Quick answer: Cloud cost optimisation in the UAE starts with the bill, not a discount. Flexera estimates 29 per cent of cloud spend is wasted, and UAE regions list roughly a fifth above US East. Tag everything, schedule non-production, rightsize and remove idle servers first, then buy Savings Plans or reservations.
Most growing UAE companies reach the same moment: the cloud bill has doubled in eighteen months, revenue hasn’t, and nobody can say exactly why. The cause is rarely one big mistake. It’s dozens of small ones: test servers running all weekend, disks nobody deleted, instances sized for a launch that never came, and every hour billed at on-demand rates in a region that costs more than the US. This playbook shows where the waste hides, what the UAE regions actually cost, how commitments work, and gives you a calculator to estimate what a first FinOps pass could save.
Sources: Flexera 2026 State of the Cloud Report, March 2026 · AWS Savings Plans pricing · Azure Reserved VM Instances · AWS EC2 on-demand pricing and Azure Retail Prices API, Linux list prices checked 1 October 2026
What the UAE cloud regions cost compared with the US and Europe
Running in-country carries a premium. We pulled AWS and Azure list prices for common general-purpose instances in the UAE, US East and Western Europe. The pattern is consistent: the UAE regions cost roughly a fifth more than US East and only a few per cent more than Frankfurt or West Europe.
| Instance (on-demand, per hour) | UAE region | US East | Frankfurt / West Europe | UAE vs US East |
|---|---|---|---|---|
| AWS m7i.large, Linux | $0.1236 (me-central-1) | $0.1008 | $0.1208 | +23% |
| AWS m7g.large (Graviton), Linux | $0.1000 | $0.0816 | $0.0978 | +23% |
| Azure D2s v5, Linux | $0.118 (UAE North) | $0.096 | $0.115 | +23% |
| Azure D2s v5, Windows | $0.210 (UAE North) | $0.188 | $0.207 | +12% |
Sources: AWS EC2 on-demand pricing (Middle East (UAE), US East (N. Virginia), EU (Frankfurt)) · Azure Retail Prices API (UAE North, East US, West Europe). List prices in USD, checked 1 October 2026; your negotiated rates may differ.
Two conclusions follow. First, moving workloads to Europe to save money rarely works: the saving is two or three per cent, and you give up latency and an easy answer to every residency question. Second, US East is meaningfully cheaper, but for anything holding personal data about UAE customers the residency argument usually wins; our guide to cloud data residency in the UAE sets out which workloads genuinely have to stay in-country. The premium is real, so every wasted hour in a UAE region costs more than the same hour elsewhere.
Two local details belong in the budget. AWS has billed UAE accounts through branches in Abu Dhabi and Dubai since 1 February 2024 and charges 5 per cent VAT on all service charges, issuing tax invoices in the billing console; make sure your TRN is entered in the tax settings. And because the dirham is pegged at AED 3.6725 to the US dollar, USD-denominated cloud bills don’t swing with exchange rates, so any month-on-month increase is usage, not currency. This is general guidance, not tax advice; ask your accountant how VAT on cloud services is treated in your returns.
Sources: AWS tax help: UAE · AWS VAT rates summary table (EMEA) · Central Bank of the UAE exchange rates
Cloud cost optimisation in the UAE: where the waste hides
Flexera’s 2026 survey of more than 750 cloud decision-makers found that wasted spend rose to 29 per cent, the first increase in five years, with AI spending adding to the complexity. In growing UAE companies we see the same eight leaks again and again. None of them requires a new tool to find; they need someone to look.
