Billing

Invoices, budgets, alerts, tags, closing an organization.

What it is

What your organization is spending, what it owes, and the caps that warn you before a month gets away from you. Find the current period under Billing, issued documents under Billing → Invoices, caps under Billing → Budgets, and settlement under Billing → Payment methods. Everything is priced in EUR and metered per resource, so a cost on the invoice traces back to the machine that caused it.

How billing works, in short

  • You pay for what you use, by the hour. A machine costs its hourly price while it exists, and never more than its monthly price in a calendar month. Stored data is charged by size. The create forms show the monthly figure before you create anything.
  • Nothing is charged when you sign up. There is no subscription.
  • One invoice a month. At the start of each month an invoice for the month before appears under Billing → Invoices.
  • You pay by bank transfer, to the bank details on the invoice. Card and direct debit are not offered yet.
  • Prices are net. VAT is added on the invoice: German VAT, or reverse charge for a company with a valid VAT ID from another EU country.
  • A budget warns you. Billing → Budgets emails you before spending passes an amount you choose.
  • If an invoice stays unpaid, it is marked overdue and the organization cannot be closed until it is paid.

Concepts

Term Meaning
Usage record One metered event: a resource, a SKU, a quantity and unit, a unit price in EUR, and the billing period it falls in.
SKU A priced line in the catalog, such as an instance-hour or a stored gigabyte-month. Prices are published at /pricing.
Billing period A calendar month, written YYYY-MM. Usage carries the period it happened in; invoices are issued per period.
Invoice The document for one period. It snapshots your billing address, VAT number and VAT treatment as they stood at issue.
Line item One row on an invoice: description, SKU, quantity with unit, unit price, and net amount.
VAT treatment Which VAT rule the invoice was issued under. Fixed at issue and never restated afterwards.
Billing account Your organization's financial record: currency, address, VAT number, payment status and balance.
Infrastructure mode managed — machines run in the platform's provider project and you are billed for them; byo — machines run in your own project on your token.
Budget A cap in EUR with a period and a scope, plus the percentages at which it emails you.
Scope What a budget watches: the whole organization, one project, or a set of tag clauses.
Alert A threshold that has fired, recorded with the spend, the cap and the number of resources in scope at that moment.
Tag A key and value on a resource. Tags decide which project and which budget a cost lands in.
Project A saved tag query. Budgets and the inventory can be pointed at one.
Blocker Something standing between your organization and closure: an unpaid invoice, or a resource still running or storing data.

How to

See what you are spending now

Billing shows the running total for the current period, broken down by service and resource, or:

GET /billing/api/summary

Read an invoice

Billing → Invoices lists them newest first with status, total, due date and a PDF; open one to see its line items, the party details captured at issue, and the VAT note. An invoice is issued on the 1st of each month for the month before and is payable within 14 days. The API returns the list and each issued invoice as an A4 PDF:

GET /billing/api/invoices?limit=50
GET /billing/api/invoices/{id}/pdf

A draft has no number yet and no PDF.

Set who invoices are made out to

Billing → Billing account holds the company name, invoice address, VAT ID (USt-IdNr.) and invoice e-mail. They were entered at sign-up; a change applies to invoices issued from then on, and issued invoices keep what they said.

PATCH /billing/api/account
{ "companyName": "Müller Maschinenbau GmbH", "addressLine1": "Hauptstraße 1", "postalCode": "80331",
  "city": "München", "country": "DE", "vatNumber": "DE123456789", "billingEmail": "rechnungen@mueller.example" }

A VAT ID is checked for its country's format as you type and against the EU VIES service when you save. One VIES does not recognise is refused. When VIES cannot be reached, the ID is saved and shown as "Not checked yet"; until a check succeeds, invoices treat it as no VAT ID (no reverse charge). The invoice e-mail is recorded for invoice e-mails, which are not sent yet.

Pay an invoice

Every invoice is paid by bank transfer within 14 days, quoting the invoice number as the reference. Billing → Payment methods and each open invoice show the account to pay into. There is nothing to add or set up; card and SEPA direct debit are not offered yet.

Create a budget

Billing → Budgets → Create budget. Name it, set the cap in EUR, pick the period, choose what it watches, and list the addresses to mail. A budget's months, quarters and years start at midnight German time, the same boundary as the invoice, so usage at 00:30 on the 1st counts towards the new month in both:

POST /billing/api/budgets
{
  "name": "Platform team",
  "amountCents": 50000,
  "period": "monthly",
  "scope": { "type": "tags", "clauses": [{ "key": "team", "values": ["platform"] }] },
  "alertThresholds": [50, 80, 100],
  "notificationEmails": ["ops@example.com"]
}

Leave scope out and the budget watches the whole organization. Add serviceFilter to count only usage from the services you name.

