At the end of the day it's a team of psychologists, computer scientists, and super computers against one human. It's not a fair fight. There are plenty of dark patterns to make you not internalize the tracking. So while you're technically right, no one expects to see this happen in practice because your model isn't accounting for this.
In some cases, it does. In medical trials, you need "informed consent", and not just regular consent. It doesn't matter what papers are signed, if a patient isn't informed what the trial is testing, what known side effects there are, and what alternatives there are, it doesn't count.
More relevant to this issue, GDPR also uses the concept of informed consent. Under GDPR, tracking is legal only if consent is informed and freely given. The example printer fails both these conditions. It is not informed consent, because the tracking is not prominently disclosed to the user, and merely mentioning it in the fine print is insufficiently prominent. It is not freely-given consent, because a service being conditional on acceptance of tracking means that there is coercion to accept the tracking.
My viewpoint on the ethics tends to follow somewhat close to the GDPR's requirements. Even if somebody clicked through a 50-page EULA, that does not give informed permission to track somebody, and so it is still spying.
Not everything is measured in dollars, and getting back the dollars spent as a result of deceitful advertising does not undo the damage caused by the deceit.