Everyone (in the US) is talking about 4GHz, it seems, as part of a next-gen auction pipeline for choice mid-band spectrum. This follows the publication of a couple of lobbying reports, advocate that the band is either cleared for full-power commercial usage or else shared with incumbent agencies for quicker leverage, plus a new FCC filing from Verizon for a one-year experimental license for a 4.8-4.9 GHz chunk of the band to extend its ISAC experiments with Ericsson and Samsung within a 5km-radius of its Los Angeles lab, plus also a CTIA address from FCC chair Brendan Carr to suggest $100bn-plus might be accrued from upcoming spectrum auctions.

Separately, and to bookmark for you, Sean has just published a glut of research notes about changing demands on 5G/6G networks as AI goes on fiber from the center (in data centers) to the edge (to sundry agent-hosting network outposts), and out to the far-edge (onto devices and machines, sometimes via wireless networks) – and as the cellular uplink starts to get maxed-out, and the physical world is suddenly animated (the dream of IoT, recast as AI). These posts – available on RCR Tech, serialised in the RCR Wireless newsletter over the next days – offer some context: that AI will change what the radio (RAN) network needs to do, and also increase demand for spectrum.

The gist is that AI, especially physical AI, will make heavier demands on 5G/6G uplink channels – versus familiar streaming-style downlink consumption in the 4G/5G era. Cameras, microphones, machines, glasses, vehicles, and robots are all just sensors – their value, initially, is in what they perceive. How they subsequently act depends on how quickly they can upload their findings to an AI brain, in the cloud or at the edge, to process and return. The unknown, of course, is how much brain-work will be done on the sensor itself, and how much, or how often, the rest will go on a cellular network – versus fiber or Wi-Fi, or even LoRaWAN. But spectrum access and optimization matters.

Not that refarming the 4GHz band – or the 4400-4940 MHz range of it that the CTIA wants the NTIA to scrutinize, sharpish (within 60 days) – is the only answer. At least 160 MHz of the upper C-band (3.98-4.2 GHz) will go under the hammer next year. The FCC has earmarked three more auctions by the end of 2028, plus multiple 6G sales. Spectrum optimization might also be improved via configuration changes, uplink enhancements, carrier aggregation, radio densification, and better antennas/beamforming – managed, as possible, in AI-RAN set-ups. There’s another reality check in Sean’s notes: John Saw says T-Mobile hasn’t seen much AI impact on its network usage yet.

Current impacts are concentrated on transport networks and data centers, he says – as RCR has written here consistently, if just to keep the industry honest. Which makes the CTIA case, about the US needing hundreds of MHz more, a future-planning argument, rather than a warning shot about dire straits – even if traffic grew 20% last year. But there’s money in it (as part of the broader refarming process), as Carr told the CTIA show. And the sharing argument, presented by Valo Analytica, is (always) a good one: don’t spend long years and many billions clearing federal users when modern sharing tech could put some of it to use much sooner. Equally, the short-term application for 100 MHz at 4.4 GHz by Verizon (Cellco Partnership) for drones and robots shows real and immediate interest.