First, and quick (because I am about to be on holiday; praise the lord): I’d like to quote the inimitable Matt Hatton, who summarized very-well on LinkedIn the silly-buggers hash Semtech has made of Sierra Wireless’s old IoT business – bought for $1.2bn in 2023 and sold for $62m today. “How to lose a billion,” wrote Hatton, following up with some sensible number-crunching. “The cellular hardware business generated revenue of $262m in the year to Jan 2026, based on the latest annual report saying hardware revenue was up $34m on the previous year (when it was $228m). That growth bucked the trend in recent years. So that’s a 0.24x revenue multiplier. The connectivity business is around $91m, based on the above – which tracks a bit down from previous years. But I don’t think that’s included.”

Besides, I’ve had Lumen on the blower in one ear telling me that north-south AI traffic, including from the industrial heartlands, has nearly-no growth, and that everything will happen in the cloud, and flow east-west between agents; and Celona in the other ear telling me that the hype about physical AI is true, for once, and there is “phenomenal” demand for on-prem private 5G (and Wi-Fi) to stand-up industrial robots (and humanoids down the line) . Which is a little screwy, right? Are armies of robots in factories and warehouses really going to be tethered to the cloud, or will – as one well-placed coal-face integrator told me a little while ago – these blue-collar automatons drive fibre-to-the-factory refresh as well? Maybe I am missing something in between. 

But what is clear: the future is not hyperscale compute at one end and dumb devices at the other. It is a hierarchy: big training workloads in the cloud, in its fluffiest sense; inference across regional and enterprise infrastructure; lightweight domain models next to machines, robots, sensors. That is the architecture being proposed for physical AI in factories, and such – and hasn’t changed very much in forever. So what is the off-ramp from the factory? I am inclined to agree with my integrator friend; that a north-south refresh is coming, even if it doesn’t shift the needle on that top-line 2% growth much. The factory still needs serious fiber backhaul to move data into local edge compute, enterprise clouds, and, where necessary, distant AI factories. Something to ponder on, anyway. Byeee.