Hourly Billing VPS: A Burst Tool, Not a Discount

Hourly billing sounds like a way to pay less. It isn’t. It’s a way to pay for less time — and those are different products.

Every major provider that bills hourly also caps the total at a monthly maximum. Vultr, for instance, bills most server types up to 672 hours per month. A month averages about 730 hours, so once your instance has run roughly 28 days, the remaining days are free. Run it all month and you pay exactly the monthly price — the same as everyone on a monthly plan.

Which gives you the only formula that matters here:

Breakeven hours = monthly cap ÷ hourly rate

Below that number, hourly saves money. Above it, you’re on a monthly plan that happens to be described in hours. For Vultr the answer is stated outright: 672 hours, or 28 days. Keep a server longer than four weeks and hourly billing has saved you nothing at all.

So hourly is not a discount. It’s a tool for servers that don’t live long.

The trap: stopped is not the same as destroyed

This is where hourly bills surprise people, and it’s worth stating as bluntly as the providers do.

Vultr’s billing documentation is explicit, and worth quoting exactly:

“The minimum billing unit is one hour, and hourly charges begin when a server is deployed, regardless of whether it is powered on or off. Servers that are stopped but not destroyed continue to incur hourly charges for reserved resources such as compute, storage, and IP address allocations.”

To stop billing for a server, you have to destroy it. Stopping it does not suspend billing.

The sidebar tells its own story: “Billing for Stopped Instances”, “Stopped Server Charges”, “Snapshot Charges”. Vultr maintains several documentation pages on this because it catches people out constantly.

That is not a quirk of one provider — it’s the standard model, and the logic is sound. When you “stop” an instance, the provider is still holding your disk, your IP address and your slot on a host. Nobody else can use them. You’re paying for a reservation, not for electricity.

What typically keeps billing after you power something off:

“I’ll shut it down over the weekend to save money” saves far less than people expect — often nothing at all. The action that stops billing is destruction, which means you need your data somewhere else first.

Before you cancel anything, destroy the instances and the reserved IPs, volumes and snapshots. Orphaned resources sitting on a closed project are one of the most common sources of a bill nobody can explain.

Billing granularity varies more than the rate

Two providers can both say “hourly” and round very differently:

• Hetzner bills from around €0.0088/hour and rounds up to full hours. A server that exists for six minutes costs an hour.
• Vultr bills hourly against a 672-hour cap for Cloud Compute, Optimized Cloud Compute and Bare Metal, with GPU instances on a 730-hour model.
• DigitalOcean bills hourly against a per-plan monthly cap.
• RunPod bills per second, displayed as an hourly equivalent — the finest granularity in this comparison.

For a server you keep for weeks, rounding is irrelevant. For a build runner that lives eleven minutes, per-second billing versus hour-rounding is the difference between paying for eleven minutes and paying for sixty.

The providers

1. DigitalOcean — hourly with a documented cap

Droplets bill hourly with a monthly cap per plan, so a Droplet running the full month costs exactly the advertised monthly figure and one running for three days costs three days.

What makes DigitalOcean the default recommendation here isn’t the rate — it’s the surrounding tooling. Hourly billing is only useful if you can create and destroy servers programmatically, and DigitalOcean’s API, CLI and Terraform provider are the best in this comparison. Spinning up a Droplet from a script, running a job, and destroying it is a well-trodden path with documentation to match.

Remember to enable backups deliberately: they’re off by default and cost 20–30% extra, as we covered in the beginner VPS guide.

Check DigitalOcean pricing → · Read our DigitalOcean review

2. Hetzner — the cheapest hours you can buy

From roughly €0.0088/hour with no minimum contract, and if you delete a server before the end of the month you’re billed only for the hours it existed.

At these rates the arithmetic gets genuinely attractive for intermittent work: a CI runner that’s live four hours a day costs a fraction of a monthly plan, and Hetzner’s underlying monthly prices are already the lowest among mainstream providers. The catch is the rounding — full hours, rounded up — so very short-lived instances are less efficient here than on a per-second provider.

Note Hetzner’s 15 June 2026 price adjustment, and that its documentation warns changes to servers on legacy pricing may move them onto current rates.

Check Hetzner Cloud pricing → · Read our Hetzner review

3. RunPod — per-second billing

RunPod bills per second, shown as an hourly equivalent, which makes it the right choice when jobs are short and irregular. A job that finishes in eleven minutes costs eleven minutes.

It’s GPU-focused — RTX 4090 capacity runs $0.34/hour on Community Cloud and $0.69 on Secure Cloud — so it’s the answer for inference and training bursts rather than for hosting a website. Our GPU VPS guide works through the utilisation maths that decides whether hourly GPU rental beats a dedicated instance.

Check RunPod pricing → · Read our RunPod review

4. Vultr — the clearest published cap

Vultr documents its billing plainly: Cloud Compute, Optimized Cloud Compute and Bare Metal bill up to 672 hours per month, GPU instances up to 730 — which Vultr calls the industry-standard monthly cap — with snapshots at $0.05/GB per month and reserved IPs charged until released. Invoices generate on the 1st of each month.

