Most hosting providers send an alert when spending climbs, but few stop it. An alert tells you; only a hard limit or prepaid credit actually stops the meter.
Among the providers we cover, UpCloud runs on prepaid credit and shuts servers down when it runs out. DigitalOcean sends spend alerts, and says plainly that they are not a cap.
The facts below come from each provider's own documentation, read on 27 October 2026. Set your controls before an agent loops or a key leaks, not after.
Who this is for
- you rent GPUs or servers by the hour, or pay for an AI API by use
- you run agents that could loop or call tools without end
- you want to know which controls really stop spending
Not for you if you pay a fixed monthly price for a plain server: its bill does not grow with use.
Key numbers
- 10 iterations the default limit on how often n8n's AI agent runs the model for one request Official page,
- 1 month after credits run out before UpCloud removes servers and frees their IP addresses Official page,
- 2 months after credits run out before UpCloud removes storage, which is billed meanwhile Official page,
Three kinds of control
- A hard limit stops or pauses the service when spending reaches a set amount. It is the only control that protects you while you sleep.
- Prepaid credit means you pay first and use the balance. When it runs out, the service stops, unless automatic top-ups refill it.
- An alert sends an email or message when spending passes a threshold. It warns you, but the meter keeps running.
Prepaid credit and the API spending limit stop the bill; DigitalOcean's alerts only warn you.
| Provider | Hard limit | Prepaid credit | Alerts | Does stopping a server stop billing? |
|---|---|---|---|---|
| DigitalOcean | no, except pausing a Serverless Inference key | no | yes: spend alerts by budget, product or key | no: destroy the server to stop billing |
| UpCloud | in effect: servers shut down when credit runs out | yes, with optional automatic top-ups | yes: emails before credit runs out | IPs, storage and backups are still charged |
| Infomaniak AI Services | yes: a spending limit set by a billing user | credit-based billing | daily usage in the Manager | not a server; you pay per use |
| n8n (your own server or n8n Cloud) | per run: agent iterations and a workflow timeout | not applicable | an error workflow can notify you | not a server bill; it limits AI calls |
Step by step, provider by provider
DigitalOcean: set spend alerts, then destroy what you do not use
DigitalOcean's spend alerts email you when spending crosses a percentage of a monthly budget you set. You can track total spend, single products, or single Serverless Inference model access keys.
Its documentation is clear that a spend alert budget is not a spending cap. The exception is a model access key alert with pausing turned on, which pauses that key when the budget is reached.
Billing for a Droplet runs from creation until you destroy it. Powering off a GPU Droplet does not stop billing, because the GPU stays reserved, so destroy GPU servers when a job ends.
UpCloud: keep the balance small
UpCloud deducts usage from a prepaid balance every hour. When the credit runs out, it shuts down all servers, which makes a small balance a natural spending limit.
Automatic top-ups refill the balance at a threshold you choose, which removes that limit. UpCloud also sends no low-balance emails while automatic payments are on.
Two catches: IP addresses, storage and backups keep being charged after the shutdown, and the balance can go below zero. Servers are removed after a month without credit, and storage after two, losing the data.
Infomaniak AI Services: set a spending limit on the API
Infomaniak bills its AI API in credits, one per request or token. A user with billing rights can set a spending limit for the product, its guide says, and the Manager shows daily usage.
| Plan | Price | New-customer offer | Link |
|---|---|---|---|
| DigitalOcean GPU Droplet, 1 x NVIDIA L40S 48 GB GPU | $1.57/hour (about $1,146/month) on demand, billed per second · VAT not stated checked 25 Oct 2026 | New accounts get a $5 credit that lasts 90 days and covers every product except SaaS add-ons checked 29 Oct 2026 | Visit |
| DigitalOcean GPU Droplet, 1 x NVIDIA H100 80 GB GPU | $4.41/hour (about $3,219/month) on demand, billed per second · VAT not stated checked 25 Oct 2026 | New accounts get a $5 credit that lasts 90 days and covers every product except SaaS add-ons checked 29 Oct 2026 | Visit |
| UpCloud GPU Server, 1 x NVIDIA L40S, 8 cores, 64 GB 48 GB GPU | €1.11/hour (about €746/month; no charge after 28 days of use)30-day month: the last 2 days are free.31-day month: the last 3 days are free.February: no free days (one in a leap year).Started mid-month: it may never reach the cap that month. on demand, billed by the starting hour; monthly cost capped at 28 days of use checked 25 Oct 2026 | No offer | Visit |
| UpCloud GPU Server, 1 x NVIDIA H100, 12 cores, 240 GB 80 GB GPU | €2.17/hour (about €1,458/month; no charge after 28 days of use)30-day month: the last 2 days are free.31-day month: the last 3 days are free.February: no free days (one in a leap year).Started mid-month: it may never reach the cap that month. on demand, billed by the starting hour; monthly cost capped at 28 days of use checked 25 Oct 2026 | No offer | Visit |
Monthly costs for hourly prices assume the server runs all month (730 hours), or less where a provider caps the monthly bill.
