Four Ways to Run Infrastructure
Most teams treat infrastructure as cloud-or-not. The actual decision has four positions.
Most companies treat infrastructure as a binary choice: cloud or self-managed. In practice there are four distinct models, each with a different balance of cost, control, and operational burden. Where you should sit depends on your workload profile, your team, and how much of the stack you want to own.
What changes at each step
Cost drops ~50%. You lose scale-to-zero. You gain dedicated hardware, flat billing, and an SRE team you can reach directly. We handle the migration. Billing starts when your workloads are running, not when hardware ships.
Per-unit cost drops further. You lose managed Kubernetes, managed databases, and 24/7 on-call. You need to hire 1-2 infrastructure engineers.
Long-term cost drops at scale. You gain complete hardware control. You take on capital expenditure, hardware refresh cycles, and facilities management.
Side-by-side comparison
| Public Cloud | Lithus | Rented Bare Metal | Owned Hardware | |
|---|---|---|---|---|
| Best for | Variable or spiky workloads, scale-to-zero | Steady-state production, €5k+/month | Teams with existing DevOps skills | Large-scale, long-horizon deployments |
| Relative cost | ~2x bare-metal equivalent | ~50% of cloud for equivalent hardware | Cheapest per unit | Lowest at scale (3-5 year amortisation) |
| Billing | Per-resource, per-hour + egress fees, highly variable | Flat monthly rate, all-inclusive | Per-server monthly, predictable | CapEx + colo + power + network |
| You manage | Application, IAM, networking, cost optimisation, upgrade cycles | Application only | Everything from OS to application | Everything from hardware to application |
| Team required | Cloud engineer or DevOps hire at scale | None (SRE included) | 1-2 full-time infrastructure engineers | Full infrastructure + ops team |
| Support model | Ticket queue, tiered response | Engineers you can message directly | Community / self-serve | Self-managed |
| Hardware | Shared tenancy, variable performance | Dedicated bare metal, consistent throughput | Dedicated bare metal | Your own hardware, any configuration |
| Lock-in | High (proprietary services, egress fees) | Low (standard Kubernetes, open-source stack) | None | None (but hardware commitment) |
| Initial provisioning | Days to weeks (architecture design, infrastructure-as-code) | 4-6 weeks (hardware, cluster, observability) | Days to weeks | Weeks to months |
What actually changes
Cost Predictability
The biggest change is the billing model. Cloud costs are usage-based: adding a feature, running a migration, or handling a traffic spike can create a surprise on the next invoice. On managed private cloud, you pay a flat monthly rate. Hardware, Kubernetes, observability, and SRE time are all included. Your finance team can forecast infrastructure spend 12 months out.
Performance Without Variance
Our inter-node latency is around 0.2ms. Public cloud instance-to-instance is typically 1-3ms with jitter. For database replication, distributed caches, and service meshes, that difference compounds across every request. Dedicated NVMe storage gives you consistent low-latency IOPS without contention. More importantly, all of it is consistent. No noisy neighbours, no shared storage, no performance variance at 2 AM when a neighbouring VM runs a batch job.
Someone Carries the Pager
If you rent bare metal directly, you are the on-call engineer. On cloud, you get a support ticket queue. On managed private cloud, you get an SRE team in a shared Slack channel with a 2-hour incident response SLA. The operational burden is the same as cloud -- you manage your application, we manage everything underneath -- but the support is hands-on rather than ticket-based. For teams without a dedicated DevOps function, the question is not just cost but who actually operates the platform day to day.
No Lock-in
One Kubernetes cluster. Standard APIs. No proprietary service mesh, no cloud-specific IAM, no vendor SDK. Your Helm charts and CI/CD pipelines work here the same as anywhere. If you leave, you take your workloads with you.
When managed private cloud is the wrong choice
We are not the right fit for every workload. Here is where the other models genuinely win.
Scale-to-zero
Dedicated hardware runs 24/7. If your workload drops to zero for hours, needs serverless functions, or is genuinely spiky, public cloud is the right answer.
Under €5k/month
Our service starts at €5,000/month. Below that threshold it is difficult to guarantee the level of support and hands-on engineering we build every engagement around.
Windows-native workloads
Our managed stack is built on Linux and open-source tooling. We can provide Windows VMs, but not as a fully managed service with the same level of support.
Not sure which model fits?
Send us your current cloud bill. We will break down what you are paying for, estimate what the same workloads would cost on managed bare metal, and tell you honestly if the switch makes sense. If cloud is the right answer for your workload, we will say so.