There is no doubt that the increased performance and reduced cost, size and energy consumption application and communication processors has opened up new vistas for machines talking to machines and the creation of intelligent, autonomous and semi-autonomous objects.
In addition we are entering a long-foreseen era of application-specific processors with an increased diversity of architectures including neuromorphic processors.
However, right now any economies of scale present tend to be for architectures that are based on human-oriented parameters and processing and latency. That tends to mean 32-bit resolution data, image sensors optimized for human viewing, and so on.
But it should be considered that communicating machines with very fast processing and short latencies created for different functions may require many other parameters quickly but not necessarily with such high dynamic range. For example V2X automotive networking or drone collision avoidance – can have very different requirements to humans, such as low-resolution data and frame rates and spectral sensitivities beyond what the human eye can see.
In addition, the IoT is not one market but a fragmented kaleidoscope of opportunities. Each sub-market has to be addressed on its own terms and the economic case made before the benefits of overlapping applications can be seen.
There's no doubt that the move towards a swarm-like infrastructural intelligence is a major economic and societal development. It encompasses the chilling idea of the "technological singularity" where computer intelligence acquires the ability to recursively self-improve and over-take that of humans.
But it also seems clear that we are at the very early start of the S-shaped adoption curve that may take decades to play out.
Related links and articles: