...
Your source is wrong slightly off as it ignores the birthday paradox. Due to it, on average you have found a collision after checking half of the keyspace with almost certainty. Thus you only need 2^159 key generations. Not that it changes the numbers in any significant way.
Why half? According to the birthday paradox, you'd have near certainty (99.9%) of finding 2 people with matching birthday with as little as 70 people. So wouldn't you need roughly one fifth (366/70) of the key space?
Also, doesn't the "2^160 generations" relate to finding any collision (defined as randomly generating 2 identical priv keys), so including zero-balance ones (also those previously generated by attacker)? If so, finding collision with specific (non-zero) addresses would be a lot harder.
And is the birthday paradox even applicable for targeting specific addresses? I thought it's only about finding any matching pair.