So I guess, 1. This chip for an 8B model even if it was done at 5nm would still be twice the size of a conventional CPU die so what are the yields for this goi

So I guess, 1. This chip for an 8B model even if it was done at 5nm would still be twice the size of a conventional CPU die so what are the yields for this going to be like for even a 30B model? 2. They say 2 months but llama 3.1 was released 2024, ~2 years which is normal lead time for silicon, I suspect this would take longer if the architecture is not llama? 3. Can google do the same thing in house with their Gemma 4 series (two year lead time puts Gemma 4 on silicon April 2028) ? I can see the benefit for hyper scalers but at the rate of model turn over does this type of investment make sense?