From niche concern to near-universal in two years
FinOps for AI is the fastest-growing practice in the history of cloud financial management. In 2024, only 31% of FinOps teams managed AI costs. In 2025 it was 63%. In 2026 it is 98%, according to the FinOps Foundation’s State of FinOps 2026 report, which covers organizations responsible for more than $83 billion in annual cloud spend (SpendArk).
The same report shows AI cost management is now the skill FinOps teams most want to add, named by 58% of respondents (CIO Dive). Practitioners are being asked to govern a cost category that barely existed in their scope three years ago.
For CFOs and CIOs, this is more than a change in a team’s job description. It signals that AI spend has become large and volatile enough to need its own discipline, with its own data, controls, and executive ownership.
What is FinOps for AI?
FinOps for AI is the practice of tracking, allocating, forecasting, and optimizing AI spend, and tying it to business value, across models, APIs, GPUs, and AI-powered software. It borrows the cloud FinOps playbook but has to work faster and closer to where the spend is created.
It is worth separating two ideas that often get mixed up. FinOps for AI is about governing what you spend on AI. AI for FinOps is about using AI to run the FinOps practice itself. Walmart’s Karl Kraft drew exactly this distinction at FinOps X 2026 (SiliconANGLE). This article focuses on the first.
| Dimension | Cloud FinOps | FinOps for AI |
|---|---|---|
| Billing unit | Compute hours, storage, data transfer | Tokens, API calls, credits, GPU hours, agent actions |
| Main cost driver | Provisioned capacity | Usage volume, model choice, and agent behavior |
| Predictability | Fairly stable month to month | Volatile, can spike within hours |
| Where spend appears | One or a few cloud bills | Cloud bills, model vendor invoices, SaaS credits, expensed tools |
| Effective control point | Reports, alerts, commitments | Runtime limits in gateways and provisioning, plus reports |
| Value metric | Cost per transaction or customer | Cost per successful task or outcome |
The shift in workload mix explains the urgency. At AI-forward enterprises, GPUs now take 18% of spend, up from 4% in 2023, and inference makes up roughly 80% of AI budgets (SpendArk). Inference is a run cost that grows with every user and every agent, so it behaves less like a project and more like a utility bill.
What capabilities does a FinOps for AI practice need?
A working FinOps for AI practice needs six capabilities, moving from seeing the spend to controlling it to proving what it returns.
- Visibility across every AI bill. Bring cloud GPU costs, model API invoices, SaaS AI credits, and expensed tools into one dataset. The FinOps Open Cost and Usage Specification (FOCUS) helps here, and its upcoming release adds AI-specific detail such as which model was billed and which person, service, or agent drove the charge (FOCUS).
- Allocation to owners. Tag and attribute spend to teams, products, customers, and agents, so showback and later chargeback become possible.
- Unit economics. Track cost per successful task, per customer, or per transaction rather than total monthly spend.
- Forecasting for variable demand. Model AI spend on usage drivers such as active users, agent runs, and tokens per task, and refresh forecasts monthly.
- Runtime guardrails. Enforce budgets, rate limits, and model routing in an AI gateway or platform layer, so controls act before the bill rather than after it.
- Value linkage. Connect each major AI cost line to the business outcome it supports, which is what finance ultimately needs.
Most organizations are still early. The State of FinOps 2026 data shows a clear payoff from automation.
| Maturity phase | Share of organizations | How it works | Typical cost reduction |
|---|---|---|---|
| Crawl | 34% | Alerts only | 5% to 10% |
| Walk | 51% | Semi-automated, manual approval | 10% to 20% |
| Run | 14% | Automated with guardrails | 20% to 30% |
Source: State of FinOps 2026 via SpendArk.
How should CFOs and CIOs set up FinOps for AI?
The most effective model puts FinOps for AI inside the technology organization with a strong line to finance, and extends an existing cloud FinOps team rather than building a new one from scratch.
Operating model. Most FinOps teams already report to the CTO or CIO, 78% in 2026, and the most common structure is a central team that enables engineering and business units, used by 60% of organizations (SpendArk). The CIO should sponsor the practice and own the tooling and data. The CFO should set budget rules, ROI thresholds, and the allocation policy, and receive the same AI cost reporting as the technology leadership.
Connect adjacent disciplines. AI cost data lives in several places. The Vantage team’s review of the year notes that FinOps is increasingly integrating with asset management, service management, and observability tools (Vantage). Bring IT asset management, SaaS management, and IT financial management into the same view so AI credits, licenses, and cloud consumption are governed together.
A first 90-day plan.
- Days 1 to 30, find the spend. Inventory every AI cost source, including model APIs, GPU workloads, SaaS AI add-ons, and expensed tools. Name an owner for each, and publish a first monthly AI cost report.
- Days 31 to 60, allocate and set limits. Apply tagging and allocation rules by team and product. Set budgets and alerts for the largest AI workloads, and route API traffic through a gateway where possible.
- Days 61 to 90, link spend to value. Define a unit cost metric for the top five AI use cases, start showback to business units, and agree with finance on the threshold above which chargeback begins.
Staffing does not have to be large. Organizations with more than $100 million in cloud spend average eight to ten practitioners, while smaller companies can capture most of the value by automating alerts, rightsizing, and guardrails.
Frequently asked questions about FinOps for AI
What is FinOps for AI? FinOps for AI applies financial operations practices to AI spending. It covers tracking, allocating, forecasting, and optimizing the cost of models, APIs, GPUs, and AI software, and linking that spend to business value.
How is FinOps for AI different from cloud FinOps? AI costs are billed in tokens, credits, and actions, change much faster, and appear across many vendors. That means FinOps for AI relies more on runtime controls and unit economics than on monthly reports alone.
What percentage of companies manage AI costs through FinOps? According to the State of FinOps 2026 report, 98% of FinOps teams now manage AI costs, up from 63% in 2025 and 31% in 2024.
Who should own FinOps for AI? Typically a FinOps team inside the CIO or CTO organization, with a strong connection to finance. The CIO owns tools and data, while the CFO owns budgets, allocation rules, and ROI thresholds.
What tools are used for FinOps for AI? Common tools include cloud cost management platforms with AI cost tracking, AI gateways that enforce budgets and routing, SaaS management platforms for AI credits, and data standards such as FOCUS for consistent billing data.
How do you start a FinOps for AI practice? Start by inventorying all AI cost sources and owners, then allocate spend and set budgets for the largest workloads, and finally define unit cost metrics that connect AI spend to outcomes.
The bottom line for CFOs and CIOs
FinOps started as a way to tame cloud bills. In 2026 it is becoming the function that governs all technology spend, with AI as its most urgent and least predictable category.
The organizations that move first will treat FinOps for AI as a shared CFO and CIO capability, not a reporting task. They will see every AI cost in one place, assign it to an owner, control it at runtime, and connect it to outcomes. That is what turns AI from a budget risk into an investment the board can trust.