Every six months a customer asks us "should I move off AWS?" The honest answer depends on what you're running, but the cost gap between hyperscaler VMs and well-chosen bare metal is bigger than most people realize. Here's the math, with the line items the cloud calculators conveniently omit.
The Headline Comparison
Take a single workload that needs roughly 32 vCPU, 128 GB RAM, 2 TB NVMe, and 50 TB/month outbound transfer. Approximate monthly cost in 2026:
| Path | Compute | Storage | Bandwidth | Total |
|---|---|---|---|---|
| Hyperscaler on-demand VM | ~$1,200 | ~$200 | ~$2,500 | ~$3,900/mo |
| Hyperscaler 1-yr reserved | ~$700 | ~$200 | ~$2,500 | ~$3,400/mo |
| Bare metal at Noded | ~$300 | included | included or low | ~$300/mo |
Numbers are illustrative. Cloud prices vary by region, instance family, and discount. Verify with current price calculators.
The compute delta is real but the bandwidth delta is what kills people. Hyperscalers charge $0.05–$0.09 per GB egress at scale. A bare-metal provider gives you a 1 Gbps unmetered port — at 50% utilization that's 162 TB/month for the same flat price.
What Cloud VMs Actually Cost You
- vCPU pricing — vCPUs are usually hyperthreads, not full cores. You pay per "vCPU" but get half a physical core.
- RAM is bundled — at fixed CPU:RAM ratios that may not match your workload.
- Block storage — billed per GB-month plus per-IOPS for higher tiers.
- Bandwidth egress — typically the largest line item for any user-facing workload.
- NAT gateway, load balancer, snapshots, logging, monitoring — each a line item.
- Cross-AZ traffic — yes, you pay to send traffic between two of your own zones.
The convenience cost is real. The platform is excellent. You're paying for the platform, not the metal.
What Bare Metal Actually Costs You
- Server rental — fixed monthly fee.
- Bandwidth — usually a generous unmetered port (1 Gbps to 10 Gbps) or a TB allowance.
- IPs — typically a /29 IPv4 and IPv6 are included; extra subnets are cheap.
- Setup fee — often waived on annual or in promotions.
- Your time — you manage the OS, patching, monitoring, backups, and HA yourself unless you pay for managed.
The trade is straightforward: you save money, you take on operational responsibility.
When Cloud VMs Are Worth It
- Spiky, unpredictable load. If you scale from 0 to 1000 instances and back in a day, on-demand pricing is the right shape.
- You use 30+ managed services. If you're deep in managed Kubernetes, managed databases, message queues, and serverless, the integration value is real.
- Compliance requirements that map to specific cloud features. Some regulated industries lean on hyperscaler attestations.
- Globally distributed dev teams who already know the platform and don't want to learn another stack.
- Short-lived workloads. Per-second billing wins when nothing runs for more than a few hours.
When Bare Metal Wins
- Steady-state workloads. If your fleet looks roughly the same Tuesday at 3am as it does at noon, bare metal is cheaper every month.
- Bandwidth-heavy applications. Video, downloads, image hosting, game servers, CDN origins — anywhere egress dominates.
- Storage-heavy workloads. Multi-TB databases, backups, ZFS pools.
- Performance-sensitive workloads. Databases that benefit from full-core CPUs without noisy-neighbor variance.
- You already have an ops team. The savings fund the team's salary several times over.
The Hybrid Pattern
Most experienced operators end up running both. Predictable, bandwidth-heavy, performance-sensitive workloads on bare metal. Spiky frontends, dev/test environments, and managed-service integrations on cloud VMs. The two connect over IPSec, WireGuard, or a private interconnect.
You don't have to pick one philosophy. You should pick what fits each workload.
How Noded Helps
Noded.cloud offers dedicated servers with generous unmetered bandwidth, VPS plans for smaller workloads, colocation for your own hardware, and IP transit for customers running their own ASN. We're not trying to be AWS — we're the layer underneath that makes the cost math work.
If you're trying to model a migration off cloud, talk to us. We'll be honest about what makes sense to move and what's better left where it is.
FAQ
Will moving off the cloud break my deployment?
If you've used cloud-specific managed services heavily, yes — you'll need to replace those. If your workload is mostly EC2 plus a database, the migration is usually straightforward.
Can I run Kubernetes on bare metal?
Yes — Talos, k3s, kubeadm, and managed bare-metal Kubernetes offerings all exist. The capacity planning is different from EKS/GKE/AKS but the API is the same.
Isn't the cloud more reliable?
Hyperscaler regions have well-publicized multi-hour outages too. Reliability is a function of how you design the system, not who owns the metal. A well-built two-region bare-metal setup beats a single-region cloud deployment most days.
What about disaster recovery?
You can replicate to another bare-metal site, to cloud, or both. Off-site backups and a documented runbook are what matter — not which logo runs the spare.
How do I size a bare-metal replacement for my cloud fleet?
Map your peak utilization, not your provisioned size. Most cloud workloads run at 20–40% utilization. Bare metal lets you run hot because you bought the whole machine.