How RamNode Replaced Nearly 40 Leased Servers With Eight OpenMetal Bare Metal Servers

RamNode runs a public cloud that sells VPS, VDS, and virtual infrastructure to developers and businesses. In its Netherlands region, the hardware was getting old. The region ran on nearly 40 cloud nodes leased from a long-time partner, many of them eight to ten years old. Each node carried about four customers on the common 8 GB VDS plan. Adding capacity meant adding physical servers, which raised cost without a clear performance gain.

The team looked at the obvious fixes. They considered colocating new servers with another provider in Amsterdam. They returned some of the leased equipment, and the savings were marginal. Neither path solved the underlying problem, which was a cloud that could only grow one aging server at a time.

RamNode ended up on eight OpenMetal bare metal servers, plus one custom spinning-disk server for specialized workloads. This article walks through what they built, what the results look like, and how to run the same math for your own fleet. It is written for infrastructure leads at hosting providers, platform teams, and SaaS companies that run their own virtualization and are weighing leased servers, colocation, and hosted bare metal.

Why Aging Leased Servers Get Expensive to Keep

RamNode gave four reasons for moving. Replacement hardware from the leasing partner was outdated and lower performing. Cost pressure stayed high, and returning some equipment reduced it only marginally. Density had a ceiling, so growth meant more physical servers. And RamNode wanted to modernize without rebuilding its cloud stack.

That last reason shaped the whole project. RamNode wanted new hardware and a lower cost base, and it wanted its orchestration, APIs, provisioning, and billing logic to carry over.

Three Ways Out: Lease Again, Colocate, or Move to Hosted Bare Metal

Lease Newer Servers From the Same Partner

This is the lowest-effort option, and sometimes the right one. If your leasing partner can supply current-generation hardware at a good rate and your density is fine, staying put avoids a migration. RamNode’s experience shows the risk. Replacement hardware came in older and slower than they wanted, and returning part of the fleet barely moved the cost.

Colocate Servers You Buy

Colocation gives you full control over the exact hardware you run, and it suits teams that already own equipment they want to keep amortizing. It also brings capital purchases, facility contracts, and the work of running physical infrastructure: racking, cabling, spares, and break-fix. For a team that wants to focus on its platform, that is a lot of non-platform work. Our colocation alternative page covers the comparison in more detail.

Move to Hosted Bare Metal

Hosted bare metal gives you dedicated servers that only you use, and the data center work belongs to us. RamNode’s draw was newer hardware without colocation fees, long-term facility contracts, or the burden of running physical infrastructure, with predictable monthly costs. You keep control of the software layer. You give up the ability to pick any server on the market, since you choose from the configurations we offer.

What RamNode Built

RamNode’s hardware is a mix of OpenMetal enterprise bare metal servers, both Large v4 and Large v3 configurations. Each server has dual Intel Xeon Gold processors with 32 cores and 64 threads in total, 512 GB of DDR5 memory, two 6.4 TB NVMe data drives, and two 960 GB boot drives.

Three decisions stand out.

• Storage consolidation. Storage services moved onto two controller nodes. In other RamNode regions, dedicated storage nodes handle that role.
• Network adaptation. RamNode adapted its network configuration to OpenMetal’s environment.
• Phased migration. RamNode moved in phases to avoid customer disruption.

What stayed the same is as important as what changed. RamNode’s automation pipelines, APIs, provisioning, billing logic, and deployment scripts carried over, along with its software-level networking. In RamNode’s words, the cloud stack did not need to change. If you run your own platform, that is the property to test first, because it decides whether a move is a hardware project or a rebuild.

Vanessa Vasile, RamNode’s Director of Infrastructure, said, “Moving our Netherlands region onto OpenMetal completely changed how we operate.”

The Numbers and What They Mean

Cost

RamNode’s press release reports about 70% lower monthly infrastructure cost than RamNode’s prior operating range, and its cost also came in nearly 80% below its original peak spend. Read the figures as RamNode’s own reported results against its own history.

Density

Customers per node on the 8 GB VDS plan rose from about 4 to about 64, which RamNode describes as a 16x increase. That is a per-node figure for one plan.

Performance

RamNode reports improved I/O throughput and consistent performance on NVMe Gen4 storage and modern Xeon processors. If performance is the deciding factor for you, plan to measure it on your own workload during the first phase of a migration.

How to Run This Math for Your Own Fleet

You can test a consolidation case with a few inputs from your own environment. The example below is hypothetical. It uses round numbers to show the method and is not RamNode’s data.

1. Count what you carry today. If 40 servers each carry 4 customers on a given plan, the fleet carries 160 customers on that plan.
2. Estimate the density you can reach on new hardware. At 64 customers per server, 160 customers need 3 servers (160 divided by 64 is 2.5, rounded up).
3. Add headroom. Hosting providers need room for growth, failover, and uneven load, so size above the bare minimum.
4. Compare monthly cost per customer. Divide total monthly server cost by customers carried, before and after. Use your own current lease cost and our published bare metal pricing for the new side.

