[pachico]

It would be helpful to know which services you use. Do you use ML services or instances with GPU? Where is most of the cost?