Regions & Providers
Spheron sources GPU and CPU capacity from six providers across North America, Europe, the Middle East, and Asia Pacific. Most GPUs are NVIDIA; Spheron AM supplies AMD Instinct, and Intel Gaudi2 is available through Spheron MS.
GPU inventory changes as providers add and withdraw hardware. The tables below list representative capacity; check the GPU catalog for live availability and pricing.
Provider Capabilities
| Provider | Instance Types | CPU Node | Volume Support | GPU Tiers | Regions |
|---|---|---|---|---|---|
| Verda | Spot, Dedicated | Yes (Spot and Dedicated) | NFS volumes (up to 10 TB, multi-instance) | B300 SXM6, A100 80GB SXM4, RTX Pro 6000, A6000, V100 | EU (Finland 1, Finland 2, Finland 3) |
| Spheron AI | Spot, Dedicated (varies by region) | Yes (Dedicated) | Block volumes (up to 50 TB, up to 10 per instance) or shared volumes (up to 2,048 GB), depending on region | H100 (NVL/PCIe/SXM), H200 SXM, A100 80GB (NVL/PCIe/SXM4), L40, RTX 4090, RTX Pro 6000, L4 | Canada 1, Montreal 1, Norway 1, Oslo 1, US, Noida, Chennai (Tier 3 DC) |
| Spheron ES | Spot, Dedicated | No | virtiofs shared volumes (up to 256 TiB, multi-instance) | B300 SXM6, B200 SXM6, H200 SXM5, H100 SXM5, L40S, RTX Pro 6000 | EU North 1 (Finland), EU West 1 (France), UK South 1, ME West 1 (Middle East), US Central 1 (US); Tier 3 DC |
| Spheron MS | Dedicated (VM and bare metal), Spot (where available) | No | Block volumes (one per instance, bound at launch) | H200 SXM5, H100 (SXM5/PCIe), A100 80GB (SXM4/PCIe), A100 40GB (SXM4/PCIe), RTX Pro 6000, RTX 5090, RTX 4090, L40S, L4, Gaudi2, A6000, A5000, A4000, A16, A10, V100 | US, Canada, Norway, Poland, Israel, India, Japan, Singapore, Australia |
| Spheron AM | Dedicated | No | No | AMD Instinct MI300X (192 GB HBM3e), in 1x, 2x, and 4x | Michigan 1 (US) |
| Spheron Con | Dedicated (containers: a Docker image, not a full VM) | Yes (CPU containers, 2 to 32 vCPU) | Spheron Con volumes (10 to 4,096 GB, one per container, chosen at deploy, shareable across containers) | 28 GPU types, from RTX 2000 Ada and A40 to H100, H200, B200 and B300, in 1, 2, 4 or 8 GPU sizes | North America, Europe, and Asia (Japan) |
| Massed Compute | Dedicated | Yes (Dedicated) | No | H100 PCIe, A100 80GB PCIe, L40S, L40, RTX 6000 Ada, RTX Pro 6000, A6000, A30 | US regions (us-central, us-east) |
Instance Lifecycle Support
Stopping an instance keeps its disk while the GPU stops billing. Restarting reboots it in place. Neither is available on every provider, and the two do not line up with each other.
| Provider | Stop and start | Restart |
|---|---|---|
| Spheron AI | Yes, depending on machine type and region | Yes, depending on region |
| Spheron ES | Yes | No |
| Spheron MS | No | Yes, on virtual machines |
| Spheron AM | No | Yes |
| Spheron Con | GPU containers only | Yes |
| Massed Compute | No | Yes |
| Verda | No | No |
On a provider that cannot stop an instance, the instance bills at the full hourly rate until you destroy it. Some machine types also carry a provider minimum runtime longer than the 20 minute default, charged in full even if you terminate early; the deploy wizard shows the figure before you commit. See Instance lifecycle for stopped rates, capacity risk on start, and the API endpoints.
Provider Details
Verda
Verda operates in EU regions across three Finland data centers: Finland 1, Finland 2, and Finland 3. It provides both Spot and Dedicated instances with NFS volume support.
CPU Node: Verda is the only provider offering CPU nodes at spot rates, from $0.09/hr. Sizes run 4 vCPU / 16 GB, 8 vCPU / 32 GB, 16 vCPU / 64 GB, and 32 vCPU / 128 GB. See CPU Nodes.
Verda volumes (NVMe_Shared NFS type) support multi-instance attachment: a single volume can be mounted on multiple GPU instances simultaneously. Up to 10 volumes can be attached to one instance.
Local storage for Dedicated instances: When deploying a Verda Dedicated instance you choose between two root-disk sizes: 500 GB (default) or 1 TB. Select 1 TB when your workload requires more local dataset or checkpoint space than the default provides.
