Skip to content
NeuralYug

Insights · Cloud

FinOps for small teams: where your cloud bill actually goes

08 Sept 20268 min readNeuralYug

The short version

Flexera's 2026 State of the Cloud Report puts average wasted cloud spend at 29%, up from 27% in 2025 and the first increase in five years. That figure comes from a survey of 750+ mostly larger organisations, many with dedicated FinOps teams. Small teams without anyone watching the bill often do worse, because the obvious leaks have never been looked at.
A NAT gateway bills twice: by the hour just for existing, and by the gigabyte for every byte it processes. In the Mumbai region, as of September 2026, that is $0.056 per hour (about $41 a month idle) plus $0.056 per GB. Traffic to S3 that goes through it can often be moved to a free S3 gateway endpoint instead.
Only for the part of your usage that is steady and will still be there in a year. AWS says Compute Savings Plans cut up to 72% off On-Demand prices, but you commit to an hourly spend for one or three years. Rightsize and switch off waste first, then commit to your lowest normal month, not your average one.

Most small teams waste somewhere between a quarter and a third of their cloud bill. Flexera's 2026 State of the Cloud Report puts average wasted spend at 29%, up from 27% a year earlier and the first rise in five years. It rarely goes on one big mistake. It leaks through a handful of line items nobody watches: idle servers, forgotten disks, NAT gateways, data transfer and dev machines that never switch off.

This post is about finding those leaks on AWS, Google Cloud, Azure or DigitalOcean when you are a five-to-fifty person company paying in dollars from Nepal. It is not about where to host. It is about what happens to the bill once you are already there.

How much of a cloud bill is actually wasted?

Flexera surveyed more than 750 cloud decision-makers for its March 2026 report. 85% said managing cloud spend is a top challenge, and 63% now have a dedicated FinOps team. Read those two numbers together. Even organisations that pay people to watch cloud costs full-time still waste close to three dollars in ten.

The FinOps Foundation's State of FinOps 2026, with 1,192 respondents managing more than $83 billion in annual cloud spend, tells the same story from the practitioner side. Workload optimisation and waste reduction is still the number one priority. Respondents also say they have already cleared the big rocks, and what is left is a long tail of small fixes that take more effort each.

Both surveys lean towards large companies, so treat 29% as a reference point, not a forecast for your account. But the second finding is quietly good news for a small team. If nobody has ever looked at your bill line by line, your big rocks are probably still sitting there.

Why does a cloud bill surprise hurt more in Nepal?

Because the bill is in dollars and your revenue mostly isn't. On 21 March 2026 Nepal Rastra Bank fixed the rate at Rs 150.24 to the US dollar, the weakest in the rupee's history, as The Rising Nepal reported. Two years earlier it was Rs 132.76. A $1,000 monthly bill that cost Rs 1,32,760 in March 2024 cost Rs 1,50,240 in March 2026. That is 13% more in rupees for the exact same usage.

Then there is the card. NRB's April 2026 circular, covered by myRepublica, lets IT-service exporters pay up to $100,000 a year for software and technology, and lets people earning income online pay up to $25,000 a year for technology services through banking instruments. Useful, but plenty of SMEs still pay the cloud bill on a founder's prepaid dollar card with a bank-set limit. A surprise spike can push the charge past that limit, and a declined payment turns a cost problem into an uptime problem.

So for a Nepali team, predictability matters almost as much as the total. A bill that is $50 higher but steady is easier to live with than one that jumps $300 in a month nobody planned for.

Where does a small team's cloud bill actually go?

In our experience the same culprits show up on almost every small account. Prices below are AWS list prices for the Asia Pacific (Mumbai) region as of September 2026, before VAT and your bank's foreign-exchange margin. Mumbai is the region most Nepali teams pick for latency.

  • Idle and oversized servers. Instances sized for launch-day fear and never revisited, often running at single-digit CPU.
  • Unattached disks and old snapshots. A deleted server can leave its volume behind. gp3 storage costs $0.0912 per GB-month and snapshots $0.05 per GB-month, whether anything uses them or not.
  • NAT gateways. $0.056 an hour just to exist (about $41 a month) plus $0.056 for every GB that passes through, including traffic to AWS's own services.
  • Data egress. After the first 100 GB a month, data out to the internet from Mumbai costs $0.1093 per GB for the first 10 TB.
  • Dev and staging left on 24/7. A week has 168 hours. A team works about 60 of them.
  • Over-provisioned managed databases. A production-sized RDS instance, sometimes with Multi-AZ, behind a staging app nobody uses at night.
  • Log retention. CloudWatch Logs charges $0.67 per GB ingested and $0.03 per GB-month stored, and new log groups keep data forever unless you set a retention period.
  • Public IPv4 addresses. $0.005 an hour each, in use or idle.
  • No commitments. Paying On-Demand rates for servers that have run every hour of every day for two years.