| Leak | Why it happens | How to find it | Fix |
|---|---|---|---|
| Non-production running 24/7 | Development, test and staging built for a project and never scheduled | Filter by environment tag; look at night-time and weekend CPU | Schedule to working hours, or delete when the project ends |
| Oversized instances | Sized for a launch peak, a vendor’s recommendation or a guess | 30 days of average CPU and memory; AWS Compute Optimizer or Azure Advisor | Drop one size, load-test, repeat |
| Stopped but still billed | An Azure VM shut down from inside the OS stays allocated and billed | List VM power states; look for “Stopped” rather than “Deallocated” | Deallocate, or delete and keep a snapshot |
| Orphaned disks and snapshots | Disks outlive the servers they were attached to; snapshots never expire | Storage inventory sorted by age and attachment | Lifecycle and retention rules, applied by default |
| Old volume types | AWS gp2 volumes from older templates | Volume type report | Move to gp3, which costs 20% less per GB |
| Public IPv4 addresses | Every server given its own public address | IP address inventory | Remove unused addresses; put services behind one load balancer |
| Logs and monitoring | Debug logging left on, retention set to forever | Ingestion volume by source | Sensible retention, sampling, archive tiers |
| Licence-included images | Windows and SQL Server images chosen by default | Spend broken down by operating system | Bring your own licence where eligible, or move to Linux |
Sources: Microsoft Learn: Azure VM states and billing · AWS EBS gp3 · AWS public IPv4 charge · Flexera 2026
Two of these deserve a note because they surprise people. Since 1 February 2024 AWS has charged $0.005 per hour for every public IPv4 address, whether attached to something or not: about $3.65 a month each, which adds up quickly in an estate where every test server got its own address. And on Azure, a VM shut down from inside Windows sits in the “Stopped” state and keeps billing for compute; only “Deallocated” stops the meter, and even then disks and networking continue to charge.
AWS’s own migration guidance offers a useful yardstick for the oversized category: it calls applications averaging under 5 per cent CPU and memory use “zombie” applications, and those between 5 and 20 per cent over 90 days “idle”. If a server falls in either band, the question isn’t which size to drop to but whether it should exist.
Source: AWS Prescriptive Guidance: migration strategies
If a server falls in either band, the question isn’t which size to drop to but whether it should exist.
Make the bill legible before you cut it
The FinOps Foundation describes the practice as three repeating phases: Inform, Optimise and Operate. Most companies skip straight to optimising and watch the bill creep back within a quarter, because savings only stick when someone can see the spend and is accountable for it. In the Foundation’s 2025 survey of 861 practitioners managing around $69 billion of cloud spend, workload optimisation and waste reduction was the top priority by a clear margin, and governance was the top priority for the year ahead.
Sources: FinOps Foundation: framework phases · State of FinOps 2025
Making the bill legible is a week of work for a small estate. Separate accounts or subscriptions for production and non-production. Four mandatory tags on everything: owner, environment, product and cost centre, enforced by policy rather than goodwill. Budgets with alerts at 80 and 100 per cent for each account, sent to the owner, not just to finance. Anomaly detection switched on so a runaway service is caught in a day, not at month end.
Then pick one unit cost that the business understands: AED per order, per active customer, per booking or per enrolled student. Total spend going up is fine if unit cost is going down. A cost that has no owner won’t go down, however many dashboards you build, because nobody is accountable for switching anything off.
Rightsizing, scheduling and newer hardware: savings with no commitment
These three levers cost nothing to pull and carry no lock-in, which is why they come before any commitment purchase.
Rightsizing. AWS Compute Optimizer is free to use; you pay only for the resources and CloudWatch monitoring, plus an optional fee of about $0.25 a month per resource for enhanced metrics. Azure Advisor gives similar recommendations. Treat the output as a shortlist: drop one size at a time, watch memory and response times for a week, and note what changed so it can be reversed.
Source: AWS Compute Optimizer pricing
Scheduling. A week has 168 hours. A development environment needed from 8am to 6pm, Monday to Friday, needs 50 of them, so running it around the clock means roughly 70 per cent of the hours you pay for are unused. Since the UAE federal working week moved to Monday to Friday in January 2022, with most private employers reported to be following, weekends are a clean shutdown window for non-production in most companies. Schedules should be the default for every non-production resource, with an opt-out that has to be justified.
