For a while, back in 2012 (when Bitcoin SHA256 ASIC preorders were going on), quite a few of us thought scrypt ASICs wouldn't be worth producing for a long time.
In theory, an ASIC could be designed for mining any individual currency, and a well-designed one would beat out GPUs for efficiency, but as Vann said above, it's generally more difficult (and thus cost-prohibitive) to create ASICs for those with more memory-intensive algorithms.
The idea with the more resistant cryptocurrencies like Ethereum and Zcash is not necessarily to have an entirely "ASIC-proof" algorithm behind them, but one resistant enough that nobody finds ASICs worth developing, and that's how it will *probably* remain for some time.
Considering all the people with GPUs, if the current coins were magically flooded with application-specific devices that crushed GPUs in terms of both efficiency and effectiveness (highly unlikely), I could see a new algorithm being used, or an existing coin with a more obscure mining algorithm becoming more popular. But if we were in a world where that happened anytime soon, there's no telling what would happen.
If you check out the pages on Zcash, they're careful to say that Equihash ASICs, while currently "not economically implementable", aren't impossible to make, so they're still unsure about what will happen in the distant future.
https://z.cash/blog/why-equihash.htmlhttps://z.cash/support/faq.html