Meanwhile, we do not know whether this will be a revamped Hopper GH100 compute GPU or something brand new.
Although, the maximum potential of the HBM3 DRAM has yet to be realized but I think NVIDIA may entirely skip it in favor of HBM3E for the Hopper-Next GPUs which are expected to be codenamed "Blackwell", IMO. Blackwell will use a completely new architecture.
BTW, if suppose HBM3 supplies become an issue, then NVIDIA can also choose to go with Samsung which is also preparing its own "Snowbolt" HBM3P memory that would offer up to 5 TB/s bandwidth per stack.
Samsung's roadmap also shows HBM3P with PIM (Programming In Memory) by 2025 and HBM4 by 2026.