Source: Mondaq: a new working week in the UAE from 1 January 2022
Newer and ARM-based instances. AWS says Graviton instances cost up to 20 per cent less than comparable x86 instances. In the UAE region the numbers bear that out: m7g.large lists at $0.1000 an hour against $0.1236 for m7i.large, 19 per cent less. Many modern web stacks (Node, Python, Java, Go, containers, managed databases) run on ARM with little or no change; Windows workloads and some older binaries don’t. Moving gp2 volumes to gp3, which AWS prices 20 per cent lower per GB with a 3,000 IOPS baseline at any size, is the storage equivalent and usually a zero-downtime change.
Sources: AWS Graviton · AWS EC2 on-demand pricing, Middle East (UAE) · AWS EBS gp3
Commitments: Savings Plans, reservations and Spot
Once the estate is the right size, the remaining steady usage shouldn’t be paying on-demand rates. Both main providers sell discounts in exchange for a one- or three-year commitment, and AWS sells spare capacity cheaply if you can tolerate interruptions.
| Option | Vendor’s maximum discount | What you commit to | Flexibility | Best for |
|---|---|---|---|---|
| AWS Compute Savings Plans | Up to 66% | An hourly spend for 1 or 3 years | Any instance family, size, region or OS; also covers Fargate and Lambda | Companies still changing their architecture |
| AWS EC2 Instance Savings Plans | Up to 72% | An hourly spend on one instance family in one region, 1 or 3 years | Can change size within the family in that region | Stable workloads on a settled instance family |
| Azure Reserved VM Instances | Up to 72% | Reserved capacity for 1 or 3 years | Tied to the VM type and region reserved | Steady production VMs |
| AWS Spot Instances | Up to 90% | Nothing | Can be reclaimed with a two-minute warning | Batch jobs, CI builds, stateless workers |
Sources: AWS Savings Plans pricing · Azure Reserved VM Instances · AWS Spot Instances · AWS Spot interruption notices
The headline percentages are maximums. For a common size in a UAE region the real figure is lower: a D2s v5 Linux VM in Azure UAE North lists at $0.118 an hour, about $1,034 a year on pay-as-you-go, against $636 for a one-year reservation (38 per cent less) and $1,222 for three years against roughly $3,101 (61 per cent less). Run the numbers for your actual instance types before you build a business case on “up to 72 per cent”.
Source: Azure Retail Prices API, UAE North consumption and reservation prices for D2s v5, checked 1 October 2026
Run the numbers for your actual instance types before you build a business case on “up to 72 per cent”.
How OXEDGE does this
We run a cloud cost review that ends with changes made in your accounts, not a slide deck.
- Read-only review of your AWS or Azure bill and usage: an account, tag and owner map, and the eight leaks above checked one by one.
- Quick wins applied with your team: schedules, rightsizing, storage and IP clean-up, each change logged with a rollback note.
- A commitment plan sized from 90 days of real usage, with a monthly coverage and utilisation report.
- Budgets, anomaly alerts and a unit-cost dashboard handed over with a one-page monthly review runbook.
Four rules keep commitments from becoming a new kind of waste. Buy only after rightsizing, or you lock in the oversized estate for the length of the term. Cover the floor of your usage, typically the lowest hourly spend of the last three months, not the average. Prefer the flexible products (Compute Savings Plans on AWS) and one-year terms while the architecture is still moving. And review coverage and utilisation every month: an unused commitment is pure loss, and a commitment bought once and forgotten tends to drift out of line with the estate after the first re-architecture.
Licences: the Windows and SQL Server premium
Licensing is the line item most often missed in a cloud cost review. In Azure UAE North the same D2s v5 VM costs $0.118 an hour on Linux and $0.210 on Windows, so the licence adds 78 per cent to the compute price. SQL Server Standard or Enterprise on top multiplies it again.
