The worlds themselves don't need to be generated on-device, just rendered there (even as a video stream.) The AIs that will build virtual worlds will live in datacenters and stream to headsets. Local AI will likely focus on predicting movement, upscaling, adding frames, and reducing latency from the source in a similar way that RTX cards do now. I would be very surprised if we don't see online AI-generated virtual worlds before the 2030's.
Ok, so why 'real-time generated' at all? Now we're just doing VR MMORPG with an expensive implementation detail that doesn't make an end-user difference?
Real-time generated because it will respond to the players to feel more alive. Worlds right now are mostly static. Imagine Star Citizen but you don't have to wait for decades for it to be developed because the galaxy creates itself as you explore it. No more invisible walls.