Check what limits density for your plan. A plan sized at 8 GB multiplied by 64 customers adds up to 512 GB, the same as the memory in each of RamNode’s servers. How much of the density gain comes from the hardware, and how much from how a provider allocates memory, CPU, and I/O across customers, varies from one provider to the next. Work out your own allocation policy before you borrow someone else’s ratio.

Then check the model against real results. A phased order lets you measure how your first workloads behave on the new servers and adjust the later orders to match what the sizing math predicted.

How a Bare Metal Migration Gets Planned

We do not keep a large pool of idle servers waiting for orders, so a migration of this kind is planned with you. We work through the target configuration, the order, delivery against your migration dates, and the setup of the servers. We set timelines for your specific order during planning and do not publish a single figure. A phased migration, like RamNode’s, fits this model well, because the first servers can carry the first workloads while later ones are planned to arrive when they are needed.

The planning conversation covers the same questions you would ask for any large order: which workloads move first, what the network needs to look like, what the cutover plan is, and how much headroom to buy. Customers who run their own virtualization on bare metal choose their own software. Our write-up on running Proxmox VE on OpenMetal bare metal is one example of that kind of stack.

Where Bare Metal Fits Next to Hosted Private Cloud

Bare metal and hosted private cloud solve different problems.

• Bare metal: dedicated servers with your own software on top. It fits teams with a platform, hypervisor, or orchestration they already run and want to keep, as RamNode did.
• Hosted private cloud: OpenStack and Ceph on dedicated servers, with a GUI and API through OpenMetal Central. It fits teams that want a cloud platform delivered, with Assisted Management or Complete Management as options.

The two can sit side by side. Bare metal servers can share VLANs with a private cloud, which our integrated bare metal page describes. A team might run a platform on bare metal today and add a private cloud later, or the reverse.

Two caveats apply. Bare metal gives you hardware, and your team owns everything above it, including the hypervisor, patching, and monitoring. And a hosting-provider case like RamNode’s depends on having a platform that already works. If you do not, hosted private cloud is usually the shorter path.

Where Other Options Win

Colocation wins when you own hardware you want to keep using, need specific equipment that is not in our catalog, or want to control every part of the physical stack. Leasing from your current partner wins when it can supply current-generation hardware at a rate that works and your density is already where you want it. Public cloud wins when demand is spiky or short-lived, since you pay by the hour for capacity you use for a week and return. Bare metal rewards steady, forecastable load on a platform you control.

Who This Fits

• Hosting providers and platform teams running their own virtualization, with orchestration and billing they want to keep.
• Teams with aging leased servers where density and cost are both under pressure.
• Buyers who want newer hardware without colocation contracts or physical operations work.
• Teams that can plan a phased migration against a date.

Who This Does Not Fit

• Teams with owned hardware they want to keep amortizing.
• Projects that need specific equipment outside our catalog.
• Workloads with short, unpredictable bursts.
• Teams without a platform to run on top of the servers.

RamNode’s move shows that a hosting provider can modernize hardware and cut cost without rebuilding its stack, provided the density math and the migration plan are tested before the order is placed.

Test the Math on Your Own Fleet

If your leased or colocated fleet is aging, start with the sizing steps above and bring the results to a planning conversation. Contact our team to scope a configuration and a phased order against your migration dates, or review bare metal pricing to fill in the cost side of your model.

Frequently Asked Questions

How many servers did RamNode replace, and with what?

RamNode’s Netherlands region ran on nearly 40 leased servers, many of them eight to ten years old. It now runs on eight OpenMetal bare metal servers, plus one custom spinning-disk server for specialized workloads, deployed in Amsterdam.

What cost reduction did RamNode report?

RamNode experienced about 70% lower monthly infrastructure costs than its prior operating range.

Did RamNode have to rebuild its platform?

RamNode reports that its automation, APIs, provisioning, billing logic, and deployment scripts carried over. Storage services were consolidated onto two controller nodes, and the network configuration was adapted to OpenMetal’s environment.

Does the 16x density figure apply to my workload?

It applies to RamNode’s 8 GB VDS plan, per node. Your density depends on your plan sizes, how you allocate memory, CPU, and I/O, and the hardware you choose. Model your own numbers and check them against the first phase of your migration.

How long does a bare metal migration take?

It depends on the size of the order and the migration plan, so we set the timeline during planning and do not publish a single figure. RamNode migrated in phases.

Should I choose bare metal or hosted private cloud?

Choose bare metal if you already run a platform or hypervisor you want to keep. Choose hosted private cloud if you want OpenStack and Ceph delivered with a GUI and API. The two can share VLANs, so you can combine them as your needs change.

Can I start with a few servers before ordering the full fleet?

Yes. A phased order lets the first servers carry the first workloads while later ones are planned to arrive when you need them. We set the phasing with you during planning.