Nah, price isn't going to be a motivating factor. If AMD came up with a card that had 3x the VRAM of the latest NVIDIA offering there would be research groups who would be interested because loads of models are hardware bottlenecked.
The very nature of GPUs implies that they are optimized for parallelizable workloads.
If I can split my training across 3x the chips at 1/10th the cost for each chip, why wouldn’t I do that?