The only thing weird about it is that even if it was already implemented in the first commercial computer made with vacuum tubes, because its implementation in hardware is cheap and it can be much faster than a software implementation, it was omitted in most later computers, with the exception of the most expensive computers, which were typically used mainly by government agencies, until AMD decided to implement it in AMD Barcelona (2007).

Once AMD had it, Intel followed soon in Nehalem (2009), and now most CPUs have it.

The article is somewhat misleading about population count being available in ARM NEON since 2005 (i.e. since Cortex-A8).

The 64-bit Armv8-A ISA, which was introduced in 2012, has the CNT instruction, which is indeed a population count instruction.

The 32-bit Armv7-A ISA has the VCNT instruction, which is a variant of population count, but which is much less useful than the 64-bit POPCNT provided by Cray 1, AMD Barcelona, Intel Nehalem or Armv8-A, because it works only on 8-bit words, so executing the equivalent of a 64-bit POPCNT requires a sequence of several instructions.