Publishing the cap in hours rather than burying it is genuinely useful — it’s what let us state the 28-day breakeven at the top of this article with confidence. Combined with 30-plus regions, it’s a strong option for short-lived servers in specific locations.

We don’t have an affiliate relationship with Vultr, so there’s no tracked link here.

5. Kamatera — hourly with configurable specs

Kamatera bills hourly and lets you compose the machine rather than pick from fixed plans, which pairs well with hourly billing: you can size a box precisely for a job, run it, and destroy it.

Note from our free VPS audit that Kamatera’s 30-day trial requires a credit card and carries up to $2 in verification charges, despite frequently appearing on “no credit card” lists.

6. Cloudways — hourly on a managed platform

Cloudways bills hourly for its managed hosting across the cloud provider you choose, which is unusual in managed hosting — most managed hosts sell monthly plans only. If you want the platform’s staging, backups and support without a monthly commitment, this is close to the only option.

You are paying a managed premium over raw hourly rates, and the underlying instance characteristics come from whichever provider you select. Its billing documentation is worth reading before you commit — it splits the rules by underlying provider (DigitalOcean, Linode and Vultr behave differently from AWS and Google Cloud) and has separate sections for server-deletion billing and add-on billing, which are the two places managed-platform bills tend to surprise people.

Check Cloudways pricing → · Read our Cloudways review

7. Railway — usage-based rather than hourly

Railway bills on consumption rather than instance-hours, which suits applications that idle most of the time and burst occasionally. It’s a platform rather than a VPS — you deploy an app, not a server — so it belongs here only if what you actually want is to stop thinking about instances.

Check Railway → · Read our Railway review

When hourly genuinely wins

Use hourly for:

• CI/CD runners that exist for the length of a build.
• Batch jobs — encoding, data processing, scraping runs, model fine-tuning.
• Load testing, where you want ten servers for an hour and none afterwards.
• Evaluating a provider before committing. A day costs pennies.
• Staging environments you can genuinely destroy, not merely stop.
• Seasonal capacity — a few extra servers for a launch or a sale.

Don’t bother for:

• Anything running continuously. You’ll hit the cap and pay the monthly price regardless.
• Production websites. They run 24/7 by definition.
• “I’ll turn it off at night.” Unless you destroy and rebuild it — with the data elsewhere — you keep paying for reserved resources.

The honest summary: hourly billing is an automation feature, not a pricing feature. Its value comes from being able to create and destroy servers from a script. If you’re clicking buttons in a dashboard, you’ll never extract the benefit, and a monthly plan from a cheaper provider will cost less.

Worth knowing what sits at the opposite end: the budget hosts with the lowest headline prices generally require multi-year prepayment. Hostinger’s advertised VPS rates need a 24-month term, which is the polar opposite of hourly — as we set out in our under-$5 VPS comparison, the cheapest sticker prices and the most flexible billing almost never come from the same provider.

How we picked

We prioritised providers that publish their billing model — hourly rate, monthly cap, and rounding behaviour — since the cap is what determines whether hourly saves anything. Figures come from provider documentation and pricing pages read in August 2026, and the 672-hour cap is quoted from Vultr’s own billing documentation.

We have not audited billing behaviour by running instances and reconciling invoices across every provider, and behaviour around stopped instances varies by resource type. Before relying on “stop to save money,” check your specific provider’s documentation — and test it on one cheap instance for a month before scaling the approach.

Most links here are affiliate links. Note that the article’s central conclusion — that hourly billing saves nothing for the always-on workloads most readers actually have — argues against the more expensive cloud providers we’d earn more from.

FAQ

Is hourly billing cheaper than monthly?

Only if the server lives less than the cap. Providers bill hourly up to a monthly maximum — Vultr’s is 672 hours, about 28 days — so a server running all month costs exactly the monthly price. Hourly saves money on short-lived servers and nothing at all on permanent ones.

Do I stop paying when I shut down my VPS?

No. Stopped instances continue to bill for reserved resources at most providers, because your disk, IP and host slot are still held for you. Only destroying the instance stops the charges — and reserved IPs, snapshots and volumes bill separately until they’re deleted too.

Which VPS provider has the finest billing granularity?

RunPod bills per second. Most cloud providers bill hourly, and some round up to full hours — Hetzner does, so a six-minute server costs an hour. For very short jobs, per-second billing is materially cheaper; for anything measured in days, rounding is irrelevant.

What’s a monthly cap on hourly billing?

A ceiling ensuring you never pay more than the monthly plan price even if the hours would add up to more. Divide the cap by the hourly rate to get your breakeven in hours — beyond that point, additional hours in the same month are effectively free.

Can I use hourly billing to run a site only during business hours?

In principle, but rarely in practice. You’d need to destroy and recreate the server rather than stop it, which means the data must live elsewhere — a managed database or object storage — and something must orchestrate the rebuild. For a small site the engineering costs more than the savings.

Does Hostinger offer hourly billing?

No. Hostinger’s VPS pricing is built around long prepaid terms, with the best rates requiring 24 months. It’s an excellent product at what it does, but it sits at the opposite end of this spectrum — if billing flexibility is your requirement, look at Hetzner, DigitalOcean or Vultr instead.