well - the thing is that difficulty can go down as well.
Therefore it is self balancing, sure if 80% of mining power disappeared all of sudden, it would take a LONG time before reaching difficulty recalculation, but it would eventually happen.
It is self balancing over:
Mining costs (electricty + hardware + human time) and Bitcoin value
Costs go significantly upwards -> Bitcoin value is bound to follow -> BTC value doesn't follow, miners are stopped or moved over alternative cryptocurrencies
Bitcoin value drops -> Miners are being stopped or moved over to alternatives -> Mining costs go downwards
Bitcoin value increases -> More miners will be purchased -> Mining costs go up -> BTC Value eventually follows
Very simple really.
Even hosting business has diminishing returns, yet, that industry is larger than ever, growing at ever growing pace.
Same for networking business (Those who lay the fiber), returns are diminishing, yet they continue on investing
Infact ... Almost every single business follows the same pattern of diminishing returns, for some industries, it's the inflation which does it, and for most technological innovation
When the diminishing returns and growth reaches an equilibrium -> There is barely growth, and the returns don't diminish significantly anymore, it's called a mature industry. ie. farming is a mature industry
When the returns diminish faster than growth it's a industry dying, the MAFIAA would like you to be believe entertainment industry is like so, but they are in a huge growth period right now.
Gone industries in the past are things like steam engine builders (barely any steam engines are being produced today), automotive phones (everyone has a cell phone, and having a phone just for the car doesn't make any sense, it was a fad to begin with), and what else has gone by the way of technological innovation.
If the diminishing returns is stopped, innovation is stopped, progress is stopped.
I think eventually bitcoin mining will reach a stage where it's roughly 2 years for hardware break even including operational costs, and the additional years is your profit. That's when ASIC manufacturing has reached the same processing node and complexity as CPUs today and been there for a significant time. and we are there already, with newest chips being 28nm, just like the latest AMD CPU series - intel is at 22nm now.
It is possible that large processing node, like 65nm ASICs will stay marginally profitable even then for some time - when compared to electricity and human resources costs.
As a comparison point i'm using dedicated hosting industry, some 5 years old servers are still being sold today, and many are in production still todate.
Hell, the bulk of our cheapest option is with Q1'10 launched CPU, and we are still taking high end servers into use older than that (Intel L5520, which is 45nm) :)
Tho, in our use every single drive is new and minimum 2Tb size, not decade old models :) Lifespan for those is 3years and then phase out, 60%+ is expected to survive the 3 years, and target average lifespan of disks is 3years, so there will be some disks being used for more than 4 years, but a marginal group.