to connect The Open, the AIG Women’s Open, and its Scottish golf headquarters. Different battlefield resilience; different wireless tech, too; same pitch about the higher-grade networks – for secure and performant comms and AI apps for facilities, matchplay, broadcast, fans.

Per the header: the plumbing is the product.