AI FinOps != Cloud FinOps: Building Observability for AI Spend

Vantage Head of Product Rem Baumann and Demandbase's Gaurang "G" Mavadiya (Cloud Engineering) explain why the traditional cloud FinOps levers, reserved instances, savings plans, and month-end anomaly alerts, fall apart when applied to AI spend that is hard to see, hard to attribute, and hard to control. G walks through how Demandbase's small FinOps team routes every AI request through an LLM proxy to get visibility, control, and attribution, using scoped proxy keys, tags, and header-level metadata to answer who is spending what, and how much it costs to serve AI to customers versus R&D. They cover enforcement patterns for runaway agents and noisy tenants, using FOCUS to normalize proxy logs, and how both teams lean on their own AI agents (Demandbase's "Claudia" and Vantage's agent plus MCP server) to monitor AI cost at a volume no two-person team could handle manually. 00:00 Intro: Why AI Spend Breaks Traditional Cloud FinOps 06:18 The Problem: Hard to See, Hard to Attribute, Hard to Control 12:31 The LLM Proxy: Visibility, Scoped Keys, and Attribution Boundaries 18:43 From Observability to Enforcement, and Tokenomics in Practice 28:44 What's Next, Takeaways, and Where Observability Meets FinOps 35:24 Audience Q&A Watch the Day 1 Keynote and Tokenomics Foundation announcement: https://www.youtube.com/live/W4g8xvduIbs Watch the Day 2 Keynote: https://www.youtube.com/live/vyQTW0PLw7A Learn more about Token Economics: https://tokeneconomics.com https://x.finops.org https://join.finops.org #FinOps #FinOpsX #AISpend #Observability #Tokenomics