Moore's law was very different because the underlying process was being scaled. CPUs just process information, and the inherent limit on the amount of "stuff" needed to perform say an addition has an incredibly small physical limit [1] which existing chips don't approach. So through scaling the feature size of chips, CPU manufacturers were able to take the same amount of silicon, the same size wafer, and get "double" the computational power (roughly speaking) out of it. So with about the same amount of "assembly line" work you can churn out something that is twice as fast in 18 months. After a decade its 100 times as fast, but your "assembly line" looks about the same.
That's not "efficiencies of scale" - that's fundamentally making something much better due improved physics, with the same amount of work. It particular, it would apply even to "small" producers.
Now don't get me wrong, CPU manufacture was also subject to traditional efficiencies of scale: the biggest fabs got bigger, and a few large players squeezed out the rest and were able to sell more and average our their R&D costs over more sales, but that effect is small compared to the million-fold improvement in the underlying physical design.
Batteries have no such scaling. The power stored is basically related to number of lithium ions and the capacity of the anode to accommodate them all. There are some small efficiencies: you can increase efficiency from 80% to 90%, but never to 2000%. You can make materials thinner or cheaper. You can change form factors to use materials more efficiently. You can standardize on battery sizes to make more use of a single production line. You can secure long term lithium contracts and open more mines.
These are all the traditional "efficiency of scale" things and they all hit a wall pretty quickly. You could probably easily sustain a 15% reduction for a while longer, but certainly not 40 years like Moore's law.
> The same principally applies to almost anything.
It doesn't, just look around.
What else has decreased in cost by a factor of a million over the past few decades? Cars have increased in popularity by many-fold since the 40s , and after an initial period of traditional "efficiencies of scale", costs have remained relatively fixed.
Look at any random food product that suddenly increases a surge in popularity, perhaps reaching a 10x sales multiplier: final cost and production costs don't drop 10x.
If all of a sudden we start eating 2 avocados every meal they aren't going to start costing 10 cents.
> In fact, for typical physical products the saying is that double the production will halve the cost.
I can imagine this rule is true... up to a point!
That's the "traditional efficiencies of scale" at work: it's usually an S-curve [2]. If you want some custom widget, you are probably going to have to pay $100 for the first one, and $1 each or whatever for your run of 100. When you order one million, maybe it drops to 1 cent. When you order a billion, they don't cost 0.001 cents though.
You don't have to take my word for it though: just look at any two big companies, but where one is bigger than the other, and look at their costs of production. Let's say Coke sells 5x as much as Pepsi: does Pepsi cost 5x as much to produce? Not all, they are virtually identical.
Many models of vehicles sell 10x or 100x of some unpopular rivals, but the production cost is about the same.
---
[1] https://en.wikipedia.org/wiki/Landauer%27s_principle
[2] Approximately, at least - although it depends heavily on the product. For example, the initial part of the curve might not be very flat for some things (e.g., with large fixed one-off costs) - but they almost all share the "rightmost" flat part of the S-curve.