AI FinOps: Measuring and Controlling AI and GPU Spend
For FinOps practitioners, platform engineers and engineering managers who need to see where AI money goes, allocate it to teams and products, and cut waste in GPU, model-API and token spend.
The curriculum is being prepared and reviewed. Enrollment is not open yet.
About this course
AI spend behaves differently from the cloud spend most FinOps practices were built for. GPU capacity is scarce and often bought in reserved blocks; model APIs bill per token with prices that change by model, cache state and batch mode; shared clusters make it hard to say which team used what; and the business value of an AI feature is rarely in the bill at all. This course adapts the FinOps Foundation's framework to that reality.
You start with the cost anatomy of AI: accelerated instances, storage and networking, managed AI platforms, and per-token model APIs, including how prompt caching, batch processing and reasoning tokens change the bill. You then build visibility: ingesting cloud and AI-vendor bills into a FOCUS-aligned dataset, tagging and metadata for AI workloads, and allocating shared GPU clusters by GPU-hours with OpenCost or Slurm accounting and shared model gateways by token usage with LiteLLM-style metering. The middle of the course covers unit economics and showback or chargeback models that teams accept. The optimization modules cover GPU utilization and right-sizing, commitments, spot and capacity reservations, and token-level levers such as model routing, caching, prompt compression and batch APIs. The course ends with forecasting, budgets and anomaly detection, and an AI FinOps operating model. Build-versus-buy and on-premises break-even analysis are covered in inf-260. Pricing and product facts are current as of October 2026.
What you will be able to do
Break an AI bill into GPU compute, storage, networking, platform and per-token API components
Normalize multi-cloud and AI-vendor billing data into a FOCUS-aligned dataset
Design a tagging and allocation scheme for shared GPU clusters and shared model gateways
Define and compute unit economics such as cost per inference, per resolved ticket or per training run
Improve GPU utilization and choose between on-demand, spot, commitments and capacity reservations
Reduce token spend with caching, batch APIs, model routing and prompt optimization while monitoring quality
Build forecasts, budgets and anomaly alerts for AI workloads
Set up an AI FinOps operating model with clear owners, reviews and guardrails
Syllabus
6 modules · 21 lessons · about 12 hours
1
Module 1: The anatomy of AI spend
Where the money goes in AI infrastructure and model APIs.4 lessons
Why AI breaks traditional FinOps: scarcity, token pricing and fast-moving rates
GPU infrastructure costs: instances, storage, networking, egress and idle time
Model API pricing: input, output, cached and reasoning tokens, batch discounts and rate tiers
Managed AI platforms and marketplaces: Amazon Bedrock, Vertex AI and Microsoft Foundry billing
2
Module 2: Visibility and data
Getting all AI cost data into one trustworthy view.3 lessons
The FinOps Framework applied to AI: principles, capabilities and maturity
FOCUS in practice: normalizing cloud and SaaS AI bills into one schema
Ingesting usage: cloud billing exports, AI vendor usage APIs and gateway logs
3
Module 3: Allocation and unit economics
Assigning shared AI costs fairly and tying them to value.4 lessons
Tagging and metadata strategy for AI workloads, models and experiments
Allocating token spend through model gateways, API keys and per-team budgets
Unit economics: cost per inference, per task and per customer, and showback versus chargeback
4
Module 4: Optimizing GPU spend
Paying less for the accelerated compute you actually need.3 lessons
Measuring real GPU utilization with DCGM: allocated versus active versus useful
Right-sizing and sharing: instance choice, MIG, time-slicing and autoscaling inference
Commitments, spot and capacity reservations: Savings Plans, CUDs and Capacity Blocks
5
Module 5: Optimizing model and token spend
Cutting the per-request cost of AI features without degrading quality.4 lessons
Model routing and cascades: sending each request to the cheapest model that meets the bar
Prompt caching, batch APIs and output-length control
Self-hosting versus API at the workload level: a per-feature cost model
Guarding quality while cutting cost: evaluation gates for optimizations
6
Module 6: Forecasting, governance and operating model
Running AI FinOps as a continuous practice.3 lessons
Forecasting AI spend: usage drivers, adoption curves and price changes
Budgets, quotas and anomaly detection for GPUs and tokens
The AI FinOps operating model: roles, cadences, policies and a 90-day plan
Exam and certificate
The exam unlocks once you have completed every lesson. It draws 20 questions from a larger bank, spread across the modules, with the answer options shuffled for each attempt. You have 40 minutes and need 80% to pass. After a failed attempt you can try again after 24 hours, up to 3 attempts in total. Feedback is given per module, never as the answer key.
Passing earns a certificate of completion and assessment from the Wirewalk Institute of AI with a unique ID that anyone can check at institute.wirewalk.com/verify. It is not an accredited qualification or a vendor certification. How certification works.
The Wirewalk Institute of AI is independent. This course teaches publicly documented products and is not affiliated with or endorsed by FinOps Foundation (FinOps Framework, FOCUS), AWS (Cost and Usage Reports, Savings Plans, Capacity Blocks), Google Cloud (Billing export, committed use discounts), Microsoft Azure (Cost Management), OpenAI, Anthropic, OpenCost, Kubecost, LiteLLM, or NVIDIA DCGM. Product and company names are trademarks of their respective owners.
A plain-language tour of the stack that runs AI: what a GPU server is, why clusters need special networks and storage, how jobs are scheduled, what an AI data center needs in power and cooling, and how cloud and on-premises options differ. No prior infrastructure background required.
The operator's view of data-center GPUs: what the current NVIDIA and AMD systems are, how they are packaged and connected, how the driver, CUDA/ROCm and container stack fit together, how to monitor and triage failing GPUs, how to share and size GPU memory, and what power, cooling and sourcing decisions follow.
The Linux skills that differ on an AI cluster: choosing and imaging the OS, packaging GPU and RDMA drivers that survive kernel updates, NUMA and PCIe topology tuning, cgroup v2 isolation, shared identity and file systems, scientific software environments with Lmod, Spack and containers, and patching a fleet without breaking jobs.