Spheron AI
Spheron AI provides GPU and CPU instances from Tier 3 data centers in North America, Europe, and India. Its regions are Canada 1, Montreal 1, Norway 1, Oslo 1, the United States, Noida, and Chennai. Inventory includes H100 (NVL/PCIe/SXM), H200 SXM, A100 80GB (NVL/PCIe/SXM4), L40, RTX 4090, RTX Pro 6000, and L4. Stock rotates by region; check the catalog for what is deployable right now.
Some options depend on the region you deploy into. The deploy wizard only offers what the selected region supports.
| Option | Canada 1, Montreal 1, Norway 1, Oslo 1, US | Noida, Chennai |
|---|---|---|
| Instance types | Spot and Dedicated | Dedicated |
| Volumes | Block volumes, up to 50 TB each, up to 10 per instance, attach and detach while running | Shared volumes, 50 GB to 2,048 GB, one per instance, chosen at deploy time and mounted automatically |
| OS images | 11 images to choose from | One image per region |
| Startup scripts | Supported | Not supported |
| Stop and start | Depends on machine type and region | Every machine |
| Restart | Supported | Not supported |
Volumes: The size slider in the create-volume wizard adapts to the selected region's limits. See the Spheron AI volume mounting guide for both volume types.
Root disk: In Noida and Chennai, you choose a 200 GB, 500 GB, or 1 TB root disk in the wizard. 500 GB is preselected, and the hourly price shown already includes the disk you pick.
OS images: In Canada 1, Montreal 1, Norway 1, Oslo 1, and the US, you choose from 11 images, including Ubuntu 22.04/24.04, AlmaLinux 9, Debian 12, and CUDA-enabled variants (R535, R550, R570). The default is Ubuntu Server 22.04 LTS with CUDA 12.8 and Docker pre-installed. In Noida, GPU machines run Ubuntu Server 22.04 LTS with the R580 driver, CUDA 13.0, and Docker. In Chennai, they run Ubuntu Server 24.04 LTS with the R595 driver, CUDA 13.2, and Docker. CPU nodes in both regions run Ubuntu Server 24.04 LTS with Docker.
SSH access: Port 22 opens automatically on every new Spheron AI instance; no firewall configuration is needed for SSH. In Noida and Chennai, a deployment requires an SSH key, and you log in as the ubuntu user. Copy the exact command from the instance details panel.
CPU Node: Spheron AI offers Dedicated CPU nodes from 4 vCPU / 4 GB up to 32 vCPU / 64 GB in Norway 1 and Canada 1, and from 2 vCPU / 8 GB up to 32 vCPU / 128 GB in Noida and Chennai. CPU nodes in Noida and Chennai do not take volumes.
Spheron ES
Spheron ES (Extra Supply) provides on-demand and spot GPU instances across five regions: EU North 1 (Finland), EU West 1 (France), UK South 1 (United Kingdom), ME West 1 (Middle East), and US Central 1 (United States). Inventory currently spans B300 SXM6, B200 SXM6, H200 SXM5, H100 SXM5, L40S, and RTX Pro 6000, all flagged with the Tier 3 DC compliance badge.
Spot pricing: Spheron ES spot follows a live price set per GPU type and region. It can change as often as every 15 minutes and never exceeds the dedicated price for the same machine in the same region. See Spot pricing.
Spheron ES does not offer CPU nodes. Deploy CPU work on Verda, Spheron AI, or Massed Compute.
Shared volumes use the virtiofs protocol and support multi-instance attach, in-place resize, and up to 256 TiB per volume. See the Spheron ES volume mounting guide for setup.
Spheron MS
Spheron MS (Multi Supply) provides Dedicated and Spot GPU instances across a wider geographic footprint than any other provider on the platform: the United States (Austin, Ashburn, Chicago, Culpeper, Dallas, Dulles, Houston, New York, Phoenix, Salt Lake City, San Jose), Canada (Calgary), Norway (Oslo), Poland (Warsaw), Israel (Jerusalem), India (Mumbai), Japan (Tokyo), Singapore, and Australia (Sydney). Inventory runs from H200 SXM5 and H100 SXM5 at the high end down to L4 and A10 for inference, and includes Gaudi2.
Each offer names its region, and the region on the offer is the region you deploy into. A machine that is out of stock shows as sold out rather than disappearing, so you can set a stock notification instead of wondering where a GPU went.
Instance types: Dedicated, plus Spot where the underlying cloud publishes spot capacity. A Spheron MS spot offer is listed separately from the on-demand offer for the same machine, and its price never exceeds that on-demand price. The provider can reclaim it at any time. Where an offer is a full physical server it is listed as bare metal and labelled on the card. See Instance types.
Volumes: A Spheron MS volume is selected during deployment and only then. There is no attach or detach for an instance that is already running, each instance takes one volume, and the volume must be in the same region as the instance. Size and name are fixed at creation. See the Spheron MS volume mounting guide.
Boot disk: Each machine type ships a fixed boot disk, so its size appears on the offer rather than being something you set. Pick a machine type with the disk you need, or attach a volume for the rest.
