AI infrastructure pages separate GPU training reservations, inference unit cost and idle accelerators from a generic compute invoice.
ML and platform leads who already run GPU or managed inference and need a unit-cost model finance can review.
Chapter 3 of Generative AI on AWS explains tokenizers, embeddings, the transformer and scaling laws. This part measures the Qwen tokenizer across five languages, sizes text-embedding-v4 vectors for a search index, and shows where each internal step shows up on a Model Studio bill, with scripts you can run to check the numbers yourself.
Chapter 2 of Generative AI on AWS teaches prompt structure, in-context learning and sampling parameters. This part runs every technique against qwen3.8-flash on Model Studio, adds the two things the book could not cover in 2023, built-in thinking mode and JSON output, and shows what each choice does to the token bill.
The planning worksheet is the AI and GPU workload cost estimator. The practice name is What is AI FinOps?. Every row is a burning GPU-hour meter. LaunchTime older than the last completed job, with no serving traffic, is idle. Instance families differ by region. p5 in us-east-1 does not mean ap-southeast-1 has the same quota or price.
A chapter-by-chapter series that applies Generative AI on AWS to Alibaba Cloud Model Studio, PAI and Qwen, with runnable demos and token cost for each step.
Generative AI on AWS describes a six-stage project life cycle. On Alibaba Cloud the same stages run on Model Studio, PAI and the monitoring stack. This part maps each stage, prices four Qwen models for one realistic workload, and walks through a first API call from Singapore with cost controls in place.
Scoped review for this topic: AI FinOps consulting.