What six common leaks cost per month (US$)

One month of each, at list price, before VAT and bank FX margin

AWS Mumbai list prices, Sep 2026
0132538501 TB of old snapshots500 GB unattached gp3 volume1 idle NAT gateway50 GB/month of logs ingested300 GB egress over free tier5 idle public IPv4 addresses

Together: about $221 a month, roughly Rs 33,200 at the March 2026 rate of Rs 150.24 per dollar. None of these lines serves a single customer. Calculated from AWS list prices, ap-south-1, effective Sep 2026: snapshots $0.05/GB-mo (1,000 GB); gp3 $0.0912/GB-mo; NAT gateway $0.056/hr x 730 hr; CloudWatch Logs ingestion $0.67/GB; data transfer out $0.1093/GB; public IPv4 $0.005/hr x 730 hr. FX: Nepal Rastra Bank rate via The Rising Nepal, 21 Mar 2026.

What six common leaks cost per month (US$) — data table
CategoryUS$ per month
1 TB of old snapshots50
500 GB unattached gp3 volume45.6
1 idle NAT gateway40.9
50 GB/month of logs ingested33.5
300 GB egress over free tier32.8
5 idle public IPv4 addresses18.3

Illustrative, not an average account. Each figure is a single list-price line you can check against your own bill.

How is a cloud bill actually metered?

Nobody sets a total. The big clouds meter almost everything separately. Compute bills by the second or hour, storage by the GB-month, network by the GB, and managed services by requests or capacity units. Your invoice is the sum of hundreds of small meters, each running on its own. Simpler providers like DigitalOcean bundle more into a flat monthly price, which is a big part of their appeal to small teams. Even there, extra volumes, snapshots, backups and bandwidth overage are billed on top.

Two things follow. Costs keep running after a person stops thinking about them, so a test server from March is still billing in September. And the expensive line is often not the obvious one. Teams tend to watch server prices closely and never look at the NAT gateway or log ingestion, which quietly pass every byte through a meter.

How does a small-team FinOps loop work?

FinOps sounds like a department. For a small team it is a habit: one owner, half an hour a week, and a short loop that turns the raw bill into decisions. The order matters, because each step depends on the one before it. You can't rightsize what you can't attribute, and you shouldn't commit to spend you haven't cleaned up yet.

The small-team FinOps loop

From metered usage to a bill someone owns. Tap a component to see what it does and why it sits where it does.

How it works
Where cost is createdSee itGet warnedCut it

Tap any component above for its role and the real tech.

  1. Running resources (Service, EC2, RDS, EBS, NAT gateway, S3, logs): Every resource emits usage: instance-hours, GB-months, GB transferred, GB ingested. Nothing here knows or cares about your budget. This is where the 29% leaks happen.
  2. Cost allocation tags (Service, project, env, owner): Three tags are enough for a small team: which project, which environment (prod, staging, dev) and who owns it. Untagged spend is unowned spend. Enforce tags at creation time in Terraform or a policy, because tagging after the fact never finishes.
  3. Billing export (Data, AWS Data Exports (CUR) / GCP billing export to BigQuery / Azure cost exports): The line-item record of every meter, landed in storage you can query. This is the raw truth. The console's summary view is fine for a glance, but questions like 'what did staging cost last month' need the export.
  4. Cost dashboard (Service, AWS Cost Explorer / a simple BI view): Spend grouped by tag, service and day. The weekly review happens here: sort by change since last week, not by total, because the thing that grew is the thing to look at.
  5. Budgets and anomaly alerts (Service, AWS Budgets, Cost Anomaly Detection): A monthly budget with alerts at 50%, 80% and 100% of forecast, plus anomaly detection for sudden spikes. For a team paying on a limited dollar card, this is the single most important box on the diagram.
  6. Named cost owner (External, email or team chat channel): Alerts go to a person, not a shared inbox nobody reads. One owner per project tag. They don't have to fix everything, they have to notice and ask.
  7. Rightsizing and cleanup (Service, AWS Compute Optimizer, Trusted Advisor): Uses weeks of real utilisation to recommend smaller instances, and flags unattached volumes, idle load balancers and unused IPs. Do this before buying any commitment, or you lock in the waste.
  8. Off-hours scheduler (Queue, Instance Scheduler / cron-triggered Lambda): Stops dev and staging outside working hours. Running roughly 60 of 168 weekly hours cuts those environments' compute hours by about 64%.
  9. Commitments (Service, Compute Savings Plans / Reserved Instances): Once the baseline is clean, commit to the steady part of it. AWS says Savings Plans save up to 72% against On-Demand. Commit to your lowest normal month, not your average, so a quiet quarter doesn't leave you paying for capacity you've stopped using.

The loop is the product. Tools are interchangeable across AWS, Google Cloud and Azure; the order is not.

Which fixes should a small team do first?

Start with the ones that are low effort and can't break anything. Deleting an unattached volume after a snapshot is safer than resizing a production database. Commitments come last, because they lock in whatever you are running on the day you sign.

