Cloud cost optimization works in a fixed order. Name an owner, clear out idle and oversized resources, then buy reservations or savings plans sized to the cleaned-up baseline. Microsoft’s own guidance puts it bluntly, discounts reduce rates, not waste. A mid-market team without a FinOps department can do this in 90 days with native tools, a tagging policy, a monthly review and one unit-cost metric. Watch AI and GPU spend closely, because that’s where waste is growing.
Cloud cost optimization means cutting waste and lowering rates without hurting performance. Order matters. Assign an owner, remove idle and oversized resources, then buy commitment discounts sized to the clean baseline.
Here’s a pattern our cloud engineering team runs into a lot. A company moves to Azure, the first year’s bill comes in about where the business case said, and then it creeps. Ten percent, then twenty. Nobody did anything wrong exactly. Dev environments stayed on over weekends, a proof of concept never got shut down, and someone bought reservations for VMs that were later resized.
That creep is now measurable industry-wide. Flexera’s 2026 State of the Cloud report found respondents estimate 29% of their cloud spend is wasted, the first increase in five years, with AI workloads a main driver. It’s self-reported. If anything, it undercounts. Nobody reports waste they haven’t found yet.
This playbook is for IT and finance leaders at companies with 100 to 2,000 employees, usually Azure-heavy with Dynamics 365 and Microsoft 365, sometimes with AWS on the side, and no dedicated FinOps team. It’s the same sequence we follow on client cost reviews. Still planning the move itself? Start with our cloud migration strategy guide and come back after go-live.
What is cloud cost optimization, and how is it different from FinOps?
Cloud cost optimization is the set of actions that lower your bill, such as rightsizing, deleting idle resources and buying discounts. FinOps is the operating model that keeps those savings from leaking back, with owners, tagging, reporting and a regular review between engineering and finance.
Think of optimization as a project and FinOps as the habit. Different jobs. Companies that only do the project tend to watch costs drift back within a year or so, because once the consultant leaves and the cleanup spreadsheet goes stale, nobody in particular owns the bill, and new resources start appearing without tags, budgets or anyone asking whether they’re still needed.
The FinOps Foundation has widened the definition, too. Its 2026 framework now covers public cloud, SaaS, data center, data cloud platforms and AI as separate technology categories. In plain terms, it’s no longer just the Azure invoice.
Who should own cloud costs when you don’t have a FinOps team?
Make the cloud or infrastructure lead accountable, give finance ownership of the forecast and the showback report, and make each application owner responsible for their own spend. One monthly 45-minute review between those three is enough to start.

That mirrors how the industry has settled it. In the FinOps Foundation’s State of FinOps 2026, a survey of 1,192 practitioners covering more than $83 billion in annual cloud spend, 78% of FinOps practices reported to the CTO or CIO and only 8% to the CFO. Cost lives with technology. Finance keeps it honest.
| Task | Cloud or infrastructure lead | Finance (controller or FP&A) | Application owners |
|---|---|---|---|
| Tagging policy and enforcement | Accountable | Consulted | Responsible for their resources |
| Rightsizing and cleanup | Accountable | Informed | Responsible |
| Reservations and savings plans | Recommends | Approves | Consulted on future usage |
| Budget, forecast and showback report | Consulted | Accountable | Informed |
| Anomaly alerts | Accountable | Informed | Responsible for responding |
| Monthly cost review | Runs it | Attends | Attend when their spend moved |
A simple cloud cost ownership model for a team with no FinOps department.
Why not give it all to finance? Because finance can see the bill but can’t resize a VM. And why not all to IT? Because engineers rarely feel the budget the way a controller does, and a cloud lead juggling patching, outages, security reviews and a migration backlog will push cost work to next month every single month unless someone outside IT is expecting the number. Split it.
What should you do in the first 30, 60 and 90 days?
Spend the first 30 days on quick wins and visibility, days 31 to 60 on tagging, showback and a unit-cost metric, and days 61 to 90 on commitment discounts and architecture changes. Buying discounts before cleanup locks in waste.
| Days | Focus | Work | Done when |
|---|---|---|---|
| 1 to 30 | Quick wins and visibility | Act on Azure Advisor or AWS Cost Optimization Hub recommendations, delete unattached disks and idle IPs, schedule dev and test shutdowns, set budgets and anomaly alerts | Every subscription or account has a budget and an owner |
| 31 to 60 | Foundations | Enforce required tags through Azure Policy or infrastructure as code, build a showback report, pick one unit-cost metric | At least 90% of spend is tagged to an owner and cost center |
| 61 to 90 | Structural savings | Size reservations and savings plans to the cleaned baseline, review architecture for the top ten cost drivers, set a quarterly commitment review | Commitment coverage matches steady usage, and the review is on the calendar |
A 90-day cloud cost optimization plan.
