Blog Post

Azure Compute Blog
5 MIN READ

Generally available: single instance launch mode and vCPU-based capacity for Azure Compute Fleet

Jerry Steele's avatar
Jerry Steele
Icon for Microsoft rankMicrosoft
Nov 17, 2026

Today we're announcing the general availability of two Azure Compute Fleet features that make it easier to get the capacity you need and run it your way:

  • Single instance launch mode: a one-time, launch-only request. Fleet finds and provisions the VMs, then hands them to your own tools, which manage their lifecycle.
  • vCPU-based capacity: set your target capacity in vCPUs instead of a VM count. Fleet works out how many VMs to create from the sizes you allow.

You can use them separately or together. vCPU-based capacity works with the new single instance launch mode and with Managed mode, the existing default, where the fleet stays in place so you can adjust capacity over time. Both features also work with multi-region compute fleet, currently in preview, so one request can spread capacity across several Azure regions.

Azure Compute Fleet deploys up to 10,000 VMs in a single request across a mix of VM sizes, availability zones, and Spot and Standard pricing. If your application can run on several VM sizes, adding more of them to your request improves your chances of getting the capacity you need. Fleet chooses from those options based on your allocation strategy, so you don't have to build that logic yourself.

These features fit any workload that needs capacity at scale, such as stateless web services, batch jobs, CI pipelines, and platforms built on their own schedulers or control planes. Financial risk analysis, rendering, and other HPC and grid workloads are common examples too.

Benefits of single instance launch mode

Many teams already have a scheduler, autoscaler, or custom platform that decides when to add and remove VMs. Until now, a fleet stayed in place to manage the VMs it created, which doesn't fit when another tool already does that job. Single instance launch mode separates getting capacity from managing its lifecycle. It's a one-time, launch-only request: you send it once, and Azure Compute Fleet picks the sizes, zones, and pricing to fill it, then creates the VMs. After that, the fleet resource deletes itself. The VMs remain as standard Azure VMs, and your tools take it from there.

  • Keep your existing tools: Fleet finds and provisions the VMs. Your automation or scheduler manages their lifecycle from there.
  • Provision at scale in one request: get up to 10,000 VMs across 1 to 10 VM sizes, your chosen zones, and Spot and Standard pricing in a single request. Being flexible on sizes and zones gives Fleet more places to find capacity.
  • Manage standard VMs: each VM is a standalone Azure VM, so you can use the VM APIs, Azure CLI, Azure PowerShell, or the portal.
  • Find your VMs easily: set a name prefix for the VMs, and tags on the fleet carry over to every VM it creates.

Benefits of vCPU-based capacity

  • List your sizes, and Fleet works out the mix: give Fleet the VM sizes your workload can run on and a target capacity in vCPUs, such as 1,000. Fleet decides how many VMs of each size to create, based on your allocation strategy and the capacity available. You don't have to work out the math yourself.
  • Keep the same target as sizes change: your request stays in vCPUs whatever mix of sizes Fleet provisions.
  • Add sizes with less effort: because your target doesn't depend on any one size, adding more sizes is easy, and each one gives Fleet another way to reach your total.
  • Use it in either mode: set a vCPU target on a Managed mode fleet that stays in place, or on a Single Instance Launch Mode request that hands the VMs to your own tools.

For example, a 1,000-vCPU request might be filled with 50 VMs with 4 vCPUs, 50 with 8, and 25 with 16. That's 125 VMs and 1,000 vCPUs.

A few things to know before you start. In Single Instance Launch Mode, Fleet doesn't replace evicted Spot VMs, and you can't change a request after you submit it. With vCPU-based capacity, you choose the capacity type when you create the fleet. Fleet provisions whole VMs, so you might get a few more vCPUs than you asked for. You pay for the VMs that are provisioned, and your usual quotas apply.

How it works

  1. Create a fleet with capacityType set to VCpu and mode set to Launch (Single Instance Launch Mode). List the VM sizes and zones your workload can use, and pick an allocation strategy.
  2. Fleet chooses VMs from your list and provisions enough whole VMs to reach your vCPU target.
  3. Use the List VMs API to get the VM resource IDs and check their status.
  4. Manage the VMs with your own tools. The fleet resource deletes itself after about five hours, and the VMs stay until you delete them.

This ARM resource asks for 1,000 Standard vCPUs across three VM sizes with the LowestPrice allocation strategy. You'll also need a baseVirtualMachineProfile parameter with your image, OS, and network settings. The quickstart shows how to set it up.

{
  "type": "Microsoft.AzureFleet/fleets",
  "apiVersion": "2026-08-01",
  "name": "myCapacityRequest",
  "location": "eastus",
  "zones": ["1", "2", "3"],
  "properties": {
    "mode": "Launch",
    "capacityType": "VCpu",
    "vmNamePrefix": "myapp",
    "vmSizesProfile": [
      { "name": "Standard_D4s_v5" },
      { "name": "Standard_D8s_v5" },
      { "name": "Standard_D16s_v5" }
    ],
    "regularPriorityProfile": {
      "capacity": 1000,
      "allocationStrategy": "LowestPrice"
    },
    "computeProfile": {
      "computeApiVersion": "2024-03-01",
      "baseVirtualMachineProfile": "[parameters('baseVirtualMachineProfile')]"
    }
  }
}

Here, capacity: 1000 means 1,000 vCPUs, not 1,000 VMs.

Provisioning runs in the background. Use the List VMs API to track it:

GET https://management.azure.com/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.AzureFleet/fleets/myCapacityRequest/virtualMachines?api-version=2026-08-01

Poll until each VM shows Succeeded or Failed, and save the resource IDs before the fleet resource goes away.

Example: IBM Spectrum Symphony

The IBM Spectrum Symphony Provider Plugin for Microsoft Azure Compute Fleet uses both features. It turns workload demand into vCPU-based Fleet requests and uses Single Instance Launch Mode to provision execution hosts. Symphony keeps control of scheduling and the hosts, and can release the Azure resources when demand drops. Teams can extend an existing Symphony environment to Azure without changing how they schedule work. Read about the integration.

Availability and pricing

Single Instance Launch Mode and vCPU-based capacity are generally available today in all Azure public regions. There's no extra charge for either feature or for Azure Compute Fleet. You pay only for the VMs that Fleet provisions.

To learn more, visit the single instance launch mode and vCPU-based capacity documentation pages, then follow the quickstart to create your first fleet.

Updated Nov 12, 2026
Version 1.0
No CommentsBe the first to comment