Not only is it impractical in rural areas of the US like others have said, but in most small-to-medium sized cities as well. Some places simply won't hire folks if they don't have access to a car. Some small cities have no taxi service, and the larger (30-80k) that do often charge too much to use on a regular basis.
Not to mention the same holds true for rural areas of other countries as well, as the bus system might not come often enough to work, let alone buy groceries. Or you might have to deal with weather which makes bike riding (let alone walking) inconceivable.
Where I live, the closest store is 5 miles, but it's just a very basic and small general store. To reach an actual store with fresh produce and more, it's a 100 mile round trip. That may be on the somewhat extreme end of things, but variations on that are a normal way of life for people in rural areas.
Cellphones may be popular, but they are not ubiquitous. Land lines (and equivalent) work fine for many purposes (including me).
Since humans are terrible at assessing risks that have probabilistic or delayed effects, most people - including some engineers that should know better - don't understand or underestimate how powerful patter-of-life analysis can be when it can utilize entire lifetimes of timestamped location data.
The point is you shouldn't assume everyone has a cellphone in their pocket. Therefor a car that broadcasts its location would compromise an existing expectation of privacy.
The rapidly growing use of ALPR devices is causing a similar compromise of privacy, but that's a separate problem.
My point is that it's not compromising an existing expectation of privacy if 1) you have the option to buy an older car, and 2) you are aware the new car you are considering buying is connected to the Internet.
Now your car is broadcasting its location at all times via cell signal, and that data may be (almost certainly is) being recorded.