Pricing model
Overview
ECI offers three pricing models depending on your use case and time horizon. Each has different rates, commitment terms, and reclamation risk; picking the right one for your workload can save a lot of money.
For hourly rates by instance type, see the official price list.
Comparison
| Item | Spot | On-demand | Reserved |
|---|---|---|---|
| Commitment | None | None | Yes |
| Capacity acquisition | Opportunistic from idle capacity | Immediate from the available pool | Guaranteed for the term you reserved |
| Can be reclaimed while running? | Yes (force-reclaimed when capacity tightens) | No | No |
| Best for | Cost savings, interruption-tolerant jobs | Short-term, flexible work | Long-term, steady operation |
Block storage, object storage, and public IPs are billed independently of the VM pricing type. Stopping the VM does not stop these charges. To stop them entirely, delete the resource.
Spot
A discount on ECI's idle GPU capacity relative to on-demand. The discount varies with supply and demand, and a running VM can be force-reclaimed when capacity is tight. See Operating spot VMs for reclamation detection and checkpointing.
Good fit
- ML training jobs that checkpoint
- Batch processing that can resume after interruption
- Short-term work where cost is the top priority
Not a good fit
- 24/7 inference servers
- Real-time services where saving state is hard
- Production environments where interruption is costly
On-demand
You commit to nothing and are billed for time the VM is running. Capacity is taken from the zone's available pool, so you can start as soon as capacity exists, and you can stop or delete at any time. If the pool is temporarily exhausted, creation may be restricted.
Good fit
- Development, test, and PoC environments
- Workloads with irregular usage
- First-time ECI users
Reserved
A separate sales contract that pre-reserves capacity for a fixed term. The reserved capacity is guaranteed during the term, at a stable discounted rate.
Good fit
- Services running continuously for 3+ months
- Projects that need fixed budgets up front
- Anyone who needs guaranteed capacity
How it works
- When reserved pricing exists and you have quota left, the reserved option appears in VM creation
- Beyond the quota, you can keep creating on-demand VMs
- When the contract nears expiration, the portal home and VM creation screen show the days remaining
For onboarding, price-list lookup, resource status, and expiration management, see the Reserved resources guide.
When a reservation expires, the contract itself is deactivated and the VMs under it can no longer be used (the portal shows "This VM's reservation has expired. Please contact your administrator."). VMs do not automatically fall back to on-demand rates, so check the expiration date in advance and ask the sales team to renew or extend.
FAQ
Can I use reserved and on-demand at the same time?
Yes. Beyond the reserved quota, you can create additional resources on-demand.
Where do I pick the pricing type?
On the Basic Info step of the VM creation wizard, under Pricing type. Reserved only shows up if your organization has reserved pricing enabled; spot is only selectable for GPU instance types.
Can I switch an on-demand VM to reserved later?
No, pricing type can't be changed after creation. To use a different type, create a new VM. You can keep the data by detaching the existing block storage and attaching it to the new VM.
What happens to the storage data if a spot VM is reclaimed?
The block storage attached to the VM is kept. VM memory and local ephemeral data can be lost.
Next steps
- Operating spot VMs: reclamation detection and checkpointing
- Usage lookup: review usage and analyze costs
- Official price list: hourly rates by instance type