This is the same kind of failure you used to see in 10base5 and 10base2 networks, where one computer having a loose terminator could cause the entire LAN to be offline. With 10baseT (and I assume automotive ethernet) the bus is contained in the hub/switch/router instead of needing to be terminated (as in CAN bus, from my brief research anyway)
Definitely nothing stopping it, according to the wiki star architectures are supported, but I don't think many of the parts suppliers build them that way, would be my guess.
I think the question would be why would you want to? At the point you're doing that, ethernet already exists including plenty of industrial specced hardware.
Presumably so you can reuse all of your existing CAN car parts. Even the lights are on CAN now, it's going to be a lot of stuff to transition to Ethernet.
And you have thieves stealing cars by hacking into the easily-accessible car light CAN bus - a CAN network switch could also act as a firewall to fix those thefts without re-engineering every single part.
It makes sense for Tesla since they're more vertically integrated but other car manufacturers seem more "parts bin" and having to migrate your whole car over to a new architecture will limit the parts you can use massively until the whole industry switches.
A few reasons: Ethernet electrical is more expensive and ethernet drivers are bigger code size then CAN, and more complex (buggier, you have a lot more buffers and way more state to manage with ethernet versus CAN). Plus CANFD can accommodate different frequencies, so cheaper hardware can participate on the same bus. This may change over time as larger cores are used (Renesas comes to mind) and bigger ethernet stacks that are already scrutinized can be shoehorned in.
