You linked to an article about Google renting an insignificant amount of additional capacity.
Google runs AI / HPC workloads on their own hardware and has been doing that for more than a decade. Google Gemini was trained on TPUs developed in house. It does not run on Nvidia hardware.
