From AI factory to AI factory-factory 

Firstly, the agenda for Industrial Wireless Forum on November 10 is now confirmed – and it looks fantastic. And if you set the recent moves by Nokia and Ericsson to retreat or retread in the private 5G space, say, against all the hyperactive chatter in the tech market about edge inference for autonomous mobile robots and drones, and whatever else gets bundled as ‘physical AI’, then it feels like this will be the definitive edition – with lots of questions to answer, and lots of conclusions to make. It’s going to be a good show; we hope to see you there. And for one thing, Qualcomm is red-hot right now on IoT – which is an industry sub-sector that might be considered to be both old-hat or all-the-rage. After all, what is IoT – if it is not (also) private 5G and physical AI, and everything in between? IoT is dead; long live IoT. Per Sean’s twin write-ups from the ‘edge’ king’s (Dragonwing) IoT Day in San Diego, Qualcomm is assembling the ecosystem, including with smart acquisitions, to make edge AI work in myriad IoT devices across industrial markets – while tackling the harder question about how to deliver deterministic IoT from probabilistic AI.

Both his research notes are worth your time, and there is a temptation from reading them, somewhat fanciful, to say that, where Nvidia is doing cloud AI, Qualcomm is doing edge AI. It is too simplistic, of course; Nvidia has Jetson for edge deployments, plus AI-RAN etc, and Qualcomm has powerful on-prem AI appliances and increasingly capable robotics processors – plus it is going into data centers to boot. Both are building full-stack platforms, and both want to support physical AI from development through deployment. But there is instructive distinction in their emphasis and heritage: Nvidia wants, most of all, to own the compute platform that makes the centralized AI brain whizzy in AI ‘factories’; as Sean writes, Qualcomm wants to supply the entire edge platform – compute, connectivity, systems – to make its cloud commands register in real time in sundry industrial appendages, including in some-day everyday end-devices, like drones and robots (IoT devices, big and small) in proper factory-factories, and everywhere else they go to work.

The interesting question is whether Qualcomm can turn that sprawling collection of technologies into a repeatable commercial proposition (of the sort Nokia, for one, figured was too complicated and fragmented), and whether the physical-industrial world is prepared – business-critical, life-critical, mission-critical – to absorb as much intelligence as the AI industry would like to sell it. How much of this actually gets deployed? Who makes the money from its design, deployment, operation, ownership? Questions there for Industrial Wireless Forum; catch you November 10.