Their statement is not “in the general case” though, it’s specifically for postgres.

Postgres uses a fixed-size null bitmap and variable-size rows, so a NULL value takes one bit in the bitmap (and additional nullable columns may require a wider bitmap), but they are skipped in the row itself.

Postgres uses 1 bit per NULL value, though the reality is more complex as everything is padded for alignment. So fewer than 8 nullable columns are free, and above that you pay probably 8 bytes at once (I'm not totally sure on the exact numbers here).

So for Postgres it is generally true that storing NULLs is very cheap in most cases.