Billed by the second or hour while the server exists, not only while it works.
| Plan | Price | New-customer offer | Link |
|---|---|---|---|
| Infomaniak AI Services: swiss-ai/Apertus-v1.5-70B | €0.70 / €2.50 per 1M tokens, input / output · pay as you go, no commitment; billed at the end of each month checked 25 Oct 2026 | 1M free credits to start with AI Services, then pay as you go checked 29 Oct 2026 | Visit |
| Infomaniak AI Services: mistralai/Mistral-Small-4-119B-2603 | €0.20 / €0.75 per 1M tokens, input / output · pay as you go, no commitment; billed at the end of each month checked 25 Oct 2026 | No offer | Visit |
Prices per million tokens; set the spending limit before you share an API key.
Inside n8n: cap AI steps and stop loops
An agent that calls tools in a loop spends on every call. n8n has three settings that stop it:
- Max Iterations on the AI Agent node: how many times the model may run for one request. It defaults to 10.
- Timeout Workflow in the workflow's settings: cancels a run after a set time. n8n Cloud enforces a maximum per plan.
- Error Workflow: another workflow that runs when this one fails, for example to send you a message.
Pair these with a spending limit at the AI provider, so a loop that slips through still has a ceiling. Our guide to what OpenClaw and Hermes Agent cost a month shows where agent costs come from.
Stopped is not always free
- DigitalOcean bills powered-off Droplets and GPU Droplets until you destroy them.
- UpCloud keeps charging for IP addresses, storage and backups after it shuts servers down.
- Snapshots, backups and storage volumes are billed separately; check your provider's list of what keeps charging.
Before you leave, list what still exists, not what is running. On an hourly GPU cloud, the cheapest server is the one you destroyed.
If a key leaks: the first ten minutes
- Revoke the key in the provider's control panel and create a new one.
- Check the usage or billing page for spending since the leak.
- Destroy any servers or resources you did not create.
- Search your code and chat history for where the key appeared, and remove it.
- Set a spending limit or alert now, if you had none, and contact the provider's support about unexpected charges.
Keep keys in environment variables or a secrets file, never in code you share. Our n8n security guide and OpenClaw security guide cover this for agent servers.
How we checked
We read DigitalOcean's, UpCloud's, Infomaniak's and n8n's own documentation on 27 October 2026. Each statement in the table is as the provider documents it.
Our own measurements will be added here, with their dates, when our test bench runs them.
Mistakes to avoid
Treating an alert as a limit
Instead: An alert emails you; the meter keeps running.
Powering off a GPU server to save money
Instead: On DigitalOcean it is still billed; destroy it.
Turning on automatic top-ups without a cap in mind
Instead: They remove the natural limit of prepaid credit.
Leaving an agent without an iteration limit
Instead: Keep n8n's Max Iterations low and set a workflow timeout.
Other questions
Can I set a hard spending limit on DigitalOcean?
Not for the account: its documentation says spend alerts are not a cap. A Serverless Inference model access key can be paused when its alert budget is reached.
Does UpCloud stop charging when my credit runs out?
It shuts down all servers, but IP addresses, storage and backups keep being charged, and the balance can go below zero.
How do I stop an n8n AI agent from looping?
Keep Max Iterations on the AI Agent node low (it defaults to 10) and set a workflow timeout in the workflow's settings.
Sources and test notes
- DigitalOcean documentation: spend alerts, read
- DigitalOcean documentation: Droplet pricing and billing, read
- UpCloud documentation: account balance, read
- UpCloud documentation: getting started FAQ (prepaid system, automatic payments), read
- Infomaniak: getting started with AI Services (billing and spending limit), read
- n8n documentation: AI Agent node, Tools Agent (Max Iterations), read
- n8n documentation: workflow settings (timeout, error workflow), read
Every page read on 27 October 2026. Each control is as the provider documents it.