This article explains how to design AI cost allocation for production workloads across AWS, Azure, and Google Cloud. It covers why traditional cloud tagging models are insufficient for AI workloads and introduces an architecture for request-level attribution using metadata instrumentation, gateway layers, and project boundaries. The article also analyzes the full cost surface of AI systems, including model inference, vector databases, embeddings, GPU compute, data transfer, a

FOCUS is becoming the standard way to normalize cloud and SaaS billing data, but it is often misunderstood. This article explains what FOCUS actually standardizes, where it stops short, and how experienced FinOps teams use it as a foundation rather than a replacement for optimization and allocation logic.

Traditional FinOps relies on delayed billing data. AI workloads evolve in minutes. This gap is where costs explode. In this post, I explain why agentic FinOps requires a real-time runtime, not better dashboards. Using MCP (Model Context Protocol), FinOps agents can query live cost, usage, and policy data across clouds, detect anomalies as they happen, and act before waste becomes sunk cost. MCP is not hype. It is the missing plumbing for scalable FinOps in the age of AI.

AI Agents & MCP in cloud Financial Management and CloudOps: Beyond the Hype