The Cloud Bill Nobody Shows You
A gaming company was paying $30,000 a month to AWS. They moved to bare-metal Kubernetes hosted in a Tier 5 colocation facility. Their monthly infrastructure bill is now $6,000 — with enough physical capacity to run ten times their current workload at that same price.
There is a number on your AWS or Azure invoice every month. It is the number your finance team sees, the number that gets approved in budget reviews, and the number your cloud vendor wants you to focus on. It is not the real number.
The real cost of cloud infrastructure has three components. The invoice is only one of them. The other two — egress fees that scale with your own success, and the engineering hours consumed by complexity that exists to serve the vendor rather than your business — rarely appear in any procurement conversation. By the time they become visible they are already structural, already embedded in your architecture, and already very expensive to remove.
This is not an accident.
The elasticity you are paying for versus the capacity you could own
Cloud vendors sell elasticity. The pitch is compelling: pay only for what you use, scale instantly on demand, never over-provision. For an early stage company with unpredictable growth this is a genuinely reasonable trade.
For a company with stable, understood workloads it is one of the most expensive decisions on the balance sheet.
A gaming company running twelve full racks of bare-metal Kubernetes in SWITCH Data Centers in Nevada — one of the most premium colocation facilities in the world, Tier 5 rated with exceptional power redundancy and connectivity — pays approximately $6,000 per month for colocation. That facility has the physical capacity to host ten times their current workload at the same price. The headroom is already paid for.
Their previous AWS bill was $30,000 per month.
The difference is $24,000 every month. $288,000 every year. Not because they compromised on infrastructure quality — SWITCH is not a budget facility. Not because they reduced their capability — they have more headroom now than they ever had on AWS. Because they stopped renting capacity they could own.
Cloud elasticity is valuable when you need it. When you do not need it you are paying a permanent premium for an option you are not exercising.
The egress trap
Cloud vendors price ingress at zero. Data coming in costs nothing. This feels generous until you understand what it is: a mechanism to ensure that leaving costs as much as possible.
Egress — data leaving the cloud — is priced per gigabyte. The rates vary by provider and region but the structure is consistent: the more your applications serve, the more your users consume, the more successful your product becomes, the larger your egress bill grows. You are being charged for your own success by the infrastructure provider whose pitch was that they would help you scale.
For a gaming company with active users, real-time game state, asset delivery, and analytics pipelines, egress is not a marginal line item. It is a significant and growing fraction of the total infrastructure cost — one that scales directly with the metric every business wants to maximise: usage.
On bare-metal infrastructure the concept of egress fees does not exist. Your hardware connects to your network. Data leaving your servers costs the same whether it is one gigabyte or one petabyte — the bandwidth already provisioned in your colocation contract. The relationship between your growth and your infrastructure cost becomes linear and predictable rather than a variable that accelerates against you.
A 1Gbps dedicated connection included in a colocation contract delivers consistent, unmetered throughput. The same data volume through a cloud provider's egress pricing would appear as a line item on every invoice, growing every month the business grows.
Complexity as a business model
The third hidden cost is harder to quantify but more damaging over time.
Cloud platforms are not designed to be simple. They are designed to be comprehensive — a managed service for every problem, an integration for every tool, a certified partner for every implementation pattern. The complexity this generates is not a side effect. It is the product. Every managed service adopted is a dependency. Every dependency makes the next migration decision harder. Every hard migration decision is another year of invoices.
The engineers who manage this complexity are not building your product. They are managing the operational surface area of a vendor relationship. IAM policies, VPC configurations, service quotas, region-specific limitations, API deprecation cycles, cost allocation tags — none of this creates value for your users. All of it consumes engineering time that could be directed at the actual problem your business exists to solve.
Sun Microsystems understood something in the early days of enterprise computing that the cloud era has largely abandoned: security and operational integrity are not states you achieve and certify. They are continuous processes requiring engineers who understand every layer of the stack they operate. When your infrastructure is a managed abstraction your team cannot fully inspect or reason about, that continuous adaptation becomes impossible. You become dependent on your vendor's response to threats your vendor may not have disclosed to you yet.
The shared responsibility model — the framework every major cloud provider uses to define what they secure and what you secure — is an elegant document. It is also a liability transfer mechanism. The responsibilities it assigns to you require deep expertise in a platform you do not own, running on infrastructure you cannot inspect, with a threat surface that changes on the vendor's schedule rather than yours.
Bare metal returns that responsibility to where the expertise actually lives — with engineers who built and operate the stack, who understand every configuration decision, and who are not dependent on a vendor's disclosure timeline to know what is running in their environment.
What the numbers actually say
The gaming company comparison is not an outlier carefully selected to make a point. It is the predictable outcome of a straightforward analysis applied to a stable workload.
$30,000 per month in cloud infrastructure. $6,000 per month in premium Tier 5 colocation. Ten times the physical capacity headroom at the lower price. Zero egress fees. Complete operational ownership of every layer of the stack.
The cloud invoice was not paying for better infrastructure. It was not paying for greater reliability. It was not paying for capabilities unavailable on bare metal. It was paying for the convenience of not having made a different decision sooner — and for the egress fees, the managed service dependencies, and the operational complexity that accumulated quietly in the years that followed the original decision.
For organisations with stable workloads, predictable growth, and engineering teams capable of operating their own infrastructure, the economics of cloud repatriation are not ambiguous. The question is not whether the numbers work. The question is how long the current arrangement has been running and how much the delay has cost.
Is repatriation right for your organisation
Repatriation is not the correct answer for every situation. Early stage companies with genuinely unpredictable scaling requirements, organisations without the engineering maturity to operate bare metal, and workloads with extreme geographic distribution requirements all have legitimate reasons to remain in cloud.
But for organisations that have been running stable workloads in cloud for more than two or three years, that have engineering teams capable of operating Linux infrastructure, and that have a workload profile they can size hardware against — the break-even point on bare metal hardware investment typically arrives within the first year. Everything after that is savings.
The $24,000 monthly delta in the gaming company example represents $288,000 per year in infrastructure cost that is no longer leaving the business. At that rate, enterprise-grade bare metal hardware pays for itself before the second annual renewal of the cloud contract it replaces.
The conversation worth having is not whether cloud or bare metal is philosophically superior. It is whether the premium your organisation is currently paying for cloud infrastructure is purchasing something of equivalent value — or whether it is paying for a decision that made sense years ago and has not been revisited since.
Sugau Infrastructure specialises in bare-metal Kubernetes deployments, cloud repatriation, and sovereign AI infrastructure for enterprises that require operational ownership of their compute stack.