Running a machine learning project can be an expensive endeavor in terms of computing and storage resources; whether it is the computation required for training a large dataset or the ongoing computation cost of running a model in production for real-time inference, costs can add up significantly over time. Some End-to-End ML platforms and tools are starting to adopt cost management practices like show back and chargeback so that TV costs of the model, both for training and production, can ultimately be factored into the return on investment of the overall project. Advanced ML teams work hard to be transparent about the underlying infrastructure costs that will be required so that they can be factored in by the model development team. That way they can ensure that the value of the system they are building will outweigh the costs of running it.
Cost management and transparency emerged as a key function and the world of auto-scaling in cloud computing and it will likely become a key function in ML Platforms as well.