The 90% tagging target is ours, not an industry benchmark. It’s high enough to make showback credible and low enough that you won’t spend a month chasing the last stray storage account.
Which quick wins cut the bill fastest on Azure and AWS?
Idle and orphaned resources first, then oversized compute, then schedules for non-production environments. Native tools already list most of these. Azure Advisor and AWS Cost Optimization Hub both surface rightsizing and idle-resource recommendations at no extra charge.

On AWS, Cost Optimization Hub pulls rightsizing, idle-resource and commitment recommendations across accounts and Regions into one list, and de-duplicates the savings so you don’t count the same dollar twice. On Azure, Advisor does a similar job inside Cost Management.
- Unattached managed disks and old snapshots. Boring, and almost always there.
- VMs running at single-digit CPU for weeks. Resize or shut them down.
- Dev and test environments that run 168 hours a week when people use them for maybe 50
- Do you already own Windows Server or SQL Server licenses with Software Assurance? Azure Hybrid Benefit can apply them to cloud VMs.
- Storage sitting in hot tiers years after anyone last read it, which lifecycle rules can move to cool or archive automatically.
The catch with recommendation tools is that they show you the problem and stop there. Somebody still has to act. That’s what the ownership model above is for.
Reserved instances vs savings plans, which should you buy first?
Rightsize first, then follow Microsoft’s order. Exchange or trade in reservations that no longer fit, buy reservations for stable, predictable workloads, and cover the flexible remainder with savings plans. Reservations give deeper discounts. Savings plans tolerate change.

Microsoft’s guidance on choosing between the two lays out that exact sequence, and includes a line worth printing out. Discounts reduce rates, not waste. Buy a three-year reservation for an oversized VM and you’ve just committed to the waste for three years. Ouch.
| Option | Maximum discount | Term | Flexibility |
|---|---|---|---|
| Azure reserved VM instances | Up to 72% | 1 or 3 years | Narrow, but exchanges and trade-ins are possible |
| Azure savings plan for compute | Up to 65% | 1 or 3 years | Applies across VM families and regions |
| Azure savings plan for databases | Up to 35% | 1 or 3 years | Covers eligible database services |
| AWS Savings Plans | Up to 72% | 1 or 3 years | Compute plans apply across instance family, size, Region, Fargate and Lambda |
| AWS Database Savings Plans | Up to 35% | 1 year, no upfront | RDS, Aurora, DynamoDB, ElastiCache and other database services |
Commitment discount options on Azure and AWS, with published maximum discounts.
Two warnings. Azure savings plans can’t be canceled or refunded, and any hourly commitment you don’t use simply expires. And those maximums are maximums. Real discounts on a typical mix of VMs land well below the headline.
Our rule of thumb for a first purchase is to cover only the usage that has been flat for the last 60 to 90 days, after cleanup, and leave growth on pay-as-you-go until it proves itself. It’s conservative. We’d rather leave a few points of discount on the table than lock in a commitment that a re-architecture makes useless six months later.
How do tagging, showback and chargeback make savings stick?
Tags tie every cost to an owner and cost center. Showback reports those costs to the teams that create them, without billing them. Chargeback actually bills them. Start with showback, because seeing their own number changes behavior faster than any policy.
Two Azure details trip people up. Microsoft notes that tags aren’t retroactive in cost reporting, so cost from before you tagged stays untagged. And reservation purchases can’t be tagged at all, so you’ll allocate those through amortized cost views instead.
Then pick one unit-cost metric the business already understands. Cloud cost per order shipped, per invoice processed, per active user. When the bill grows 15% and orders grew 25%, that’s a good month. Without the unit metric, it looks like a bad one. Context is everything.
How do you control AI and GPU spend?
Treat AI as its own cost category. Put experiments in separate subscriptions or accounts with hard budgets, schedule GPU shutdowns, estimate token and GPU costs before a feature ships, and review AI spend monthly with the same owners as everything else.

The shift has been fast. In State of FinOps 2026, 98% of practitioners said they now manage AI spend, up from 31% in 2024. Flexera tied 2026’s rise in cloud waste largely to AI. A GPU VM left running over a long weekend can cost more than a month of the small VMs your team spent a week rightsizing. Expensive weekend.
What we see go wrong most is the pilot that never shut down. Someone spun up a GPU cluster or a high-throughput model deployment for a proof of concept, the demo went fine, and the meter kept running. A tag with an expiry date and an alert on it would have caught it. Our guide on moving AI from pilot to production covers how to budget for the production version properly.
