Financial TimesAmericas' Fastest-Growing 2025
Certified PartnerOdoo Gold · Adobe Gold Solution · Shopify Premier partnershipsOdoo Gold · Adobe Gold · Shopify Premier partnerSee all
One business dayA senior architect replies to every inquiry · +1 (631) 897-7276Architect reply within one business dayLet's talkCall us
Inc. Regionals 2025Fastest-growing private companies in AmericaInc. Regionals 2025 honoreeSee awards
Americas' Fastest-Growing Companies 2025Recognized by the Financial TimesFT Fastest-Growing Companies 2025See the rankingRanking
4.8 / 5 on Clutch265 verified client reviews4.8/5 on Clutch · 265 reviewsRead the reviews
4.8 · 265 Clutch reviews
US United States
February 9, 202111 min readUpdated on October 14, 2026

How to control cloud costs: 9 steps, ownership first and discounts last

· Vice President, Microsoft Technologies
How to control cloud costs: 9 steps, ownership first and discounts last

To control cloud costs, work in a fixed order: give every dollar an owner, remove what nobody uses, resize what's left, and only then buy discounts. Most teams run it backwards. A budget scare hits, someone signs a three-year commitment, and the waste gets locked in at a lower rate.

The order matters because the waste is real. Flexera's 2026 State of the Cloud report, a survey of 750+ cloud decision-makers, puts wasted cloud spend at 29%. That's the first increase in five years, and Flexera ties it to AI workloads. On a $5M annual bill, close to $1.5M is buying nothing. In the same survey, 85% named managing cloud spend as a top challenge, so if your finance team keeps asking about the bill, you're in the majority.

What you need before you try to control cloud costs

Four things, and none of them is a new tool.

  • Billing access for someone in finance and someone in engineering. The FinOps Foundation frames FinOps as financial accountability shared between engineering, finance and business teams. If only one side can see the bill, the other side can't act on it.
  • At least 3 months of usage data, ideally 12. You'll need it to size commitments in step 6 and to negotiate in step 9.
  • A short tag set agreed in writing: cost center, application, environment and owner.
  • One named owner for the program, with a standing monthly slot on the calendar.

A note on scale: if the whole bill is a few thousand dollars a month, skip the formal program. A weekly look at your provider's cost console will do more for you than a FinOps process.

The nine steps below follow the FinOps Foundation's three phases: Inform (see and allocate the spend), Optimize (cut usage, then lower rates) and Operate (keep doing it). Spread across one quarter, that's a realistic plan for getting cloud costs under control.

Weeks Steps What you should have at the end
1–2 1. Tags and cost allocation Spend split by team and app, with the untagged share measured
3–4 2. Team budgets and forecast alerts Alerts that reach someone who can act
5–8 3–5. Idle cleanup, rightsizing, storage and data transfer A lower, steadier baseline
9–10 6–7. Commitments and spot capacity Discounted rates on work you know you'll run
11–13 8–9. SaaS and license review, contract prep, monthly review A repeatable rhythm, not a one-time cleanup

Step 1: Give every dollar an owner

You can't control spend you can't attribute. Enforce the tag set in your infrastructure-as-code templates, not in a wiki page, and measure the untagged share every week until it's small.

Two platform details trip teams up:

  • AWS tags stay invisible in cost reports until someone activates them. Cost allocation tags are switched on in the Billing console and can take up to 24 hours to appear.
  • Azure resources don't inherit tags by default. Turn on tag inheritance in Cost Management so subscription and resource group tags reach usage records. Changes apply to the current month within about 24 hours.

Kubernetes needs its own answer, because a node tag can't say which team's pods ran up the bill. OpenCost, a CNCF incubating project, splits container costs by namespace, deployment and label.

Then send the numbers back to the teams that spent them. Showback reports each team's spend. Chargeback moves it into their budget. Start with showback, since chargeback lands better once teams trust the data.

Step 2: Put a budget on every team and a person on every alert

A company-wide budget tells you the bill is over plan. A budget per product or team tells you who is over, while there's still time to do something about it.

AWS Budgets alerts on actual and forecasted spend and can run budget actions, such as applying an IAM policy that blocks new resources. Its data refreshes up to three times a day. Azure budgets evaluate costs every 24 hours and can trigger action groups for automation.

Neither one caps spending by default. When an Azure budget is exceeded, Microsoft says "resources aren't affected, and your consumption isn't stopped," so usage simply carries on. AWS can block new resources, but only if you've set that action up in advance. A budget is a smoke detector, not a sprinkler. Decide now who receives each alert and what they're expected to do within a day.

Step 3: Switch off what nobody is using

Development and test environments rarely need to run nights and weekends. A machine that runs 50 business hours a week instead of 168 costs about 70% less on hourly pricing.

