I’m really looking forward to GPU-CXL integration.

"CXL-GPU: Pushing GPU Memory Boundaries with the Integration of CXL Technologies" https://arxiv.org/abs/2506.15601