Skip to main content

GPU Stock Information

Check current and upcoming GPU stock by model, region, and count, in the console or through the API.

GPU availability on Hyperstack is real time. It shifts as GPUs are deployed and freed across the platform, and it varies by GPU model, region, and GPU count. This page explains how to check current and upcoming stock, both in the console when you deploy and through the Retrieve GPU stocks API, and how stock affects deploying new resources.

Live GPU stock​

The dashboard below reads live availability from the GPU stock API and refreshes automatically. Filter by region, and check the per-model availability and the GPU counts you can deploy right now. Availability changes from moment to moment, so if a flavor is unavailable, check back shortly as capacity frees up. For the most reliable figures, query the API directly.

How GPU stock works​

Every GPU you deploy draws from a shared, finite pool of hardware in each region. Because the pool is shared, availability changes from moment to moment: a model that is in stock now can be unavailable later, and capacity that is gone can return as other workloads are released. Hyperstack tracks availability along three dimensions:

  • GPU model: each model, such as H100-80G-PCIe or L40, is counted on its own. Spot variants are tracked as separate models.
  • Region: stock is held per region, so a model can be in stock in CANADA-1 and out in US-1. See Regions for the full list.
  • GPU count: the same model can have plenty of single-GPU availability but none at 8×, because a multi-GPU deployment needs that many GPUs free on one machine.

Check stock in the console​

When you deploy a virtual machine or a Kubernetes cluster, the GPU Flavor section shows live availability on every flavor card. This is the quickest way to see what you can deploy right now, without leaving the deploy form.

  1. Navigate to the Deploy Virtual Machine page. To check availability for Kubernetes worker nodes, open the Create a Kubernetes Cluster page instead, which uses the same flavor cards.

  2. Find the GPU model you want in the GPU Flavor grid. Each card represents one flavor family.

  3. Read the card's availability:

    • In stock: the card shows the per-hour price and a GPU-count dropdown (1×, 2×, 4×, 8×, and 10× on flavors that offer it). Choose a count to select it and see its price.
    • Out of stock: the card shows an Out of Stock badge and a Click here to reserve link in place of the price. See When a GPU is out of stock.
The GPU Flavor grid on the Deploy Virtual Machine page. In-stock cards show a per-hour price and a GPU-count dropdown, while out-of-stock cards show an Out of Stock badge and a Click here to reserve link in place of the price.The GPU Flavor grid on the Deploy Virtual Machine page. In-stock cards show a per-hour price and a GPU-count dropdown, while out-of-stock cards show an Out of Stock badge and a Click here to reserve link in place of the price.

Availability is shown per region, so check the region you intend to deploy in. For Kubernetes, only worker node flavors draw from GPU stock; master nodes are CPU-only and free. See the Kubernetes cluster guide.

Check stock with the API​

To monitor availability programmatically, or to surface it in your own tooling, call the Retrieve GPU stocks endpoint. A single request returns current and upcoming stock for every GPU model in every region.

The request takes no parameters and authenticates with your API key:

Check GPU stock
curl -X GET "https://infrahub-api.nexgencloud.com/v1/core/stocks" \
  -H "accept: application/json" \
  -H "api_key: YOUR_API_KEY"

The response groups GPU models by region. Each entry in the stocks array carries a region, a stock-type of GPU, and a models array.

Below is a sample response, trimmed to a single region and one GPU model:

Response
{
  "stocks": [
    {
      "region": "CANADA-1",
      "stock-type": "GPU",
      "models": [
        {
          "model": "H100-80G-PCIe",
          "available": "10+",
          "planned_7_days": "0",
          "planned_30_days": "0",
          "planned_100_days": "0",
          "configurations": {
            "1x": 24,
            "2x": 4,
            "4x": 0,
            "8x": 0,
            "10x": 0
          }
        }
      ]
    }
  ]
}

A full response returns every model in CANADA-1, NORWAY-1, and US-1.

Each model reports:

FieldDescription
modelThe GPU model name, such as H100-80G-PCIe. Spot variants appear as separate models.
availableHow many of the model are available, reported conservatively. 10+ means at least 10; 0 means none.
configurationsThe number of VMs you can deploy at each GPU count: 1x, 2x, 4x, 8x, and 10x.
planned_7_days, planned_30_days, planned_100_daysAnticipated additional stock over the next 7, 30, and 100 days. null when no forecast is available.

When a GPU is out of stock​

GPU availability returns as workloads are released, so an out-of-stock model is often a temporary state. When the model you want shows 0 or an Out of Stock badge, you have a few options:

  • Reserve the capacity. On an out-of-stock flavor card on the VM or Kubernetes deploy form, click Click here to reserve to request that configuration from the Hyperstack team. See Check stock in the console.
  • Try another region or GPU count. The same model may be in stock in a different region, or available at a lower GPU count.
  • Monitor and retry. Check the Retrieve GPU stocks API, then deploy or restore once availability returns.
  • Request a dedicated cluster. For guaranteed, large-scale capacity, request a dedicated GPU cluster.

Back to top

On this page