Put it on a schedule rather than on someone's memory. Instance Scheduler on AWS (version 3.2.13, October 2026) starts and stops EC2 instances, Auto Scaling groups and RDS databases from resource tags. On Azure, VM auto-shutdown turns machines off at a set time each day.

Next, go looking for leftovers: unattached disks, old snapshots, unused IP addresses and load balancers with no targets. AWS Compute Optimizer identifies idle resources across EC2, EBS, RDS, NAT gateways and more. Azure Advisor flags VMs with almost no CPU or network use over a lookback window you choose. Old snapshots and orphaned disks are pure storage cost with no workload behind them.

Step 4: Rightsize to what workloads actually use

Rightsizing matches instance size to measured utilization instead of the size someone guessed at launch. Compute Optimizer analyzes 14 days of CloudWatch metrics by default, or 93 days with its paid enhanced metrics. Azure Advisor lets you set a lookback from 7 to 90 days.

Pick a window that covers your whole business cycle, including month-end close. Leave memory-heavy workloads alone until memory metrics are collected, because CPU numbers on their own can mislead. To reduce cloud infrastructure spend without hurting performance, change one workload at a time and watch latency and error rates for a week before moving to the next.

Do this before step 6, not after. Azure's documentation warns that Advisor's resize recommendations don't account for reservations or savings plans you've already bought. Following them after you commit can, in some cases, raise costs.

Step 5: Move cold data down a tier and watch what leaves the cloud

Storage grows quietly while everyone watches compute. Most data is written once and rarely read again.

Amazon S3 Intelligent-Tiering moves objects not accessed for 30 days to an Infrequent Access tier, saving 40%. After 90 days, they move to Archive Instant Access, saving 68%. Objects smaller than 128 KB aren't auto-tiered, so check object sizes before you count on the savings. On Azure, Blob Storage lifecycle management moves data to the cool, cold or archive tiers after a period without access or changes that you define. The policies themselves are free.

Data transfer deserves its own line in the review. AWS includes 100 GB of free data transfer out to the internet each month, pooled across services. Past that, egress, cross-region replication and NAT gateway traffic add up fast. Where services sit relative to each other is an early architecture choice whose cost lasts for years.

Step 6: Commit to your baseline, not your peak

Only now does buying discounts make sense. For predictable workloads, commitments are the largest single lever: you promise a level of usage for 1 or 3 years, and the provider cuts the rate.

Option Maximum published discount Flexibility
AWS Compute Savings Plans Up to 66% Applies across EC2, Lambda and Fargate
AWS EC2 Instance Savings Plans Up to 72% Tied to an instance family in a region
AWS Reserved Instances (Standard) Up to 72% Least flexible; Convertible RIs trade discount for exchanges
Azure savings plan for compute Up to 65% Hourly dollar commitment across compute services
Azure Reserved VM Instances Up to 72% Specific VM types; exchange rules tighten in 2027

Those maximums assume 3-year terms and specific configurations, so your real rate will be lower. AWS bills any usage above a Savings Plan commitment at On-Demand rates.

Size the commitment to the floor of your usage, not the average. Leave growth and spikes on pay-as-you-go and add commitments in layers each quarter. On AWS, Compute Savings Plans are usually the simpler first purchase, because one commitment covers EC2, Lambda and Fargate. Instance Savings Plans and Standard RIs go deeper but tie you to an instance family, so save them for very stable workloads.

Watch the calendar, too. Microsoft says Azure reservations bought on or after February 1, 2027, will no longer be eligible for exchange. On Google Cloud, the equivalent lever is committed use discounts, sold as spend-based or resource-based commitments.

Step 7: Run interruptible work on spot capacity

Spot capacity is spare provider capacity sold cheap, on the condition that the provider can take it back. AWS Spot Instances can cost up to 90% less than On-Demand and come with a two-minute interruption notice. Azure Spot Virtual Machines offer up to 90% off pay-as-you-go, with a 30-second eviction notice when scheduled events are on.

CI/CD pipelines, batch jobs, rendering, data processing and test environments suit it well. Stateful services that can't checkpoint don't, and neither does anything with a hard deadline.

You don't have to watch prices. AWS sets the Spot price and adjusts it gradually with long-term supply and demand. The real effort goes into jobs that survive an interruption. Spot usage also sits outside your Savings Plans: it isn't covered by them and doesn't count toward the commitment.

Step 8: Pull SaaS and licenses into the same review

The bill finance worries about isn't only infrastructure. Seats in collaboration, CRM and analytics tools often renew without anyone checking who still logs in. The FinOps Foundation's 2026 State of FinOps survey of 1,192 practitioners found 90% are now asked to manage SaaS and 64% manage licensing. Almost all of them, 98%, manage AI spend as well. Whatever review you build for cloud infrastructure will soon be expected for every bill that arrives priced on usage.

