Python's memory-profiler reveals np.random taking up too much memory

Question

I am using Python's memory-profiler to better understand how much memory my code is using and when setting up this simple example, I found a much higher memory usage than expected:

import numpy as np
from memory_profiler import profile

@profile
def func(n):
    x1 = np.random.random(n).astype(np.float32)
    x2 = np.random.random(n).astype(np.float32)
    x_sum = x1 + x2
    return x_sum


@profile
def main():
    """
    Main script calling func
    """
    x = func(1024**2)


if __name__ == "__main__":
    main()

Output:

Filename: mem-test-5.py

Line #    Mem usage    Increment  Occurences   Line Contents
============================================================
     4     52.7 MiB     52.7 MiB           1   @profile
     5                                         def func(n):
     6     56.8 MiB      4.1 MiB           1       x1 = np.random.random(n).astype(np.float32)
     7     68.7 MiB     11.9 MiB           1       x2 = np.random.random(n).astype(np.float32)
     8     68.7 MiB      0.0 MiB           1       x_sum = x1 + x2
     9     68.7 MiB      0.0 MiB           1       return x_sum


Filename: mem-test-5.py

Line #    Mem usage    Increment  Occurences   Line Contents
============================================================
    12     52.7 MiB     52.7 MiB           1   @profile
    13                                         def main():
    14                                             """
    15                                             Main script calling func
    16                                             """
    17     64.8 MiB     12.1 MiB           1       x = func(1024**2)

x1 has the size expected (1024 x 1024 x 4 = 4194304 bytes = 4 MiB) but why does creating x2 right after take up 12 MiB when it should also be only 4 MiB? Interestingly, this problem does not occur if using np.ones instead of np.random or if not casting x1 and x2 to float32 (that is, leaving them at default float64):

Filename: mem-test-5.py

Line #    Mem usage    Increment  Occurences   Line Contents
============================================================
     4     52.5 MiB     52.5 MiB           1   @profile
     5                                         def func(n):
     6     60.4 MiB      7.9 MiB           1       x1 = np.random.random(n)
     7     68.6 MiB      8.1 MiB           1       x2 = np.random.random(n)
     8     76.6 MiB      8.0 MiB           1       x_sum = x1 + x2
     9     76.6 MiB      0.0 MiB           1       return x_sum


Filename: mem-test-5.py

Line #    Mem usage    Increment  Occurences   Line Contents
============================================================
    12     52.5 MiB     52.5 MiB           1   @profile
    13                                         def main():
    14                                             """
    15                                             Main script calling func
    16                                             """
    17     60.6 MiB      8.1 MiB           1       x = func(1024**2)

I don't know if this is an issue with np.random, np.float32, the memory profiler or something else, but it's weird! Any insight much appreciated. Also, can anyone can reproduce this behavior?