See the price before you create something

Every create form (instance, volume, NAT gateway, autoscaling group, database, cache, file system, Kubernetes node group, load balancer, app, VPN gateway, function, queue, bucket) shows what it will cost: per hour, per month with the monthly cap, and the cap. The same estimate is one API call, and --dry-run prints it for any CLI command:

POST /billing/api/estimate
{ "operation": "rds.instances.create", "body": { "instanceClass": "performance-2", "topology": "cluster" } }
$ dcloud ec2 instances run --server-type performance-1 --image-id 114690387 --region-id de1 --dry-run
POST https://…/ec2/api/instances
…
Cost: 11,99 €/month max · 0,0192 €/h (excl. VAT)
  1 × Instance (performance-1, 1 vCPU, 2 GiB): 0,0192 €/h, 11,99 €/month (monthly cap 11,99 €)

operation is the API operation id the CLI and MCP use. Prices come from the SKUs the meter rates usage against. Lines marked metered, not charged yet (billed: false; today: network load balancers lb11–lb31, elastic IPs, NAT gateways, file systems, database storage and backups, snapshots, object storage) are recorded and shown at what they would cost, but not on your invoice until billing for them is switched on; the estimate's eurPerMonth keeps them out and notBilledEurPerMonth carries them. A create form's price shows everything you configure, with the part not charged today under it. Pay-per-use services list their unit prices instead. Allowed with billing:EstimateCost, or with the operation's own action. The MCP server's dry-run plans carry the same estimate.

See every price and what you used this month

Pricing (in the organization menu) is the one price list: every machine type by family and price, marked when it is sold out; then every service with the price of each thing it meters (a service that runs on machines costs the machine price of the size you choose); and an estimate for any mix of them, with VAT, a link that shares it and a CSV. Every line of an estimate is at its list price, and the total says how much of it is not charged today.

Billing › Usage lists the same meters with what your organization used and paid so far this month. What is metered and not charged yet shows what it would have cost. Both pages are generated from the meters each service declares, so they cannot leave one out.

Three statuses: Charged — on the invoice at the listed price. Metered, not charged yet — the price is set and usage is recorded, but it is not billed yet. Preview, not charged — the service is in preview, costs nothing now, and its price may still change. In both cases you will be told before charging starts.

See what one resource costs

The overview of an instance, database, cache or VPN gateway has a Cost card: this month so far, last month, what it is charged per hour now ("€0/h while stopped"), and each machine's progress towards its monthly cap. Once a machine reaches the cap, the rest of the month costs nothing and the card says so.

GET /billing/api/resource-cost?kind=instance&id=i-4d99096e

kind is instance, database, cache, vpn or volume. Needs billing:GetCostAndUsage; without it the card is not shown.

Let a budget stop non-production resources

A budget can propose stopping what it is paying for that nobody needs in production. Add an action:

PATCH /billing/api/budgets/{id}
{ "action": { "thresholdPercent": 100, "selector": [{ "key": "environment", "values": ["dev", "test", "staging"] }] } }

When the spend reaches thresholdPercent, the hourly sweep writes one proposal for the period: the running instances and databases, and the autoscaling groups, inside the budget's scope whose tags match selector (default environment in dev, development, test, staging, preview), each with the hourly charge it would stop. Untagged resources never match. A resource tagged budget-action=keep is never proposed. Kubernetes nodes are left to their node groups; autoscaling members are covered by scaling their group to 0 (min and desired), and the proposal records the sizes before. The budget's recipients get a mail with the list.

Nothing stops until someone approves. On Billing → Budgets, or:

GET  /billing/api/budgets/action-proposals
POST /billing/api/budgets/action-proposals/{id}/approve
POST /billing/api/budgets/action-proposals/{id}/dismiss
GET  /billing/api/budgets/{id}/action-preview      # what it would propose right now

Approving needs billing:ApproveBudgetAction. Each stop then goes through the service's own route with your credential: it also needs your ec2:StopInstances, rds:StopDBInstance or autoscaling:UpdateAutoScalingGroup on that resource, and is in the audit log as yours. Each resource gets a result: done, skipped (someone stopped it first) or failed (with the reason, e.g. 403). If you may stop none of them, nothing happens: the approval answers 403 and the proposal stays pending for someone who may. Otherwise a proposal is decided once; dismissing it closes the period, and an undecided one expires when the period ends. To undo, start the resources again, or set the group's sizes back to the previous values the proposal shows.

