Virtual machines priced by what you reserve.

Spark instances run on compute nodes Carpathian owns in Des Moines. You price a machine by its parts, vCPU, memory, and disk, instead of picking whichever plan tier happens to carry the one number you need. Compute is never metered, so what you reserve is what you pay for, whether the instance idles for a month or runs flat out.

carpathian.ai / dashboard / spark-servers

Instance

base configuration
spark-app-01RunningUbuntu 24.0410.42.8.14ConsoleResize
vCPU1 on the base instance
Memory1 GB on the base instance
Disk20 GB on the base instance
ScalingEach component on its own
1 vCPU / 1 GB / 20 GBroot accessnightly backup archive
Compute
Not metered
Billing
Prorated daily
Uptime SLA
99.5%
Minimum term
None
Priced by component
vCPU, memory, and disk carry their own rates, so you size each one on its own
Compute is not metered
No per-hour billing, no compute overage, and no CPU credits to run down
Root from first boot
SSH, RDP, or screen sharing on Linux or Windows Server, with no web terminal in the way
99.5% uptime SLA
Backed by uptime checks running against every host we operate

What every instance includes

The base instance is 1 vCPU, 1 GB of RAM, and 20 GB of disk. Everything above that is a component you chose to add, and the three scale independently, so a workload that wants memory does not oblige you to buy cores it will never use.

The list below is not a tier. It is what an instance carries the moment it boots, on the base configuration and on the largest one you can assemble, with none of it held back behind a higher plan.

  • Full root or administrator access over SSH, RDP, or VNC-based screen sharing
  • Ubuntu, Debian, CentOS, or Windows Server
  • A base network speed tier and data allowance
  • Uptime checks, SSL monitoring, and DDoS mitigation
  • Nightly backup archives and snapshots you can take yourself
  • Standard disk images, exportable, in no proprietary format
  • A 99.5% uptime SLA backed by monitoring on the host
  • Resizing that adds vCPU, memory, or disk without a rebuild
carpathian.ai / instance / base configuration
vCPU1 on the base instance
Memory1 GB on the base instance
Disk20 GB on the base instance
ScalingEach component on its own
Uptime checksOn every instance
SSL monitoringOn every instance
DDoS mitigationOn every instance
Backup archivesNightly
SnapshotsOn demand

Compute is never metered

The invoice follows what you reserved. Nothing on the compute side of it moves because of how hard the machine worked, and the one resource that does carry a usage limit answers with a speed change rather than a charge.

Priced per component

vCPU, memory, and disk are separate lines with separate rates, so you size each one against what the workload asks for. A database server that wants memory and a build runner that wants cores end up as two different machines rather than two rungs of the same ladder. The rate for each component is published on the pricing page, and the price of an instance is the sum of what you selected. Nothing is bundled in to round the number up.

Prorated to the day

Billing runs monthly and prorates to the day. Add disk partway through a month and you are charged for the days you had it rather than for the whole cycle you were in the middle of. There is no setup fee to clear before the first month and no minimum term holding the account open, so an instance that exists for three weeks is billed for three weeks.

The bandwidth default is a speed change

Every instance includes a base network speed tier and a data allowance. When you reach the allowance, the default is that speed falls back to the base tier until the next billing cycle and nothing further is charged, which means a traffic spike changes how fast the machine serves and leaves the invoice where it was. Overage billing exists for the case where full speed past the allowance matters more than a fixed bill, and it applies only if you opt into it. Beyond the base tier you choose a network plan for your organization, and that plan covers every server on the network rather than being charged per machine, so a second and third instance do not each arrive with their own bandwidth line.

One invoice

Compute, network, and the monitoring that comes with the instance arrive on one monthly invoice. There is no per-hour compute line to reconcile against a usage graph, no compute overage, and no credit balance that quietly empties during a busy week and throttles the machine you already paid for. What you agreed to reserve is what shows up.

