FinOps for ML and GenAI on Google Cloud: Cost per Task, Not Cost per Token

Are you still tracking raw token consumption, or do you know the true business cost of every completed task?
Gartner projects that 40% of agentic AI projects will be canceled by 2027 because teams struggle to define their return on investment. While monitoring token spend is straightforward, attributing that spend to specific features, teams, or business outcomes—and proving ROI—is where most cloud initiatives fail.
Join GDG Cloud Thessaloniki for an evening dedicated to unlocking financial predictability in AI: FinOps for ML and GenAI on Google Cloud: Cost per Task, Not Cost per Token. We will break down a battle-tested FinOps methodology designed for Google Cloud, shifting engineering focus from low-level token accounting to high-impact unit economics.

### What We’ll Cover

• The ROI Bottleneck: Why agentic AI projects stall out financially and how to pivot metrics from abstract token billing to actionable "Cost per Completed Task."
• Cost Drivers Across the Lifecycle:
• Classic ML Footprint: Managing training clusters, serving endpoints, pipeline orchestrators, and persistent storage overhead.
• GenAI Pricing Models: Choosing strategically between Standard, Flex, Batch, Priority, and Provisioned Throughput.
• Architectural Optimization Tactics:
• Smart Caching & Routing: Implementing implicit, explicit, and semantic caching alongside dynamic model routing and strict output constraints.
• Agent Loop Safeguards: Setting hard boundaries and circuit breakers against runaway agent execution loops.
• Granular Attribution & Kernel-Level Visibility:
• Labeling strategies across jobs and API calls—and identifying blind spots like shared LLM gateways and Provisioned Throughput traffic.
• Leveraging eBPF for deep, kernel-level traffic observability without code changes, SDK injection, or middleware overhead.
• Budgets, Anomalies & Engineering Habits:
• Deploying ML-powered cost anomaly detection and budget triggers on Google Cloud.
• Establishing actionable FinOps team rituals that keep cost reviews sustainable rather than reactive.
• The "Five Things for Monday" Action Plan: Concrete, immediate configuration changes and audits you can run on your cloud environments right away.

### Who Should Attend

• Software Engineers & AI Developers building with Gemini and Vertex AI who need to keep operational costs aligned with system output.
• Cloud Architects & DevOps / Platform Engineers responsible for infrastructure efficiency, observability, and billing transparency on Google Cloud.
• Engineering Managers & Tech Leads needing to justify GenAI ROI, allocate cloud budgets accurately, and establish mature FinOps practices.

### Agenda

• 18:30 – 19:00 | Welcome, Check-in & Community Catch-up
• 19:00 – 19:15 | GDG Cloud Thessaloniki Intro & Community Updates
• 19:15 – 20:15 | FinOps for ML & GenAI on Google Cloud: Cost per Task, Not Cost per Token
• 20:15 – 21:00 | Interactive Networking, Drinks & Pizza

Prerequisites: Bring your curiosity and laptop (optional if you want to inspect your own GCP billing setups or take notes). No deep FinOps background required—just an interest in architecting cost-effective, scalable AI workloads!

Thanks to the Municipality of Thessaloniki and OK!Thess for hosting another great GDG event

PLEASE REGISTER HERE