If you hold Windows Server or SQL Server licences with active Software Assurance, Azure Hybrid Benefit lets you apply them in Azure; Microsoft quotes savings of up to 80 per cent on Windows Server when combined with reservations and up to 85 per cent on SQL Server against pay-as-you-go. Where there’s no Software Assurance, the options are to pay licence-included rates knowingly, move .NET applications to Linux, or move databases to a managed service where the licence model is simpler. Our comparison of AWS, Azure, Google Cloud and Oracle in the UAE covers why licensing often decides which provider a Microsoft-heavy company should use in the first place.
Sources: Azure Retail Prices API · Azure Hybrid Benefit
Estimate your own savings
Enter your monthly compute spend and a few honest estimates. The calculator applies the levers in the order this playbook recommends: schedule non-production, rightsize production, then commit to the steady share of what’s left.
Illustrative only. The calculator applies to compute; storage, data transfer, licences and support respond to different levers. Use the realised discount for your own instance types (38 per cent for a one-year Azure reservation in the example above), not the vendor’s maximum.
If the result is under a tenth of your compute spend, your estate is already in reasonable shape and the gains are in architecture rather than housekeeping.
A 90-day FinOps plan
| Weeks | What happens | Output |
|---|---|---|
| 1–2 | Read-only access; map accounts, tags and owners; split production from non-production; set budgets and anomaly alerts | A bill everyone can read, with a named owner per line |
| 3–4 | Delete the obvious: orphaned disks, snapshots, idle IPs, zombie servers; schedule non-production | First saving visible in the next invoice |
| 5–8 | Rightsize production one service at a time; move gp2 to gp3; test Graviton for compatible services; set log retention | A smaller, measured baseline |
| 9–10 | Buy commitments against the new floor, starting with flexible one-year products | Commitment coverage on steady usage |
| 11–13 | Agree a unit-cost metric; hand over a monthly 30-minute review; add cost checks to the release process | A routine that stops the bill drifting back |
Timings reflect OXEDGE delivery experience with growing UAE companies; the size and age of your estate will move them.
The last step is the one that decides whether the savings last. Cost has to become part of how engineers ship, in the same way tests and reviews are; our DevOps baseline for small teams shows where a cost check fits in a simple pipeline. The companies that keep bills down review them every month and ask one question of every new resource: who owns this, and when does it switch off?
Questions UAE buyers ask
How much more does cloud cost in the UAE than in the US or Europe?
On-demand list prices in the UAE regions run roughly a fifth above US East. An AWS m7i.large on Linux lists at $0.1236 an hour in me-central-1 against $0.1008 in US East, 23 per cent more. Moving to Europe saves only two or three per cent and costs you latency and an easy residency answer.
How much cloud spend is wasted?
Flexera’s 2026 survey of more than 750 cloud decision-makers found organisations estimate 29 per cent of cloud spend is wasted, the first increase in five years. The waste is rarely one big leak. It’s a dozen small ones nobody is watching.
What is the fastest way to cut a cloud bill without a commitment?
Rightsizing, scheduling and newer hardware. Drop oversized instances one size at a time, run development environments only during working hours, and move to newer or ARM-based instances. AWS says Graviton instances cost up to 20 per cent less than comparable x86 instances.
What are the benefits of using local cloud services in the UAE versus international ones?
The main benefits are lower latency and an easy answer to every data residency question. The price is a premium: UAE regions list roughly a fifth above US East, while moving workloads to Europe saves only two or three per cent.
Does AWS charge VAT on UAE cloud bills?
Yes. AWS has billed UAE accounts through branches in Abu Dhabi and Dubai since 1 February 2024 and charges 5 per cent VAT on all service charges, issuing tax invoices in the billing console.
What is FinOps?
FinOps is a practice with three repeating phases: Inform, Optimise and Operate. Most companies skip straight to optimising and watch the bill creep back within a quarter, so make the bill legible first with separate accounts, mandatory tags and one unit cost the business understands.
Work with OXEDGE
Find out what your cloud bill should be before you sign another commitment.
We would start with read-only access to your billing and usage data and a one-week review of where the money goes. You leave the first conversation with the three or four changes most likely to cut your bill, and an honest estimate of what each one is worth.
