Some more reassuring info:
https://litecointalk.org/index.php?topic=17948.0 March 26th, 2014:
With the internet awash with poor data and panic about scrypt asics, I decided to put together a (hopefully) readable analysis about the current state of asics, with emphasis on the differences between SHA256/Bitcion asics, and SCRYPT asics.
One of the most important things to consider is that scrypt asic's will not be as prolific as sha256 asics simply due to the memory constraints imposed by the scrypt algorithm.
Some technical facts to consider:
First true btc asics were built on a 65nm manufacturing process which at the time was about 7 years old. Despite that they were ~40x more efficient than GPUs (~2MH/w on a GPU, vs ~80MH/w on a BFL single sc)
Current 20nm (the smallest/newest production process) sha256 asics (knc Neptune for example) are orders of magnitude better than that, around 2000MH/w, ~1000x more efficient than a GPU
Until manufacturing process goes smaller than 20nm, asic's can't advance as quickly as they have in the last 2 years while "catching up" to current process... ie: they jumped from 65nm to 20nm in less than 2 years, whereas now they can only expect to jump from 20nm to 14nm in ~2 years. This is significant.
We can see that the first gridseed scrypt asic is about 12x more power efficient than the best GPU in terms of scrypt KH/w. AMD R9 270 at ~ 3KH/w, gridseed 360kh scrypt asic ~36KH/w (55nm process).
We can see that the first KNC scrypt asic is about 26x more power efficient than the best GPU in terms of scrypt KH/w. AMD R9 270 at ~ 3KH/w, KNC titan ~85 KH/w (?? process).
This is already proof of scrypt algorithm doing what it was originally intended to do, which is to make scrypt asic development less prolific than that of sha256 asics - which I would say they have achieved, notably when considering that even the first generation 65nm btc asics were 4x more efficient than the current 55nm scrypt asics vs GPUs strictly in terms of power. This is especially true when you consider the people building these scrypt asics (gridseed, knc) have already been building sha256/bitcoin asics for quite some time and are familiar with the process.
As with everything, the first person to get their asic, scrypt or otherwise, usually makes their money back pretty quickly. With bitcoin, this was far more volatile than it will be with scrypt asics as the bitcoin asics that were built on a smaller process than their predecessor were 10x more efficient.
If you're basing your scrypt asic buying decision on the history of bitcoin asic's, don't. This is because scrypt asics will not be as prolific as sha256 asics were... it is simply not possible to produce a scrypt asic that outperforms gpu's by the same amount that a sha256 asic outperforms them, among other things because of memory constraints inherent to scrypt algorithm (and limitations of sram cell size/current leakage on newer processes).
Scrypt asics on 22nm will not be 25x more efficient that scrypt asics on 55nm as we see that they are in sha256/bitcoin asics. This is due to, as previously mentioned, limitations of sram cell size and current leakage vs manufacturing process size and the cost of that miniaturization.
Also consider the fact that scrypt coins are largely popular, ie: adopted, because of their rapport with GPU mining. You had hundreds of thousands of gpus mining bitcoins that suddenly needed something else to do and spurred on the adoption of litecoin (and later other scrypt coins). Take that away, suddenly vertcoin is the new favorite. Altcoins not being the forerunner that bitcoin is - I would be surprised if they responded the same positive way to the introduction of ASICs as bitcoin did. Look at how scrypt difficulty has risen since Christmas and how alt coin prices have inversely adjusted, almost 1:-1.
The asic's shipping in the fall/early winter (namely, KNC titan) will be close to 5x more efficient in terms of initial cost vs GPU, and about 26x more efficient in terms of power usage when compared to current GPU standard. Although KNC has not disclosed the manufacturing process upon which their new scrypt asic is being manufactured, one can estimate that if it is 2.5x more efficient than gridseed's 55nm process, it is likely being produced on 28nm. If you are one to buy this ASIC, you can expect that the first 20nm scrypt asics (~Q1/Q2 2015) to roll out will be, at most, 1.4x more power efficient than your 28nm due to the way current leakage of sram scales on process miniaturization.
SO, what does all that mean?
For bitcoin/sha256 asics:
Current/Next gen bitcoin/sha256 asics are already at the forefront of manufacturing process, they cannot get more power efficient without waiting for a new manufacturing process to be available (14nm) which is not on the roadmap for their scale of production until at least end of 2015 - if Intel isn't doing it now, they wont be doing it for another year. This means that the only way to increase your SHA256 hashrate using asics is to scale-out. ie: buy more. As a result and when compared to previous improvements in ASIC generations and their resulting efficiency jumps - these current 20nm asics will not depreciate as quickly as, for example, the 55nm asics. Increasing asic efficiency 10x every year is no longer possible now that we've reached (or caught up to, rather) the smallest reasonable manufacturing process and thus, the most efficient sha256 asics (for MH/w) without improving the actual design of the circuit (minimal gains are to be had here, but not 10x). The only way for them to go now is down in price (with increase in quantity produced).
For scrypt asics:
Largely due to the familiarity of manufacturers with die shrink on exsiting sha256 asics, the second gen scrypt asics (again, referring to KNCs example) are already very close to the smalles process and thus peak efficiency. With the most efficient asic capable of being produced today only being ~40% more efficient than a 28nm example, which means the "best" scrypt asic that we'll see before mid-end 2015 will only ever be ~35x more power efficient than GPUs, still a significant margin no doubt, but not as extreme as the best 20nm sha256 asics out now which are nearly 1000x more efficient than GPU.
TL;DR: The best next gen scrypt asics (arriving Q3 this year) are roughly 22x less efficient than gpu mining than current gen bitcoin asics are, but are still substantially (~26x) more power efficient than GPUs. Once they reach 20nm process , likely Q2 2015, they will be ~35x more power efficient.
TL;DR2: Next gen scrypt asics shipping in the fall are pretty close to the best you can expect for asic efficiency within the next 18 months (within ~40%) and thus, you will not get washed out by future asics as hard as second gen bitcoin asic buyers were. BUT, asics might be far more detrimental to scrypt coin profitability per KH than bitcoin asics were to sha256 profitability.
