These kind of functional solutions are very cute mathematically, but...
I once wrote this way of generating an infinite list of primes in unlambda. That was kind of interesting to do.
That's not really true, though, because that filter is applied lazily, so it only get evaluated up until the Nth prime and no further.
The time complexity is less obvious, each prime adds an additional pass to evaluate, so that seems linear, but each new pass only applies to a list that already had all the previous passes filtered out, so that's not technically linear, but I find it hard to determine how much it actually is.
But the filter for multiples of 3 only sees the values that weren't already filtered by 2. So if we have N filters we don't evaluate all N filters for each value. The 2 is applied to everything, the 3 filter is only applied to things that are not multiples of 2, the 5 filter is only applied to things that aren't multiple of 2 and 3, etc. That doesn't sound very quadratic to me.