Should Dynamics 365 and Microsoft 365 be part of cloud cost reviews?
Yes. In a mid-market Microsoft shop, Microsoft 365 and Dynamics 365 licenses can cost as much as Azure itself. The FinOps Foundation now includes SaaS in scope, and Azure Cost Analysis can show eligible Microsoft 365 and Dynamics 365 charges next to Azure.
State of FinOps 2026 found 90% of practitioners already manage or plan to manage SaaS spend. For Microsoft customers it’s practical, too. Cost Analysis can filter by provider to show Microsoft 365 and Dynamics 365 charges bought under the New Commerce experience alongside Azure. One review, one view.
The savings are different in kind. Unassigned licenses, users on Dynamics 365 full licenses who only need team member access, sandboxes nobody uses. Our Dynamics 365 team finds those during license reviews, and in some estates they’re worth more than every Azure quick win combined.
What is FOCUS, and does a mid-market company need it?
FOCUS is an open billing data standard from the FinOps Foundation that puts cost data from different clouds and SaaS vendors into one common format. If you run more than one cloud, or Azure plus Snowflake or Databricks, it saves building a separate cost report for each.
The FinOps Foundation ratified FOCUS 1.4 on June 4, 2026, adding invoice detail so cost data can be reconciled against actual invoices. The specification site lists data generator documentation for providers including AWS, Microsoft, Google Cloud, Databricks, Snowflake and Oracle.
One practical caveat. Azure’s own FOCUS export documentation still lists versions 1.0 and 1.2-preview, so don’t build reports that depend on 1.4 columns until your provider supports them. For a single-cloud Azure shop, the native Cost Management exports are fine. Add FOCUS when a second provider shows up.
Native tools or a third-party FinOps platform?
Start native. Seriously. Azure Cost Management and Advisor, or AWS Cost Explorer, Compute Optimizer and Cost Optimization Hub, cover visibility, recommendations, budgets and anomaly alerts for a single-cloud estate at no added license cost. A Power BI report on top of the cost exports handles showback.
A third-party platform starts earning its fee when three things pile up. Multiple clouds, Kubernetes clusters where cost has to be split by namespace or team, and SaaS plus AI token spend that native tools won’t put in one view. Until then, it’s another tool someone has to own. Another login. Another renewal. And yes, we’d say that even though some clients ask us to implement those platforms.
Who we are
Upen Group is a cloud, data and digital transformation company headquartered in Irving, Texas, working with mid-market and enterprise clients across Azure, AWS and Google Cloud. Our cloud engineers handle architecture, infrastructure as code, containers and cost optimization, and our cloud strategy work sets the budgets and roadmap the savings have to fit.
The order that saves money
Cloud cost optimization isn’t a tool purchase or a one-off cleanup. Name an owner and a monthly review. Clear out idle and oversized resources in the first 30 days. Tag, show back and track one unit-cost metric in the next 30. Then, and only then, buy reservations and savings plans sized to what’s left, and keep an especially close eye on AI and GPU spend.
Bill creeping and not sure why? Ask for a cloud cost review, and we’ll find the idle resources, size your commitments to real usage and hand your team a 90-day plan to keep it down.
Cloud bill questions we get asked
How much cloud spend is wasted, really?
29%, by respondents’ own estimate in Flexera’s 2026 State of the Cloud report, and that number went up for the first time in five years. Self-reported estimates tend to run low, because people can’t count waste they haven’t found.
FinOps vs cloud cost optimization, is there a real difference?
Yes. Optimization is the cleanup. FinOps is the ownership, reporting and review cadence that stops costs drifting back.
Should we buy reserved instances or savings plans first?
Neither, until you’ve rightsized. After that, reservations for steady, predictable workloads where the deeper discount pays, and savings plans for usage that moves between VM sizes, regions or services. Microsoft’s own guidance follows that order, and it’s the one we use with clients on both Azure and AWS.
Can we cancel an Azure savings plan if usage drops?
No. Azure savings plans can’t be canceled or refunded, and unused hourly commitment expires. Reservations are more forgiving, since they can be exchanged or traded in toward a savings plan.
What are AWS Database Savings Plans?
A commitment discount AWS launched on December 2, 2025. It offers up to 35% off eligible database services like RDS, Aurora and DynamoDB on a one-year term with no upfront payment.
Do we need a FinOps team to get started?
Short answer, no. A named owner in IT, a finance partner who owns the forecast, application owners who see their own costs and a monthly review will get a mid-market company most of the way. Hire or outsource FinOps when multi-cloud, Kubernetes and AI spend make the monthly review run long.
