The question is summarized in the title.

I will elaborate it here:

Are you aware of a research paper that quantifies the Trade-off in terms of

Cloud Computing processing power typically expressed in Cloud Million Instruction per Second (MIPS) or Floating-point Operation per Second (FLOPS)

vs

Cloud Computing Cost when using virtual Central Processing Units (vCPUs), Graphics Processing Unit (GPUs) and Tensor Processing Units (TPUs) such as vCPUs alone or (vCPUs+GPU) or (vCPUs+GPU+TPU).

These configurations are typically offered by well-known Cloud Service Providers (CSP) such as Awazon Web Services (AWS), Microsoft Azure or Google Cloud Platform (GCP)?