I don't think the kernel guarantees that if X runs against Linux today, then it will run against the same X on new kernels on new hardware in perpetuity. It's more along the lines of, "if I one calls the read syscall with parameters a, b, c, it won't suddenly require a 4th parameter, or demand they be in a, c, b order."
Anyways, if the kernel developers decided they wanted to invent a brand new uAPI that does graphics, I wouldn't at all be surprised if new GPU drivers only implemented the new uAPI if their target userspace programs all use the new uAPI instead of an older uAPI. Nothing in userspace is "broken" because it will seem to the old uAPI using userspace as if there is no driver for the new GPU.
Anyways, if the kernel developers decided they wanted to invent a brand new uAPI that does graphics, I wouldn't at all be surprised if new GPU drivers only implemented the new uAPI if their target userspace programs all use the new uAPI instead of an older uAPI. Nothing in userspace is "broken" because it will seem to the old uAPI using userspace as if there is no driver for the new GPU.