does this mean you can horizontally scale to GPT-4-esque LLM locally in the near future? (i hear you need 1TB of VRAM) Is Apple's large VRAM offering like 196g

does this mean you can horizontally scale to GPT-4-esque LLM locally in the near future? (i hear you need 1TB of VRAM) Is Apple's large VRAM offering like 196gb offer the fastest bandwidth and if so how will pairing a bunch of 4090s like in the comments work?