I understood it to be a throwaway estimate for the cost of apple building out a specialized chip architecture that can handle excessive workloads from transform

I understood it to be a throwaway estimate for the cost of apple building out a specialized chip architecture that can handle excessive workloads from transformer based AI apps.