Cloud Pricing Models
How Cloud Pricing Models Work
Cloud pricing models are the ways a provider charges for computing resources. The main choice is between paying for capacity as you use it and committing in advance to a level of use in return for a lower rate.
Most bills combine several models. A team may run its steady core workload on committed capacity, cover peaks with on-demand servers and use spare capacity for batch jobs that can be interrupted. Charges for moving data are a separate line item, explained in egress fees.
On-Demand, Reserved and Savings Plan Options
- On-demand (pay-as-you-go): you pay for each period a resource runs, with no commitment. It gives the most flexibility, usually at the highest unit rate.
- Reserved or committed use: you agree to use a set amount of a resource for a fixed term, often one to three years. The provider offers a lower rate in return, but you pay for the commitment whether or not you use all of it.
- Savings plans: some providers let you commit to a steady spend per hour across a range of services, rather than to one specific server type. This keeps more flexibility while still lowering the rate.
The size of the servers you commit to matters too. Server families and sizes are described in instance types.
Spot and Interruptible Capacity
Spot instances, called preemptible or low-priority capacity by some providers, are spare servers offered at a steep discount. The catch is that the provider can reclaim them at short notice when it needs the capacity back.
Spot capacity suits work that can stop and restart safely, such as batch processing, video rendering, testing and some data analysis. It is a poor fit for a checkout service that must always respond. Well-designed jobs save progress regularly so they can resume on another server.
A Mixed Pricing Plan for a News Site
A news site has steady traffic during the day and sharp spikes after major stories. The team commits to reserved capacity for the baseline that runs all the time, uses on-demand servers for the spikes and runs image resizing jobs on spot capacity. Those jobs save checkpoints (progress markers), so a reclaimed job can continue from its last checkpoint.
The result is a bill that follows demand more closely than any single pricing model would.
Frequently asked questions
Is reserved pricing always cheaper than on-demand?
Only when you use the resource consistently. A commitment saves money if you would otherwise run the same server most of the time. If your usage drops, you still pay for the commitment, and the saving can disappear. Review usage before committing.
Are spot instances safe for production websites?
Usually not for the main customer-facing layer, because spot capacity can be reclaimed at short notice. Spot works best for interruptible work, or as one part of a fleet in which committed or on-demand servers can carry the load if spot servers are removed.
How do I choose a pricing model?
Start by measuring how steady your usage is over several months. A predictable baseline suits commitments, unpredictable peaks suit on-demand capacity, and interruptible jobs suit spot. Regular cost reviews, covered in FinOps, keep the choice up to date.