From priced to connected

Four steps, and the first of them happens before you have an account.

carpathian.ai / dashboard / provision
PriceImageProvisionConnect
  • Price the machine
    Set vCPU, memory, and disk on the pricing page, which needs no account. The base instance is 1 vCPU, 1 GB of RAM, and 20 GB of disk, and every unit above that is a line you chose to add.
  • Pick an image
    Ubuntu, Debian, CentOS, or Windows Server. Whichever you take is a standard disk image, which is also the format you get back if you ever export it.
  • Provision
    The instance is created on a compute node in US Central 001 and comes up with an address, your image, and root or administrator access already in place.
  • Connect
    SSH for a shell, RDP for Windows Server, VNC-based screen sharing for a graphical desktop. You use the client you already have rather than a console we put between you and the machine.

Why the rates look the way they do

We own the compute nodes and we own the buildings they sit in. Much of the hardware is enterprise equipment bought after its first depreciation cycle and rebuilt to our own spec, which is a long way from paying list price for the same silicon in its first year.

That is a structural cost position rather than a promotion. It is not an introductory rate that resets at renewal, and it is not a discount funded by metering something else on the way past. The published component rates are what the dashboard charges, and they are on the pricing page rather than behind a quote form.

What we do not claim

Both of our facilities are in Des Moines, Iowa. US Central 001 carries production and US Central 002 is the disaster recovery site, running the same hardware and the same network with automated failover between them. Each building has its own power and cooling, so a failure in one does not travel to the other.

That is site redundancy, and we will not describe it as anything else. It is not geographic diversity across separate power grids or separate weather systems, because both buildings sit in the same metro. If your compliance obligations require regional separation, raise it in the first conversation and we will tell you plainly whether we can meet them today.

Bare metal is in development and is not available to provision, so everything on this page describes virtual machines running on shared compute nodes.

Questions we get before the first instance

What does the base instance include?
One vCPU, 1 GB of RAM, and 20 GB of disk. Those three scale independently, so you can add memory without buying cores you will never touch. The current rate for each component is published on the pricing page.
Is compute metered?
No. You pay for the resources you reserve, so an instance that runs flat out for a month costs the same as one that idles. There is no per-hour billing, no compute overage line, and no CPU credit balance to run down.
What happens when I pass the data allowance?
By default the instance falls back to its base speed tier until the next billing cycle and nothing further is charged. If full speed past the allowance matters more to you than a fixed invoice, overage billing is available as an opt-in. Nothing changes unless you choose it.
Can I resize an instance without rebuilding it?
Yes. vCPU, memory, and disk each change on their own, and none of them require rebuilding the machine. Billing is monthly and prorated to the day, so a change partway through the month is charged for the days it applied.
Which operating systems can I run, and how do I connect?
Ubuntu, Debian, CentOS, and Windows Server, each with full root or administrator access. You connect over SSH for a shell, RDP for Windows Server, or VNC-based screen sharing for a graphical desktop.
Where do the instances physically run?
On compute nodes Carpathian owns in Des Moines, Iowa. US Central 001 carries production and US Central 002 is the disaster recovery site, running the same hardware and the same network with automated failover between them.
Is that geographic redundancy?
No, and we will not call it that. Both buildings are in Des Moines, so what you get is site redundancy: independent power and cooling in each facility and a failover that does not need anyone awake. It is not diversity across separate grids or weather systems. If your compliance requirements call for regional separation, raise it early and we will tell you plainly whether we can meet them.
What if I want to move somewhere else?
Your instances use standard disk images that you can export and run elsewhere. There is no proprietary image format, no migration fee, and no notice period, and billing carries no minimum term or setup fee.

Price the exact machine you need.

Set vCPU, memory, and disk against the published component rates and watch the price move as you change each one. Billing is monthly and prorated to the day, with no setup fee and no minimum term. If you would rather work through the sizing with someone first, the person answering is one of the engineers who runs the hardware.