Cloud Pricing Models Compared: On-Demand, Commitments, Spot, Hourly Caps and Prepaid Balances
How the pricing models of twelve cloud providers actually work, from AWS Savings Plans and Google sustained use discounts to hourly billing with monthly caps and UpCloud’s prepaid balance.
Every cloud provider says it is pay-as-you-go, and every one means something slightly different. The hyperscalers layer commitment discounts and spot markets on top of per-second billing. Developer clouds bill hourly with a monthly cap. Some prepay, some auto-renew, some bill by the minute. This guide explains each model, which providers use it and what it implies for how you run your infrastructure. For the day-to-day practice of controlling spend, see our cloud cost management guide.
The models #
| Model | How it works | Who uses it |
|---|---|---|
| On-demand, per second or hour | Pay for what runs with no commitment; the highest unit price and the most flexibility | AWS, Google Cloud, Azure, Oracle Cloud, Alibaba Cloud (hourly, with sub-hour usage billed as a full hour) |
| Reserved instances | Commit to an instance or resource type for one or three years for a significant discount | AWS, Azure |
| Savings Plans | Commit to a consistent hourly spend for lower prices, with flexibility across instance families | AWS |
| Committed use discounts | Commit to a level of resources for one or three years | Google Cloud |
| Sustained use discounts | Automatic discounts for resources that run a large part of the month; no sign-up | Google Cloud |
| Spot or preemptible capacity | Spare capacity at a large discount; instances can be interrupted, preempted or evicted | AWS Spot Instances, Google Cloud Spot VMs, Azure Spot Virtual Machines |
| License benefits | Apply existing licenses to cloud resources for a lower price | Azure Hybrid Benefit for Windows Server and SQL Server |
| Hourly billing with a monthly cap | Billed by the hour, never more than the plan’s monthly price | DigitalOcean, Linode, Hetzner, UpCloud (capped at 28 days) |
| Subscription or prepaid term | Prepay for a term at a discount | Alibaba Cloud subscription, Kamatera monthly billing, OVHcloud services |
| Automatic renewal | Services renew on their renewal date and are debited from the default payment method; manual renewal can be set for some services | OVHcloud |
| Prepaid balance | Add funds; hourly usage is deducted from the balance | UpCloud |
| Post-paid, pro-rated | Hourly usage accrues and is invoiced at month end or when a billing limit is reached | Linode |
| Per-minute billing | Variable cost by the minute depending on whether servers are on or off | Kamatera hourly plans |
What each model implies #
Per-second on-demand rewards elasticity
If you pay per second, every idle minute is waste and every autoscaling decision saves money. This is why hyperscaler architectures lean on autoscaling groups, serverless and scheduled shutdowns of non-production environments.
Commitments reward predictability
Reserved instances, Savings Plans and committed use discounts trade flexibility for price. Buy them for the baseline you are certain will run for the term, never for the peak. Google Cloud’s sustained use discounts are the exception: they apply automatically to long-running resources with no commitment.
Spot capacity rewards fault tolerance
Spot, preemptible and evictable capacity is the cheapest compute available, on the condition that your workload survives being interrupted. Batch processing, rendering, CI runners and stateless workers behind a load balancer qualify. Databases do not.
Monthly caps reward simplicity
On DigitalOcean, Linode, Hetzner and UpCloud, a server that runs all month costs its list price and never more. Budgeting is easy. The catch is the other direction: a server that exists but is powered off still bills, because its disk and address are reserved. Delete what you are not using.
Prepaid and renewal models reward attention to dates
Prepaid balances run out; auto-renewing services renew whether you remembered them or not. UpCloud offers automated top-ups, and OVHcloud lets you switch services to manual renewal. Put renewal dates in a calendar.
Beyond compute #
Compute is the headline, but three other lines shape the bill. Traffic: hyperscalers charge for outbound data; Hetzner includes a location-dependent allowance; UpCloud includes outbound traffic at no cost; DigitalOcean pools a per-Droplet allowance; Kamatera includes an allowance and bills extra per gigabyte. Storage: block volumes, snapshots and object storage tiers each have their own prices, and old snapshots are a classic silent cost. Add-ons: backups, floating IPs, load balancers and dedicated IPs are priced separately on most platforms.
Free tiers and credits #
Most providers offer a free tier, a time-limited trial credit, or both, and the two behave differently: a free tier is ongoing usage at no charge within limits, while a credit is a one-time balance that expires. Our guide to free trials and credits compares them across providers.
Worked example: one small server, five ways #
Consider a modest web server that runs 24 hours a day for a month.
- Hyperscaler on-demand: hours multiplied by the hourly rate, plus separate charges for the disk, the public IP and outbound traffic. Predictable only if you add those lines up.
- Hyperscaler with a commitment: the same server at a discounted rate because you promised to run it for a year; the disk, IP and traffic lines are unchanged.
- Developer cloud with a monthly cap: the plan’s list price, with a traffic allowance included. If you delete the server after ten days, you pay ten days of hours.
- Prepaid balance: the same hourly deduction from a balance you topped up in advance; the risk is the balance running out, not an unexpected invoice.
- Subscription or monthly prepaid: the month’s price paid at creation, cheaper per month than hourly if the server really does run all month, and no refund if you delete it early.
The lesson is that the same workload can be billed in materially different ways, and the right choice follows from how long the server will actually live and how much you value predictability over flexibility.
Frequently asked questions #
Reserved instances or Savings Plans on AWS?
Savings Plans are more flexible because the commitment is to spend rather than to a specific instance type. Reserved instances still make sense for capacity reservations and some services. Our AWS billing guide compares them.
Why did a powered-off server bill me?
On hourly-capped platforms, and for stopped instances with attached disks on hyperscalers, storage and addresses stay allocated. Only deletion stops the charge.
Is spot capacity safe for production?
For stateless, interruption-tolerant tiers, yes, with on-demand capacity as a floor. Not for stateful services.