Run a seat review against login data before each renewal. Then check the licenses you already own. Azure Hybrid Benefit lets eligible Windows Server, SQL Server and Linux licenses count toward Azure VM costs. Microsoft quotes savings of up to 85% for SQL Server versus pay-as-you-go, varying by region and instance type. If you're paying for the same license twice, this can be the quickest saving in the whole sequence.

Step 9: Negotiate with 12 months of clean data, then make it monthly

Enterprise agreements trade a spend commitment for better rates. On Azure, a Microsoft Azure Consumption Commitment (MACC) commits you to a set level of Azure spend over a defined period.

Read the downside first. Microsoft's documentation says that if you miss the MACC amount by its end date, a shortfall charge is applied for the remaining balance. Commit to what your tagged, forecasted baseline supports, not to the growth plan in last year's board deck.

Bring 12 months of usage split by workload, which steps 1 and 2 have now made possible. Ask about price protection at renewal, room to shift spend between services and how marketplace purchases count. Procurement and finance belong in the room with engineering.

After that, the work becomes a habit rather than a project. Put engineering, finance and product owners in one monthly meeting and report by business unit. Three items are enough: the variance against budget, the top 5 cost movers and the open savings actions with their owners.

If the bill spans AWS and Azure and nobody owns the full picture, our cloud services team would begin at step 1. That means a report of untagged spend and a proposed owner for each line. You can tell us what you're running and start there. For Azure-heavy estates, our Microsoft Azure cloud services team can take on the tagging, budgets and reservation planning.

The tools you already pay for

You don't need a new platform for the first 90 days. Each major provider includes native cost management tools that cover the steps above.

Job AWS Azure Google Cloud
See and forecast spend AWS Cost Explorer: 13 months of history, 18-month forecast Microsoft Cost Management: cost analysis Cloud Billing reports
Budgets and alerts AWS Budgets Cost Management budgets Cloud Billing budgets
Rightsizing and idle resources AWS Compute Optimizer Azure Advisor FinOps hub recommendations
Commitment discounts Savings Plans, Reserved Instances Savings plans, reservations Committed use discounts

Cost Explorer's console is free; its API costs $0.01 per paginated request. Microsoft Cost Management is the same product many teams still call Azure Cost Management. Google's FinOps hub puts realized savings, active recommendations and potential wasted usage on one dashboard.

When the bill still won't come down

Most stalls trace back to one of these four causes.

Rightsizing barely moved the number. The app may scale poorly by design. If every request hits a database that has to run on the largest instance, the fix is refactoring, and that's a DevOps and engineering decision rather than a billing one.

Spend swings too much to commit. Keep commitments to the floor you can see and lean on autoscaling, forecast alerts and spot capacity. For event-driven work, serverless can fit better: AWS Lambda event functions are priced on requests and the time your code runs, so idle time adds no duration charges.

AI workloads are driving the growth. GPU capacity, model inference and token usage need their own unit metrics, and they move faster than VM spend. The companion piece on AI in cloud computing covers predictive scaling and anomaly detection in more depth.

Bills from several clouds won't reconcile. Native tools stop at their own border. The FOCUS specification (FinOps Open Cost and Usage Specification) normalizes billing data across vendors, and AWS, Microsoft and Google Cloud all offer FOCUS-format cost data. That turns multi-cloud reporting into a data job instead of a spreadsheet project. Third-party cost management software starts to make sense once you need one view across several providers, Kubernetes and SaaS.

FAQ

How can businesses control growing cloud spend?

Give every dollar an owner through tagging and cost allocation, then set budgets per team so someone watches each limit. Next, remove idle resources and rightsize what's left. Only then buy Savings Plans or reservations sized to the new baseline. Review the results monthly with engineering and finance in the same meeting.

What are the best strategies to control unpredictable cloud costs?

Forecast-based budget alerts, anomaly detection and autoscaling handle unpredictable spend better than long commitments. Keep the spiky share of usage on pay-as-you-go and run interruptible work on spot capacity. Commit only to the steady floor you can see in at least 3 months of data.

Do you need a FinOps team to control cloud costs?

Not at the start. Flexera's 2026 report found 63% of organizations have a FinOps team, but a smaller company can begin with one named owner and a monthly review. Dedicated staff make more sense once the bill spans several providers or business units.

Your first move this week

Pull last month's bill and measure what share of it has no owner tag. That single number tells you whether to start at step 1 or move straight to step 3.

ShareLinkedInXEmail

Related reading

Modernizing your cloud and delivery pipeline?

Explore DevOps services