Eight fixes, roughly in the order to do them

Sort by effort or filter by the line item you're worried about

Priority list
Effort ratings are our judgement for a team with one part-time cloud owner. Savings are the vendor's stated maximums or straight list-price arithmetic, not guarantees.
Cited saving
1Set a budget with forecast alertsSurprises, card declinesLowPrevents spikes rather than cutting spend
2Delete unattached volumes, prune old snapshotsEBS storageLow100% of that line: $0.0912/GB-mo (gp3), $0.05/GB-mo (snapshots)
3Release idle public IPs, set log retentionIPv4, CloudWatch LogsLow$0.005/hr per IP; stops $0.03/GB-mo log storage growing forever
4Migrate gp2 volumes to gp3EBS storageLow20% lower cost per GB (AWS); $0.114 to $0.0912 in Mumbai
5Schedule dev and staging off-hoursCompute, managed DBsLow to mediumAbout 64% of those environments' hours (60 of 168)
6Route S3 traffic via a gateway endpoint, review NAT useNAT gateway, egressMediumRemoves $0.056/GB NAT processing on that traffic
7Rightsize instances and databases from real utilisationCompute, RDSMediumVaries by account; measure 2 to 4 weeks first
8Commit the steady baseline; Spot for batch jobsComputeMediumUp to 72% (Savings Plans), up to 90% (Spot), vs On-Demand

Effort ratings are our judgement for a team with one part-time cloud owner. Savings are the vendor's stated maximums or straight list-price arithmetic, not guarantees. AWS list prices, ap-south-1, effective Sep 2026 (EC2/EBS/NAT/CloudWatch/VPC pricing); AWS Savings Plans (up to 72%), EC2 Spot (up to 90%) and EBS General Purpose (gp3 20% lower per GB than gp2) product pages.

Rows 1 to 4 are an afternoon's work on most small accounts.

What does this look like on a real five-person team's bill?

Take the six lines in the chart above. None of them is dramatic. An old NAT gateway from an abandoned VPC, a test disk from a server someone deleted, snapshots from a backup script that never prunes, verbose debug logging left on in production. Together they come to about $221 a month, or roughly Rs 33,200 at March's exchange rate. That's close to Rs 4 lakh a year, spent on nothing a customer ever touches.

The fix for most of it is deleting things, and the hard part isn't technical. It's that nobody is sure the old disk isn't needed. That's what tags and a named owner solve: when every resource says which project and person it belongs to, 'can I delete this?' becomes a one-message question instead of a month of hesitation.

When is FinOps not worth the effort?

If your whole bill is under about $100 a month, don't build a process. Set a budget alert, delete what you recognise as dead, and get back to your product. The loop above earns its half hour a week somewhere in the few-hundred-dollar range, and becomes essential once one line item can move your bill by more than your card limit allows.

It's also not a reason to switch clouds. Moving providers to save money usually moves the same habits with you. Our cloud cost audit and FinOps blueprint walks through the same sequence at a larger scale: tag everything, rightsize from real utilisation data, then commit and add guardrails so the savings don't erode. If deployments are what keep spawning stray resources, the fix often starts in the pipeline, as in our CI/CD modernisation blueprint.

What should you do this week?

  • Turn on a monthly budget with alerts at 50%, 80% and 100% of forecast, sent to a named person.
  • Open your bill by service and find your NAT gateway, EBS, snapshot and log lines. Most teams have never looked.
  • Add three tags (project, env, owner) to everything new, and enforce them in your infrastructure code.
  • Schedule dev and staging to stop at night and on weekends.
  • Wait at least a month of clean data before buying a Savings Plan.

If you'd like a second pair of eyes on the bill, our Cloud & DevOps team runs cost reviews like this for Nepali startups and SMEs, and hands the loop over so you can run it yourselves. Send us a rough monthly figure and your provider through our contact page, and we'll tell you honestly whether it's worth a proper audit.

Frequently asked

How much cloud spend do companies typically waste?

Flexera's 2026 State of the Cloud Report puts average wasted cloud spend at 29%, up from 27% in 2025 and the first increase in five years. That figure comes from a survey of 750+ mostly larger organisations, many with dedicated FinOps teams. Small teams without anyone watching the bill often do worse, because the obvious leaks have never been looked at.

Why is my AWS NAT gateway so expensive?

A NAT gateway bills twice: by the hour just for existing, and by the gigabyte for every byte it processes. In the Mumbai region, as of September 2026, that is $0.056 per hour (about $41 a month idle) plus $0.056 per GB. Traffic to S3 that goes through it can often be moved to a free S3 gateway endpoint instead.

Should a small startup buy AWS Savings Plans?

Only for the part of your usage that is steady and will still be there in a year. AWS says Compute Savings Plans cut up to 72% off On-Demand prices, but you commit to an hourly spend for one or three years. Rightsize and switch off waste first, then commit to your lowest normal month, not your average one.

Want this run on your numbers?

We'll do the same analysis on one of your workflows in the two-week Automation Sprint.

Book a Sprint →

Related service · Cloud & DevOps

Keep reading

All insights →