Resource Pricing

How each resource is priced and billed, and where to find current rates.

Hyperstack bills virtual machines, Kubernetes clusters, storage, and public IP addresses by the hour. This page explains which resources are billed, how a virtual machine's hourly cost is calculated from them, and where to find the current rate for each one. Rates change over time, so this page does not reproduce them. For current prices, see Hyperstack pricing, or retrieve the rates billed to your own account from the Pricebook API.

How billing works

Resources are billed for every hour they exist, not for the hours you actively use them. A virtual machine that is running or stopped still holds its hardware, so it continues to be billed. Deleting a resource stops its billing entirely. Hibernating a virtual machine stops billing for its GPUs, but not for the data it keeps.

Two prices apply to your account, and they are not always the same number:

• List price: the public on-demand rate for a resource, published on the Hyperstack pricing page.
• Your rate: the rate your account is billed. It matches the list price unless your organization holds a contract that discounts a resource, or a promotional rate applies for a limited period. Retrieve it from the Pricebook API.

Where to find current prices

Current list prices for every GPU, along with storage and public IP rates, are published on the Hyperstack pricing page. The Object Storage page lists the object storage rate.

Both deployment pages in Hyperstack price your configuration before you deploy it. Navigate to Deploy a new Virtual Machine for a virtual machine, or Deploy a New Cluster for a Kubernetes cluster. Each page presents the same flavor cards, and selecting a configuration shows its hourly cost. A Running Cost total updates as you change your selection.

A cluster totals that cost differently. It is billed for its worker nodes at their flavor rate, plus its public IP addresses, while master, load balancer, and bastion nodes are not currently billed. See Cluster Billing Logic.

To retrieve the rates billed to your own account, call the Pricebook API.

What a GPU virtual machine costs

On a GPU flavor, the CPU cores, RAM, and local storage allocated to the virtual machine are included at no additional cost. Local storage covers both the root disk and the ephemeral disk, however large the flavor's allocation is. Only the GPUs are billed.

A GPU flavor's hourly cost is therefore its per-GPU rate multiplied by the number of GPUs in the flavor.

Example: A flavor with eight GPUs is billed at eight times the per-GPU rate. Doubling the GPU count doubles the hourly cost.

Resources attached to a virtual machine are billed separately from its flavor. A public IP address and any attached Shared Storage Volumes each add their own hourly charge.

Spot pricing

Spot VMs lower your hourly rate by running GPU workloads on Hyperstack's surplus compute capacity. Each spot GPU is priced as its own resource, at a discount to its on-demand equivalent.

Hyperstack can reclaim surplus capacity when it is needed elsewhere, and spot flavors do not support every virtual machine feature. Read Spot VMs before choosing one for a workload.

CPU-only flavors

A CPU-only flavor has no GPU, so its CPU cores, RAM, and local storage are each billed rather than included. Each is priced by its own resource, named with a (cpu-only-flavors) suffix.

To calculate the hourly cost of a CPU-only flavor, multiply each allocation by its matching rate, then add the results together. Multiply the flavor's CPU cores by the per-core rate, and its RAM in GB by the per-GB RAM rate. The local storage rate applies to the flavor's root disk and its ephemeral disk combined, so add both before multiplying.

note

A flavor's root disk size can differ between regions, which changes its hourly cost. Check the allocation for the region you deploy into.

Storage, public IP, and object storage

These resources are billed independently of a virtual machine's flavor:

• Cloud-SSD is the persistent storage behind Shared Storage Volumes, billed per GB per hour. Billing applies while a volume is AVAILABLE, IN-USE, or RESERVED; see Volume States and Billing.
• PublicIP is billed per hour for each public IP address.
• ObjectStorage is billed per GB per hour on the storage used by your buckets.

Snapshots are billed at the Cloud-SSD rate, on the storage they actually consume.

Hibernation and billing

When a virtual machine is HIBERNATED, its hardware is deallocated and its GPUs stop being billed. The data on its root disk is transferred to Cloud-SSD persistent storage, and is billed at the Cloud-SSD rate for as long as the virtual machine stays hibernated. This applies both to GPU flavors, whose local storage is otherwise free, and to CPU-only flavors.

Only the root disk is stored. Data on the ephemeral disk is permanently lost when a virtual machine is hibernated, so it is not billed.

A public IP address retained through hibernation also continues to be billed. For the billing status of every virtual machine state, see VM States and Billing.

Using the Pricebook API

The Pricebook API returns the hourly rate your account is billed for each resource, which is the authoritative figure for your organization. Retrieve it with your API key:

Retrieve your account's pricebook

curl -X GET "https://infrahub-api.nexgencloud.com/v1/pricebook" \
  -H "accept: application/json" \
  -H "api_key: YOUR_API_KEY"

The response is a bare JSON array, with one object per resource rather than a status and message envelope.

Resources in the response

Resource name	What it covers	Billed by
<GPU name>	One entry per GPU, named as on the flavors page. Spot GPUs carry a -spot suffix.	Per GPU, per hour
vCPU	CPU cores on GPU flavors. Included at no cost.	Per CPU core, per hour
RAM	Memory on GPU flavors. Included at no cost.	Per GB, per hour
hypervisor-local-storage	Root disk and ephemeral disk on GPU flavors. Included at no cost.	Per GB, per hour
vCPU (cpu-only-flavors)	CPU cores on CPU-only flavors.	Per CPU core, per hour
RAM (cpu-only-flavors)	Memory on CPU-only flavors.	Per GB, per hour
hypervisor-local-storage (cpu-only-flavors)	Root disk and ephemeral disk on CPU-only flavors.	Per GB, per hour
Cloud-SSD	Persistent storage behind Shared Storage Volumes.	Per GB, per hour
PublicIP	A public IP address.	Per IP, per hour
ObjectStorage	Data held in object storage buckets.	Per GB, per hour

The response also covers GPUs that are not currently available to deploy, and the AI Studio models described below. A GPU appearing in the pricebook does not mean you can deploy it; the Deploy a new Virtual Machine page lists what is actually available.

Response fields

Field	Description
id	Unique numeric identifier for the resource.
name	The resource name, matching the names in the table above.
value	The hourly rate your account is billed for the resource.
original_value	The hourly rate before any discount is applied.
discount_applied	Whether a discount applies to your rate. When true, value is lower than original_value.
start_time	When the rate takes effect, or null when it has no start time.
end_time	When the rate stops applying, or null when it has no end time.

A discount_applied of false, with no start_time or end_time, means your account is billed the open-ended list price for that resource. A contract that reserves GPUs at an agreed discount is reflected in value, and the undiscounted rate is preserved in original_value.

Calculating a flavor's hourly cost

Look up the flavor you intend to deploy, then combine its allocations with the matching value from the pricebook:

• GPU flavors: multiply the value of the flavor's GPU by the flavor's GPU count. The CPU, RAM, and local storage resources all return a value of zero, so they add nothing.
• CPU-only flavors: multiply the CPU cores, the RAM in GB, and the combined root and ephemeral disk in GB by their (cpu-only-flavors) rates, then add the three results together.

Add the PublicIP rate for each public IP address, and the Cloud-SSD rate multiplied by the size in GB of each attached volume.

AI Studio model pricing

AI Studio models are not billed by the hour, so their rates are not comparable to the hourly rates on this page. AI Studio uses three billing models: token-based pricing for text processing, per-patch pricing for image generation, and hourly pricing for tasks that use dedicated compute. See AI Studio Billing.

The Pricebook API response still carries an entry for the input price and the output price of every model, each named for its model with an (input) or (output) suffix.

Back to top