Cloud repatriation is the decision to move a workload off public cloud and back onto infrastructure you control, either on-premises or on dedicated hardware in a colocation facility. It is rarely a full cloud exit. In practice it means pulling the steady, always-on workloads (the VPN, the database, the VoIP platform) onto flat-rate hardware, and leaving the genuinely spiky ones exactly where they are. The reason is arithmetic, not ideology. Metered pricing was built to reward variability, and a workload that never varies pays the ceiling rate every month, indefinitely, on hardware it will never own.
- 86% of CIOs plan to move at least some public cloud workloads back to private or on-premises infrastructure, the highest figure Barclays has ever recorded. Only 8% to 9% plan a full exit, per IDC. Repatriation is selective, not a stampede.
- Metered cloud pricing rewards variability. A steady baseline workload never earns the discount and pays the on-demand ceiling every month.
- Egress is the line item nobody budgets. The hyperscalers dropped exit egress fees in 2024, but the everyday operational egress you actually pay is untouched.
- A dedicated server is a fixed cost. Hardware, power, cooling and bandwidth are priced once, as one flat rate. Traffic doubling in a good month does not double the bill.
- For most businesses the right answer is hybrid: dedicated for the steady core, cloud for the spiky edges.
What is cloud repatriation?
Cloud repatriation, sometimes called reverse cloud migration, is the practice of moving applications, compute or data out of public cloud and back onto infrastructure the business controls directly. That destination is usually one of two things: hardware in the company’s own facility, or dedicated servers racked in a carrier-neutral colocation data centre.
It is not an anti-cloud position, and the numbers bear that out. In its Q4 2024 CIO survey, Barclays found 86% of CIOs planned to repatriate at least some public cloud workloads, the highest rate the survey has recorded. IDC puts the share planning a full-scale exit at only 8% to 9%.
Those two figures are not in conflict, and it is worth understanding why before anyone quotes the big one at you. Barclays counts companies, not workloads. A business that moves one application back counts exactly the same as one that moves everything, which is why analysts have pushed back on the headline percentage being read as a mass cloud exodus. It is not an exodus. It is selective, cost-driven redistribution, and that is a far more useful thing to know, because it tells you the real question is not whether to leave the cloud. It is which workloads should never have been there.
The cloud pitch versus the cloud invoice
The cloud pitch was elastic scale: pay only for what you use, spin up in seconds, never think about hardware again. For bursty, unpredictable workloads, that pitch holds up. But most production workloads are not bursty. They are steady. A VPN, a database, a VoIP platform, a SaaS backend running the same baseline load every hour of every day, all year, is not what the cloud was built to price efficiently.
Run a steady workload on metered infrastructure long enough and the bill stops looking like flexibility and starts looking like rent you can never stop paying, on hardware you will never own. Compute, storage, and above all egress, the fee to move your own data back out, compound month over month while the workload itself has not changed at all.
The cloud doesn’t charge you for the compute you need. It charges you for the compute you can’t predict.
Cloud vs dedicated server: the battle card
| Dimension | Public cloud | Amanah dedicated |
| Monthly cost | Metered by the hour, unpredictable, climbs with usage | One flat rate, all-inclusive, no surprise fees |
| Data egress | Billed to move your own data out, per gigabyte | Unmetered bandwidth, 100Gbps, no per-GB toll |
| Performance | Shared, virtualized, noisy-neighbour risk | Full bare-metal resources, dedicated to you alone |
| Control | Managed abstractions; custom OS images often restricted | Root access, your OS, your hypervisor, your image |
| Contract terms | Discounts tied to multi-year reserved commitments | Month-to-month available, no long-term lock-in |
| Data residency | Often routed through or hosted in US regions | Canadian soil, Toronto facility, no cross-border hop |
| Support | Ticket queues, tiered support plans as add-ons | Local Toronto team, hands, feet and eyes on-site |
Where the public cloud actually wins
To be fair to the cloud: it earns its price for genuinely elastic, spiky or short-lived workloads. A batch job that runs for an hour a week, a dev environment torn down nightly, a traffic spike you cannot forecast, that is the cloud’s home turf, and dedicated hardware sitting idle waiting for it would be a worse deal. If your workload actually behaves like that, keep it there.
But most infrastructure is not that. A production VPN does not scale to zero at 3am. A customer database does not take weekends off. A VoIP platform runs at roughly the same baseline load, every day, indefinitely, and that is precisely the profile the cloud prices the least efficiently.
What does a steady workload actually cost?
Run the comparison on a steady, always-on workload and three structural facts show up every time. This is where most cloud cost optimization work quietly stalls: you can rightsize an instance, but you cannot rightsize a pricing model that assumes variability you do not have.
- Cloud pricing is built around variability, and steady workloads do not get the discount. On-demand rates assume you will sometimes use less. If you never do, you pay the ceiling rate every month, indefinitely.
- Egress is the fee nobody budgets for. Every gigabyte your application serves to users, replicates to backup, or moves between regions gets billed on the way out.
- A dedicated server is a fixed cost you can plan a budget around. The hardware, power, cooling and bandwidth are priced once, as a flat rate. Traffic doubling in a good month does not double your bill.
The break-even point varies by workload, but the direction does not: the longer a workload runs at a steady baseline, the more a flat-rate dedicated server outperforms a metered cloud instance on total cost, and the gap widens every month you stay.
