The cost of fixing hardware failure in a final product is not the same thing as the cost of developing and mass-producing the product.
For example, we're not talking about one button, we're talking about a lot of buttons, usually custom-made for the car in question. The whole dashboard physically has to be designed around them. Meanwhile Tesla just slaps a screen on a mount in the middle of the car and calls it a day. It's basically "we have to get everything right the first time" vs "fuck it, we can always fix things in a later software update". Which is a way to save costs by cutting corners.
The buttons all have their own complicated logic too, although I suppose that even with physical buttons one can handle almost all of that purely through software these days, so that's not really as much of an issue any more as it used to be (it does make me terrified that cars can be hacked and bricked, but I digress).
Speaking of a lot of buttons, that's the other thing: if all your buttons are virtual, you can have infinite buttons! The only thing we have to do is introduce a ton of mode switches! Which is absolutely terrible when you're driving, but nobody seems to care! So we can cram a ton of features into a screen that would otherwise require a million buttons, and use that in marketing. Even though we'd probably be better off if some time was spent to whittle things down to the essentials and design the interface around those cleanly.
Many buttons also means many more pieces to physically install, and many many more wires. And each one (or small cluster) is often accompanied by even more independently-wired small information displays (small LCDs and LEDs for showing the state / temperature / etc) which are yet more wires.
A touchscreen is largely just a single fused physical unit with ~two cables: a data ribbon and power. Utterly trivial to install and wire up in comparison.