Lifecycle: Spheron MS cannot stop an instance, so an instance bills at the full hourly rate until you destroy it. Restart is available on virtual machines and reboots in place without changing billing. Bare metal instances support neither.
OS images: Ubuntu 20.04, 22.04, and 24.04, with CUDA 12.0, 12.2, 12.4, 12.8, and 13.0 variants. Startup scripts (cloud-init) are supported on every offer.
CPU Node: Spheron MS does not offer CPU nodes. Deploy CPU work on Verda, Spheron AI, or Massed Compute.
Spheron AM
Spheron AM supplies AMD Instinct MI300X, the first non-NVIDIA silicon on the platform. Each GPU carries 192 GB of HBM3e, and the catalog publishes 1x, 2x, and 4x GPU configurations from a single region, Michigan 1 in the United States.
The 192 GB per GPU is the reason to choose it: a model that needs several 80 GB NVIDIA cards to hold its weights can fit on fewer MI300X GPUs, which changes how you shard it.
Compute stack: These machines run ROCm, not CUDA. Most PyTorch and TensorFlow code runs unchanged, but nvidia-smi does not exist and CUDA-only wheels and Docker images do not work. See ROCm and AMD Instinct before you deploy.
Instance types: Dedicated only. Spheron AM has no spot option.
OS images: One image, Ubuntu with ROCm plus Docker or Podman. There is no image picker, because the machine type determines what it boots. Startup scripts (cloud-init) are supported.
Boot disk: Fixed by the machine type, so its size appears on the offer rather than being something you set.
Volumes: Not available. A deployment that asks for a volume is refused rather than created without it. Keep anything that must survive termination off the instance.
Lifecycle: Restart reboots in place. Stopping is not available, so an instance bills at the full hourly rate until it is destroyed. Terminate an instance you have finished with.
Minimum runtime: Each machine type carries its own minimum reservation, which can exceed the 20 minute default and is charged in full even if you terminate early. The deploy wizard shows the figure before you commit.
CPU Node: Spheron AM does not offer CPU nodes.
Massed Compute
Massed Compute provides Dedicated GPU instances in US regions (us-central-1 through us-central-9 and us-east-1) with H100 PCIe, A100 80GB PCIe, L40S, L40, RTX 6000 Ada, RTX Pro 6000, A6000, and A30 configurations. Massed Compute does not support volume mounting.
CPU Node: Massed Compute offers Dedicated CPU nodes at 8 vCPU / 32 GB, 14 vCPU / 40 GB, 28 vCPU / 80 GB, and 52 vCPU / 160 GB, from $0.12/hr, in us-central-2, us-central-3, and us-east-5. See CPU Nodes.
Compliance certifications: HIPAA, SOC 2 Type II.
Region Selection Guidance
- Lowest SSH latency: Choose the provider and region geographically closest to you
- GDPR / EU data requirements: Use Verda (Finland, ISO 27001)
- Asia Pacific presence: Spheron MS (India, Japan, Singapore, Australia) or Spheron AI (Noida, Chennai)
- H200 SXM5: Spheron ES (EU North 1, EU West 1, UK South 1, ME West 1, US Central 1), Spheron MS, or Spheron AI
- B300 SXM6: Spheron ES or Verda
- B200 SXM6: Spheron ES (EU North 1, EU West 1, UK South 1, ME West 1, US Central 1)
- Multi-GPU bare-metal training: Spheron ES (B300 SXM6, B200 SXM6, H200 SXM5, H100 SXM5)
- CPU-only workloads: Verda (Finland, spot and dedicated), Spheron AI (Norway 1, Canada 1, Noida, Chennai), or Massed Compute (US)
- Shared persistent volumes: Verda (multi-instance NFS attach within the same region) or Spheron ES (multi-instance virtiofs, up to 256 TiB per volume)
- Largest block volumes: Spheron AI (up to 50 TB per volume, up to 10 volumes per instance)
- Widest choice of cities: Spheron MS, which lists GPUs in 19 locations across nine countries
- AMD Instinct MI300X, or the most GPU memory per card: Spheron AM (192 GB HBM3e per GPU, Michigan 1); see ROCm and AMD Instinct
- Intel Gaudi2: Spheron MS
- Stopping an instance to pause GPU billing: Spheron AI or Spheron ES; see Instance lifecycle
- Reserved GPU requests: Select "Any Location" on a custom reserved GPU request unless you need a specific region, so the Spheron team can match you with more capacity
What's next
- API Reference: Query available providers and GPU offers programmatically
- Instance Types: Spot vs Dedicated, and hardware isolation categories
- CPU Nodes: CPU-only instances, sizes, and regions
- Distributed Training: Multi-GPU bare-metal setup
- Instance Lifecycle: Stop, start, and restart support by provider
- ROCm and AMD Instinct: Running workloads on MI300X
- Volume Mounting: Shared storage for Verda, Spheron AI, Spheron ES, Spheron MS, and Spheron Con
- Reserved GPUs: Long-term GPU reservations