Consider trillions of IoT devices or large scale data generators.

Combined this with public cloud processing like AWS and you have a major congestion problem,

Factors like backhaul costs, cloud access hubs and data ingress/egress charges growth exponentially.  Also consider the increased latency, data privacy and security impact.

A hybrid edge/cloud model makes more sense.

This video from Sixsq/HPE addresses the use of edge computing in a large scale generator like the ESA. The model utilises orchestrated AI edge components too process ESA data locally and only report the anomalies.

The effect is vastly reduced bandwidth and cloud ingress costs but the same outcome. 

As new analytic engines are developed these can be pushed down with complete management of each computational engine.

AnotherPeak has developed three separate edge models, including supportive devices and analytics.

This covers various industry use cases, and is now exploring the generation of specific AI engines for its client base.