5G is truly about the Internet of Intelligence, where you combine your sensors with AI. Because the processing is neither machine learning, deep learning AI, nor simple data analytics, the latency and other factors necessitates that the cloud is moved to the edge.
For instance, a normal-sized plant would have needed about 5000 sensors for feeding data over 4G 10 years ago. Today, if you go to the same plant, they are thinking about how they can feed and leverage the video streams if they can cover the entire place with new cameras. Practitioner thoughts are therefore more directed towards volumetric rendering and questions about overall security and reliability.
What we need to remember here is that we cannot go all the way to the cloud since the latency is no longer 50 milliseconds. It must be less than 20 milliseconds, maybe even less than 10 milliseconds. What this means is that an all AI job is therefore already ‘always-running’.
To illustrate, think of a data center in an automotive plant. If you have three thousand cameras in place, you may need to operate about 40-50 AI servers. Then you will need 200 servers to cover the entire facility, comprising 10% of infrastructure as 5G, and 90% of infrastructure for AI, augmented reality, and edge computing.
