It is enough to identify whoever is paying for the internet access, which is enough, in itself. And it might be enough to identify the actual user with "reasonable" certainty, e.g. if the user was home alone at the time the IP was used.
Courts found that it doesn't have to be demonstrated that a user can be identified, the abstract reasonable risk that a user could be identified is enough to turn an IP address into PII (and this ruling explicitly mentions this).
In reality, as a service provider, you have no ability to determine if the client IP belongs to an individual or not - so you have no choice but to assume it does identify an individual.
You're not sending your nginx logs to Google, a well known advertiser, do you?
In this case you can store IP addresses if you have a legitimate reason (e.g. you can show you need it for troubleshooting etc), as long as it's reasonable and doesn't infringe on the rights of the user, and you have documented it along with the retention strategy.
Courts found that it doesn't have to be demonstrated that a user can be identified, the abstract reasonable risk that a user could be identified is enough to turn an IP address into PII (and this ruling explicitly mentions this).