It wouldn't, though. Instead, it would collapse under the weight of IP's management capabilities and worst-in-class feature creep. Despite decades of effort by people like Stuart Cheshire, getting two computers to talk over ethernet is still orders or magnitude more difficult than getting them to talk over a thumb drive. Why would I want that for my mouse, keyboard, and display?
Don't get me wrong, I've read Cheshire's zeroconf book and I'm a big fan of zeroconf IP and mDNS, but technological solutions to these problems are a dime a dozen. Political coordination is the difficult problem and from that perspective mDNS hasn't succeeded (Windows). The network guys can't even coordinate well enough to get name resolution working and I should trust them with doing application-layer coordination competitive with USB and PCIe? Please.
Then, re-evaluate if the increased complexity and energy consumption is worth it, e.g. just for a mouse or keyboard.
Then go ahead and specify how you want this spec to have comparable compatibility modes to USB-C, e.g. if you are going to run displays and external hard drives, as well as HIDs and power...
Considering the alternatives, USB-C is not too shabby for the intended uses.
We already had the technology to do that decades ago or more. If I had the right power and influence, I'd take the Ethernet connector and magnetics, make it smaller, add default POE support then release the smallest possible IP stack so that it can be implemented on small microcontrollers too. Very small devices would talk IP only, bigger ones can add upper protocols such as TCP, UDP etc. All that while gradually phasing out USB. This way we would have open protocols on open media with some peculiar advantages over USB, openness aside: near realtime, stackable (add a layer and you go to the Internet), intrinsic security (USB devices are getting more complex every day, yet there's no such thing as a firewall against malicious ones; Internet protocols support filtering by default), galvanic signal isolation (excellent for instrumentation or audio stuff) and the cable can safely extend a lot more than USB ones. Where low power consumption is vital, magnetic transformers could be swapped with optocouplers.
Except for cost or being "too open to be profitable" (royalties) I don't see a single reason why a beefed up Ethernet shouldn't replace USB entirely.