CDS is a relational database. This means that you can establish relationships between the tables, using 1:N, N:1, or N:N relationships.
When you create a lookup field on Entity B to entity A, you have created a relationship between the two. This single relationship would implies that many B records can lookup to (be related to) a single A records.
If you were to add a third entity, you would be able to relate it to either of the others, but nothing happens automatically. Meaning, adding a lookup from C to B, doesn't directly relate A to C.
That being said, the relationship Behavior is important to consider. By default, if you create your relationship as a lookup field, it will be relational. There are two other types of behaviors: parental and configurable cascading.
This means that you can control the behavior of the relationship between the two entities. A common example is parental, where you want B records to be deleted when parent A is deleted. You can also invert that and restrict deletion - meaning you couldn't delete A if a B exists. This prevents oprhaning of records - i.e. incomplete data.
The N:N is always referential. It just allows you to easily say that any A can be related to any B. That's it though, you can't define how they are related or understand when the relationship was made, or restrict deletion. This is when people resort to a manual N:N (intersect) entity approach. You'd create an entity like AB, and have a lookup (N:1) to each A and B. Then it's an entity that you can add fields to, set behaviors, etc. It does require that you create the records, though, vs. simply relating one to another.
TL;DR version. CDS does require that you create all the relationships that you need. You can control behaviors between relationships, but it won't create any relationships for you. This may seem like a burden, but with some basic planning (create an Entity Relationship Diagram (ERD) in Visio or equivalent), it can quick and easy to implement the data model that meets your requirements.