Moore's law still holds, and is expected to hold true until at least 2025 (see wikipedia). I don't think it will be done by then, but that is just guess work.
Moore's Law (the 2-year version upwards-revised from the original 18-month version) stopped holding last year when 10nm (Cannonlake) was delayed to this year and Intel introduced a third step in its tick-tock process. The quotes you're looking at about 2025 were from 2012 (the one cited in 2015 has no support for the quote) and all three should be removed from the article or the assertion altered.
People predicted the death of Moores law in 2005, we all know what happened. Intels CEO seems to think it still holds true, and will continue to do so for the forseeable future. (http://fortune.com/2017/01/05/intel-ces-2017-moore-law/) This is probably partly marketing, but I'm sure there is some truth to it as well.
The linked article admits that the two-year doubling ended, which is what Moore's Law has been for most of its history. Moore's Law ending doesn't mean we won't ever have another die shrink, it means that the notion that we just have to wait two years to get twice the transistors for the same cost (or die area, depending on who you ask) is no longer true (and therefore, projections based on the notion of such a cadence should be considered even more silly than they already were). I don't understand why people continue to claim that Moore's Law's death "has been predicted many times" or whatever when it already ended; what happened in 2005 was that raw clock cycles stopped improving, and guess what: they still haven't improved that much for twelve years.
Moore's law is not about single thread performance, and attacks like this are easily parallelized anyway. Not to mention that fixing them at 3Ghz is just trying to coherce the conslusion to be that we have seen little gain in the last 5 years.