you could have 2 to get 32GB VRAM, which is enough to run a decent AI model locally.