Repatriation posts usually read like sales letters. This one includes the list of things you lose, because that list is the part that decides whether the move is worth making.
We are not going to pretend that a cloud with three regions replaces one with thirty. What we will argue is that most of the workloads paying hyperscaler prices in this region are not using the thirty regions, and a fair number are not using much beyond virtual machines, block storage and a load balancer.
The Three Reasons People Actually Leave
In our experience it is almost never “cloud is too expensive” in the abstract. It is one of three specific things.
The bill stopped being predictable. Not high — unpredictable. Data transfer, NAT gateway hours, inter-availability-zone traffic and a support plan charged as a percentage of everything else produce an invoice that nobody in the business can forecast or explain.
The latency is wrong for the users. An application serving customers in the Gulf from Frankfurt or Ireland carries 100 ms or more of round trip it does not need. For an API behind a single page load that is invisible. For anything chatty, it is the whole user experience.
The support relationship is a queue. Perfectly adequate until the day it is not, and on that day the escalation path matters more than the price per hour.
If none of those three describes you, read no further — the move will cost you more in engineering time than it returns.
Start With Your Own Bill, Not a Calculator
Before comparing any UAE cloud as an AWS alternative, open last month’s invoice and find four numbers: data transfer out, NAT gateway, inter-AZ traffic, and the support plan. In the accounts we have looked at, those four together are routinely a quarter to a third of the total, and they are the ones that do not appear in anybody’s pricing calculator.
Most AWS accounts get 100 GB a month of free data transfer out, and everything past it is metered. For comparison, each of our cloud instances includes 500 GB of egress and the first public IP, and on a Virtual Data Centre the traffic allowance sits inside the flat monthly price — 2 TB on the $499 Starter pool, up to 40 TB on the largest. There is no inter-availability-zone charge because there is no inter-availability-zone billing construct.
That is the comparison worth building a spreadsheet around. Compute-for-compute, hyperscalers are often competitive. It is the traffic and the plumbing that decide it.
What AWS Will Pay For on the Way Out
This is the part nobody mentions, on either side of the argument: AWS waives data transfer out charges for customers moving off AWS. You contact AWS Support and ask for free DTO rates under the programme, the request is reviewed at account level, and once approved it comes back as credits for the data being migrated. You do not have to close the account or change your relationship with AWS to qualify.
The condition worth planning around: per AWS’s September 2025 update, approved customers have 90 days to complete the move, with extensions requiring you to notify Support. So the sequence matters — scope the migration first, then ask, then run it inside the window. Ask early and you burn a third of your allowance on planning.
Which means the exit bandwidth bill, the thing most people assume is the trap, usually is not. The real cost is engineering time, and that lands almost entirely on the services that are not virtual machines.
What the Migration Actually Moves
Our migration service is built on Coriolis, and the mechanics are worth understanding because they determine your downtime. It reads your environment through the AWS API — no agent inside the guest, no guest credentials. It copies disks while the source VM keeps running, applies OS morphing (VirtIO drivers for KVM, cloud-init for OpenStack initialisation, network interface configuration), and at cutover runs a final delta sync and boots the machine on our side. The outage is the delta sync, not the copy.
What it does not move is anything that is not a VM. RDS, DynamoDB, Lambda, SQS, managed Kafka, the glue in between — those are rebuilds or replacements, not migrations, and that is where a project’s timeline actually goes. Count them before you commit to a date.
The Honest List of What You Give Up
- The managed-service catalogue. PostgreSQL on a VM with backups you configured yourself is not RDS, and saying otherwise does nobody any favours. You are taking back operational work. Sometimes that is the point; sometimes it is a hidden cost.
- Thirty-odd regions and a global edge network. We have three sites. If you genuinely serve users on four continents, this is a bad trade.
- Elasticity economics. Spot instances, scale-to-zero, Lambda for genuinely spiky load. If your traffic is 10× at 9am and idle overnight, the hyperscaler model is worth what it costs.
- The ecosystem. Battle-tested Terraform modules, Stack Overflow answers for every error string, marketplace appliances, a hiring pool that already knows the platform. OpenStack covers more of this than people expect — there is a mature Terraform provider, and object storage is S3-compatible — but it is not parity.
If reading that list made the move feel obviously wrong, that is a useful outcome and it cost you four minutes.
What You Get Back
A bill you can forecast, because the pool price is the pool price. Latency measured from inside the country rather than to it, from Dubai DX1 or Fujairah FJ1 — or Vienna VIE2 if the requirement is GDPR rather than UAE residency. Data that sits where a PDPL conversation needs it to sit, with a data processing agreement that names the facility.
And a network that is ours. We operate AS200851 rather than reselling capacity, which means a routing problem is a conversation with someone who can change the routing rather than a ticket forwarded upstream.
How to Decide in an Afternoon
Three questions, answerable from your own console:
- What percentage of the bill is data transfer, NAT and inter-AZ? Under 10% and the economics are thin.
- How many managed services deep are you? Count the ones that are not EC2, EBS or ELB. That count is your project plan.
- Is the load steady or spiky? Steady load on reserved capacity is where this wins. Spiky load is where it does not.
If the answers point your way, send us the service list and last month’s invoice and we will map it to real numbers — including the services we cannot replace and the ones where we think you should stay put.
