Alternative B is a quite likely concept. All modules just gets a bit smarter. An external ADC may get an internal processor. Suddenly it will be able to not just capture a curve. A two-channel ADC may be able to compute RMS directly, requiring a low-bandwidth link (such as Dallas/Maxim one-wire) to send in the readings. With a number of extra channels, it'll handle 3-phase loads.

Batteries gets intelligent supervision chips that keeps track of the current charge state, and controls the charge cycles to avoid over-charging.

At a larger scale, the stupid supervision cameras doesn't just produce analog video, but gets web servers, motion detection, etc.

Cheap processing power, small sizes and low power consumption means we can put the processing power close to where the need is.

Few people on the other hand can design - or take advantage of - the other alternative, where multiple processors are clustered for redundancy or intricate load-sharing.

One of the most unstable web servers I know about is the load-balancing, redundant solution used by Keil. I don't know if it is affected by moon phase or whatever it is, but they would probably have one tenth of the downtime if they had settled for one single machine. Their site is small enough that a big enough machine isn't too expensive. And this has been true for quite a number of years.

It really is evilishly hard to gain any availability improvement from redundant designs, since there are good design rules for hardware while the software that controls the system will most probably contain a significant number of quite dire errors. There is much work to do until sw engineering will come close to most other engineering disciplines in regard to measurability, regulations, design rules etc.