Skip to main content

What Is Hyperstack?

A GPU cloud platform for AI, machine learning, and other compute-intensive workloads.

Hyperstack is a GPU cloud platform built for AI training, fine-tuning, and inference. As an NVIDIA Preferred Partner, it gives you on-demand, self-serve access to NVIDIA GPUs, billed by the minute. Run your own workloads on GPU virtual machines and Kubernetes clusters, or use AI Studio to run models with no infrastructure to manage.

GPU CloudHyperstack data centers CANADA-1 · NORWAY-1 · US-1
AI StudioModel inference

What you can run​

Hyperstack covers two kinds of work: infrastructure you deploy and control, and AI Studio, where you run models.

ProductWhat it gives youStart here
Virtual machinesGPU or CPU machines with full control of the operating system and software, which you connect to over SSH.Deploy a virtual machine
KubernetesOn-demand clusters with GPU worker nodes, deployed from the console or through the API.Deploy a Kubernetes cluster
StorageNVMe root and ephemeral disks on each virtual machine, volumes that persist independently of it, and S3-compatible object storage.Storage
NetworkingFirewalls, public IP addresses, and SR-IOV high-speed networking between virtual machines.Network
Serverless inferenceHosted text and image models from a shared catalog, called from the AI Studio playgrounds or an OpenAI-compatible API. Nothing to deploy, and you pay per use.Run a model
Dedicated inferenceThe Deploy AI wizard sizes GPUs to an open-weight model, deploys a GPU virtual machine in your environment to serve it, and issues a private, API key-protected endpoint. You pay for the machine by the hour while it runs.Deploy dedicated inference

Why Hyperstack​

Hyperstack is built for GPU workloads such as AI training, fine-tuning, and inference, and it runs on servers and networks that it owns and operates.

  • Dedicated hardware with no oversubscription. Every virtual machine flavor maps 1:1 to the physical GPUs, CPUs, memory, and storage of its host, so you receive the full resources you deploy. See Zero oversubscription.
  • Current NVIDIA GPUs on demand. The flavor catalog covers current NVIDIA GPUs, including Blackwell-generation B300, available on demand.
  • You pay for the GPUs, by the minute. On a GPU flavor, only the GPUs are billed; the CPU cores, RAM, and local NVMe storage are included. Spot virtual machines and contracts lower the rate further. See Resource Pricing.
  • High-speed networking. In CANADA-1 and US-1, compatible flavors in SR-IOV-enabled environments reach up to 350 Gbps of inter-VM bandwidth for distributed training and multi-node inference. See High-Speed Networking.
  • Certified, sustainably powered data centers. Hyperstack runs in Tier 3 certified data centers, and NexGen Cloud has achieved SOC 2 Type 2, an independent audit of the security controls that protect your data. Hyperstack offers data centers in Europe and North America powered by 100% renewable energy. See Regions & Data Centers.

Key concepts​

These concepts decide where your resources run, who can manage them, and what they cost.

Your organization​

Signing up creates a Hyperstack organization under your name, with you as its owner. You invite members and grant them access through user roles. Billing belongs to the organization, so resources that any member creates are billed to its account.

Regions and environments​

Every resource you deploy belongs to exactly one environment, and every environment belongs to exactly one region. The environment decides which data center runs the resource, which private subnet its virtual machines join, and which SSH keys and firewalls are available to it. Most deployment forms let you create an environment in place.

Each region is backed by a dedicated data center in the country its name reflects: CANADA-1 in Quebec, NORWAY-1 in Vestland, and US-1 in Texas. With AI Studio serverless inference, there is nothing to deploy. Dedicated inference deploys onto a GPU virtual machine in one of your environments, so it follows the same regions and flavors as any other virtual machine.

Flavors​

A flavor is a preset combination of GPU, CPU, RAM, and disk that you select as a single unit when you deploy a virtual machine or a cluster's worker nodes. Availability differs by region and changes with demand, so check GPU stock before you plan a deployment.

Pricing and billing​

Infrastructure is billed for the time a resource spends in a billed state, not the time you actively use it. A virtual machine that is running or stopped still holds its hardware and keeps billing. Hibernation releases the GPUs and stops their billing, and deleting a resource stops its billing entirely. In AI Studio, serverless inference is billed per token for text models and per image patch for image generation, and dedicated inference is billed at the hourly rate of its configuration while its machine runs.

Accounts are prepaid by default: you add credit, and usage draws it down as your resources run. For the full rules, see VM States & Billing and AI Studio Billing.

Console and API​

The console and the APIs work with the same resources, so a workflow you start by hand can move to code:

Reserved capacity and custom agreements

This page covers Hyperstack's on-demand platform. For guaranteed capacity, you can reserve GPUs for a fixed term at discounted rates with a contract, or request a dedicated GPU cluster provisioned exclusively for your organization. To discuss either, or another custom agreement, email the sales team.

Next steps​

Getting Started

Select what you want to run and how you want to work, and find the guide that fits.

Set Up Your AI Agent

Connect your AI agent to the docs, your infrastructure, or AI Studio models.

Resource Pricing

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