What is FinOps?
FinOps is a portmanteau of 'Finance' and 'DevOps.' It's a cloud financial management practice that brings together technical, finance and business teams around a shared goal: maximizing the business value of every dollar spent on cloud, rather than simply cutting costs.
The term is now tied to a reference framework formalized and maintained by the FinOps Foundation, an organization under the Linux Foundation, which documents the phases, principles and best practices of the discipline in a standardized way across organizations.
FinOps isn't synonymous with 'cut costs at all costs.' The goal is to align cloud spend with the business value it produces — which can also mean spending more on a high-impact project, as long as the decision is made with full visibility.
The 3-phase cycle: Inform, Optimize, Operate
The FinOps Foundation framework structures the practice around a continuous cycle of three phases, which don't run once but repeat continuously as cloud usage evolves.
Inform: gaining reliable visibility
The Inform phase builds visibility into cloud costs and usage, through cost allocation, benchmarking, budgeting and forecasting. It's the foundation of the FinOps cycle: without reliable visibility, effective optimization and governance are impossible.
Optimize: identifying efficiency levers
The Optimize phase builds on the visibility gained to identify concrete opportunities to reduce waste and improve efficiency — underused resources, unexploited pricing commitments, over-provisioned architectures.
Operate: embedding changes for the long run
The Operate phase implements the changes identified, and defines the KPIs and governance policies that durably align technical execution with business objectives — without this phase, one-off optimizations don't hold up over time.
The 6 principles that structure FinOps practice
If the 3 phases describe what a FinOps team does, the 6 principles explain why and how they should do it — they're what distinguishes genuine FinOps practice from a one-off budget review.
- Teams need to collaborate (finance, engineering, business) rather than work in silos.
- Business value drives technology decisions, not just the sticker price.
- Everyone takes ownership of their own cloud usage, not just a central dedicated team.
- FinOps data should be accessible, timely and accurate for all stakeholders.
- FinOps should be enabled centrally, to ensure consistent practices across teams.
- The cloud's variable cost model should be actively exploited, not just endured as a monthly bill.
The main cost optimization levers
Beyond the general framework, three concrete levers account for most of the savings achieved by mature FinOps programs.
Reserved instances and savings plans
Committing to a usage duration (1 to 3 years) in exchange for a discounted rate typically delivers savings in the 30-50% range compared to on-demand pricing, for workloads with predictable volume over time.
Spot instances for interruption-tolerant workloads
Spot instances tap into providers' unused capacity at a heavily discounted rate — typically 70-90% cheaper than on-demand — in exchange for a risk of interruption with little notice. They're particularly well suited to batch processing, model training, or any workload tolerant of interruption.
Rightsizing, often the fastest lever
Adjusting resource size to actual usage (rather than a generous initial estimate) remains one of the fastest levers to activate: automatically detecting and shutting down idle resources, or downsizing over-provisioned instances, often delivers measurable results within the first few weeks.
Combining levers rather than choosing between them
These three levers aren't mutually exclusive: a mature setup typically applies pricing commitments to predictable baseline capacity, spot instances to interruption-tolerant workloads, and continuous rightsizing across the board — each lever matching a different workload profile within the same organization.
According to the FinOps Foundation, a mature FinOps program reduces cloud spend by 20-25% in the first year. Some analyses of deployments combining multiple levers (rightsizing, pricing commitments, idle detection) report 50-60% cost reductions within the first 30 days for the targeted line items.
FinOps Foundation, State of FinOps 2026Precise cost attribution, harder than the reduction itself
Reducing a cloud bill is only useful if you know precisely which team, project or product is driving which share of the spend. This is often the most underestimated bottleneck in a FinOps practice.
Tagging, the foundation of reliable attribution
Without a consistent tagging strategy applied from the moment resources are created (team, project, environment, criticality, owner), it becomes impossible to answer the simplest question: who spent what. A tagging policy applied retroactively, on already-deployed infrastructure, always costs more to implement than one applied from the start.
Chargeback and showback: making cost visible to teams
Showback displays each team's share of the cloud bill, informationally, without an actual financial transaction. Chargeback goes further by actually billing each team for its usage. Both approaches make technical teams accountable for their consumption — a prerequisite for the FinOps principle of individual ownership.
The case of shared resources
Some resources (shared networking, common platform tooling, bundled licenses) don't naturally belong to a single team. Splitting their cost proportionally and transparently — rather than leaving them in an 'overhead' bucket that's impossible to optimize — is one of the trickiest but also most structuring aspects of a mature cost attribution practice.
AI and GPU costs, the new FinOps frontier
The rise of AI workloads introduced a new category of cloud spend, with cost dynamics very different from classic infrastructure — to the point of becoming, in 2026, the top priority for FinOps teams at organizations deploying AI at scale.