- 100Gbps unmetered bandwidth, no egress fees
- We’ll beat any comparable colocation quote by 10%
- 400+ carrier-neutral networks at 151 Front
Are cloud egress fees actually going away?
Partly, and not in the way that helps your monthly bill. In 2024 the hyperscalers waived egress charges for customers leaving: Google moved first in January, AWS followed in March, and Microsoft matched within a week. The trigger was regulatory, not competitive. The EU Data Act will prohibit providers from charging data egress or switching fees outright from January 2027.
Read the conditions, though. These waivers cover the exit. Azure’s requires a complete departure, including cancellation of all subscriptions. AWS excludes services such as CloudFront and Direct Connect. And crucially, none of it touches the egress you pay in normal operation. Serving data to your own users, replicating to an off-site backup, moving objects between regions: all still billed per gigabyte, every month, exactly as before. The EU Data Act changes made the door cheaper to walk through. They did not make the room cheaper to live in.
On dedicated infrastructure the line item does not exist to begin with. Amanah’s unmetered 100Gbps bandwidth means outbound traffic is not metered, tiered or tolled. There is no per-gigabyte calculation to forecast, because there is no per-gigabyte charge.
What you get back when you own the hardware
Repatriating is not just about the invoice. It is about what a dedicated bare-metal server gives you that a shared instance structurally cannot.
- Full, dedicated resources. No noisy neighbours competing for the same CPU or disk I/O. What you provision is what you get, every time.
- Bring your own OS and hypervisor. Install a custom image, run KVM, or configure it exactly to spec, with no platform restrictions in the way.
- A bill you can actually forecast. One flat, all-inclusive rate. No surprise invoices when traffic has a good month.
- Canadian data on Canadian soil. Hosted at 151 Front in Toronto, with no cross-border routing required to keep your data at home.
Does Canadian data sovereignty change the math?
For regulated workloads, it often decides it. Canadian data sovereignty is the principle that data collected in Canada stays subject to Canadian law, which in the private sector means PIPEDA. PIPEDA does not ban cross-border processing outright, but it does hold you accountable for your data wherever it goes, and it requires you to be transparent about that. If you cannot say with certainty which region a managed service replicated to overnight, that is a hard sentence to write into a privacy policy.
Dedicated hardware in a Toronto facility removes the ambiguity: you know the rack, the building and the jurisdiction. Amanah’s hardware sits at 151 Front Street, the carrier hotel that anchors Canadian interconnection, with 400+ carrier-neutral networks and a direct presence on TorIX, the Toronto Internet Exchange. Domestic traffic stays domestic, which is a compliance answer and a latency answer at the same time.
Cloud repatriation doesn’t mean ripping everything out
Moving off the cloud is not an all-or-nothing decision, and we do not pitch it that way. The workloads worth repatriating first are the ones running steady, predictable load: the VPN, the database, the always-on backend. Genuinely bursty or short-lived jobs can stay exactly where they are.
Most of the businesses we work with end up running a hybrid: dedicated hardware for the steady core, cloud for the spiky edges. Qublix Games cut its AWS costs by moving to colocation in Toronto without abandoning the cloud services that were genuinely earning their keep. If the two halves need to talk privately rather than over the public internet, Cloud Connect provides direct private links from your rack into AWS, Azure, Google Cloud and Cloudflare.
How Amanah supports the move
- Free migration consultation. We will look at your current cloud bill and tell you honestly whether repatriation makes sense before you commit to anything.
- Beat-any-quote pricing. We will beat any comparable colocation quote by 10%, with no surprise fees hiding in the fine print.
- We will receive and rack your hardware. Ship it and we will handle the rest. Our receiving service gets your gear installed without you on-site.
- No locked-in contracts. Month-to-month terms available. If dedicated does not work out for you, you are not stuck proving it for three years first.
Bring your cloud bill. We’ll show you the other number.
A free consultation and data centre tour, no commitment required. See what your steady workload actually costs on dedicated hardware.
Questions about this topic
No, and we will not tell you it is. For bursty or short-lived workloads, the cloud’s metered pricing genuinely fits better. For a steady, always-on workload, a flat-rate dedicated server almost always wins on total cost, especially once egress fees are counted honestly.
Egress fees are what cloud providers charge to move data out of their network, whether to your users, to another provider, or to a backup location. They are billed per gigabyte and rarely appear in the sales pitch. Amanah’s dedicated servers include unmetered 100Gbps bandwidth, so the line item disappears.
Only for leaving. Google, AWS and Microsoft all waived exit egress charges in 2024 ahead of the EU Data Act, which bans egress and switching fees from January 2027. Those waivers come with conditions, and none of them touch the operational egress you pay every month serving data to your own users.
No. Most businesses run a hybrid: steady, predictable workloads on dedicated hardware, and genuinely elastic or short-lived jobs left on the cloud. We will help you work out which of your workloads belong where.
You take on responsibility for the OS and stack yourself, which is exactly why some teams prefer managed cloud services. In exchange you get full root control, no platform restrictions on the OS or hypervisor you run, and no noisy-neighbour performance variance.
It does if the hardware is in Canada. Amanah’s dedicated servers and colocation sit at 151 Front Street in Toronto, so your data stays on Canadian soil under Canadian law, with no cross-border routing required. That makes PIPEDA accountability far easier to document than a managed service that may replicate to a US region.
Bring us your current cloud bill. We will walk through it with you, map it against a dedicated setup at 151 Front, and give you the real numbers, plus a free tour of the facility, before you commit to anything.