Cost dynamics specific to AI
Usage-based model pricing (per token), limited billing visibility on certain managed services, the high hardware cost of GPUs, and the split between training and inference spend create attribution and optimization challenges that classic FinOps didn't directly address.
AI-specific optimization levers
Beyond classic levers (reserved instances, spot), several AI-specific practices deliver substantial savings: idle detection and automatic shutdown of unused GPU instances, choosing the model best suited to the need rather than systematically the most powerful one, caching frequent responses, and weighing self-hosting against third-party API usage based on actual volume.
According to the FinOps Foundation's State of FinOps 2026 report, GPU spend has become the top FinOps concern for AI-focused organizations, surpassing classic cloud costs for the first time. Some deployment analyses combining smart routing, caching and rightsizing report a cost-per-answer reduction of more than 80% once implemented.
FinOps Foundation, State of FinOps 2026The most frequent cloud cost management pitfalls
Three mistakes consistently show up in organizations struggling to control their cloud bill.
Trying to optimize before having visibility
Trying to cut costs without going through a serious Inform phase leads to optimizing the wrong spend categories, or underestimating a change's real impact. Visibility always precedes optimization in a well-run FinOps cycle.
Treating FinOps as a finance-only topic
A FinOps effort driven solely by finance, without involving the technical teams who make day-to-day architecture decisions, rarely delivers lasting savings — the collaboration principle isn't optional, it's foundational.
Treating FinOps as a one-off project rather than an ongoing practice
A single cost review, followed by no regular follow-up, produces temporary gains that erode within a few months as infrastructure evolves. The Inform-Optimize-Operate cycle only has value when repeated continuously, not executed once.
The most visible line item on a bill isn't always the easiest to optimize, nor the most valuable to prioritize. Analysis based on Inform-phase data helps identify the levers with the highest real impact, not just the most visible ones.
A domain present across all 3 major entry-level certifications
Costs, pricing and commitment models consistently appear in entry-level cloud certifications — Azure Fundamentals, Google Cloud Digital Leader and AWS Cloud Practitioner all test understanding of cost levers and financial management tools specific to each platform.
Further reading
- ArticleIaaS, PaaS, SaaS: the 3 cloud service models
- ArticlePublic, Private, Hybrid, Multicloud
- ArticleCloud Regions, Availability Zones, Resilience
- ArticleIAM, Security and Cloud Governance
- ArticleAWS vs Azure vs Google Cloud
- ArticleCloud Sovereignty, Portability and Reversibility
- GuideUnderstand Azure Fundamentals
- GuideUnderstand AWS
Frequently asked questions
What is FinOps, in one sentence?
FinOps is a cloud financial management practice that brings together technical, finance and business teams to maximize the business value of every cloud dollar spent, structured by a framework formalized by the FinOps Foundation.
Does FinOps just mean cutting cloud costs?
No. The goal of FinOps is to align spend with the business value it produces, not to systematically minimize costs. This can mean increasing spend on a strategic project, as long as the decision is made with clear visibility into its impact.
What's the difference between the Inform and Optimize phases?
The Inform phase builds visibility into current costs and usage. The Optimize phase builds on that visibility to identify concrete actions to reduce waste. One can't work effectively without the other.
What's the fastest optimization lever to implement?
Rightsizing (adjusting resource size to actual usage) and idle resource detection generally deliver measurable results within a few weeks, even before negotiating longer-term pricing commitments.
Are spot instances suited to all workloads?
No. Spot instances suit interruption-tolerant workloads (batch processing, model training, test environments) but not workloads requiring continuous availability with no possible interruption, such as certain critical production services.
Why are AI costs different from classic cloud costs?
AI costs combine often less predictable usage-based pricing (per token for language models), high hardware costs for GPUs, and a split between training and inference costs, which complicates precise attribution and budget forecasting compared to classic cloud infrastructure.
What's the difference between showback and chargeback?
Showback displays each team's share of the cloud bill, informationally, without an actual financial transaction. Chargeback goes further by actually billing each team for its usage, which drives more accountability but requires more mature internal billing infrastructure.
Is FinOps only for large enterprises?
No. The principles apply to organizations of any size, even if the scale of tooling varies. A small organization can apply a simplified version of the Inform-Optimize-Operate cycle without necessarily having a dedicated FinOps team.
How much savings can you realistically expect from a FinOps program?
FinOps Foundation data points to an average 20-25% reduction in cloud spend in the first year for a mature program, with results varying based on the organization's starting maturity and the levers actually implemented.
Do you need to know FinOps for a cloud certification?
Cost, pricing and commitment model concepts are part of the domains tested in the Azure Fundamentals, Google Cloud Digital Leader and AWS Cloud Practitioner entry-level certifications, even if the term 'FinOps' itself isn't always explicitly named.