Get the monthly cost report

On the 2nd of each month at 08:00 German time, every organization that had usage gets last month's report by mail: the net total against the month before, the top services and resources, the biggest change, and every budget, with a link to that month in the cost explorer. It goes to the owner and the billing account's invoice e-mail (a budget's notification addresses get only that budget's alerts) unless you choose recipients on Billing → Reports & DATEV (which also holds the DATEV export settings):

GET  /billing/api/cost-report                       # settings, who it goes to, what was sent
PUT  /billing/api/cost-report   { "enabled": true, "recipients": ["finance@example.com"] }
GET  /billing/api/cost-report/preview?period=2026-10
POST /billing/api/cost-report/send { "period": "2026-10" }

Each month is sent once per organization; sending it again answers already_sent.

Check a scope before you save it

A scope that selects nothing looks exactly like a scope that is guarding something, until the month ends. Count what it would cover first:

POST /billing/api/budgets/scope-preview
{ "scope": { "type": "tags", "clauses": [{ "key": "team", "values": ["platform"] }] } }

This writes nothing. It needs tags:ListTags as well as billing:GetBudgets, because it reads tag membership.

See every budget with its current spend

GET /billing/api/budgets/status

Each entry carries the spend so far, the size of the scope, and the thresholds that have already fired this period. A budget whose scope currently selects no billable resource is reported as guarding nothing rather than as EUR 0.00, and one whose project has been deleted is reported as having a broken scope.

Change or remove a budget

PATCH /billing/api/budgets/{id} with any of the create fields; DELETE /billing/api/budgets/{id} removes it along with its fired alerts. Changing the cap or the scope does not un-fire an alert that already went out.

Tag a resource

Give tags at creation (--tags key=value on the CLI, the Project and tags section of every create form) or later; set the organization's required keys and its saved views under Tags & saved views in the console. See Tags, saved views & topology. A resource created without a required key is refused before anything is provisioned:

Missing required tags: env, owner

Point a budget at a project

Create the saved view under Tags & saved views, then give the budget { "type": "project", "projectId": "<id>" } (the API still calls a saved view a project). The budget then counts exactly the resources the view's tag query selects. Budgets by project are a next step.

Close the organization

Settings → Danger zone. There is no waiting period: an organization with nothing running and nothing owed closes at once.

DELETE /settings/api/organization
{ "confirmName": "<organization name>" }

The check runs at that moment, not from what the page last showed. Anything used this month is on a final invoice at the start of next month, as usual.

When something is still in the way, the settings page lists it, each row linking to the screen where it is removed. You can mark the organization as being closed while you work through the list: every console page then says so, and create forms ask before making something new.

POST /settings/api/organization/deletion
{ "confirmName": "<organization name>" }

DELETE /settings/api/organization/deletion cancels that.

Reference

How usage is metered

Every service declares its meters: what is measured, in which unit, how often, at which price, and whether it is charged. Machines (instances, database, cache, broker and VPN nodes, NAT gateways, file servers), load balancers and elastic IPs meter hourly, one record per machine (or address) per hour. Stored data meters daily by size held: volumes, database and file-system storage, backups beyond the provisioned size, snapshots and machine images, buckets (every version), container images, npm packages, Git, CI logs, artifacts and caches, monitoring logs and metrics. Outbound traffic above a machine's included quota meters daily from the provider's counter.

Metered, not charged yet

Meters added on 2026-11-16 (object storage, network load balancers, elastic IPs, snapshots and images, traffic above the quota, NAT gateways, file systems, database storage, backups, npm, CI storage, monitoring data) are recorded from the start but charged only once the installation's operator switches their group on. Until then each usage record carries what it would have cost (notChargedAmount) and charges €0: the cost explorer shows "not charged yet · would be €X", the invoice a €0 line "metered, not charged yet (would be EUR X)", and create forms the price marked the same way. Switching a group on charges from the next metered hour or day; nothing already metered is charged afterwards.

Where prices come from

Everything resold from Hetzner is priced from Hetzner's own price lists (GET /server_types and GET /pricing), read on every deploy by pnpm catalog:sync and turned into SKUs by pnpm pricing:seed:

  • A machine-hour (compute.*, rds.*, cache.*, vpn.*) is the server type's hourly price plus the public IPv4 address Hetzner bills with it, times the markup in src/lib/pricing/markup.ts. The markup is ×1.0 today for every machine price: you pay Hetzner's list price, and the cost explorer's "Compared with other clouds" shows Hetzner as its own row at that price. A machine without a public IPv4 pays the same. Where Hetzner's price differs between the locations Deutsche Cloud runs in, the dearer one applies; today Falkenstein and Nuremberg cost the same. The invoice and the cost explorer name the machine by our instance type, for example compute.performance-2.hour.
  • Machines are billed for every started hour and never more than their monthly price per calendar month (the billing period, German time), like Hetzner. The cap is per machine: a three-node database has three. It is Hetzner's monthly price for the server and its IPv4, times the markup (performance-1: €0.0192 an hour, at most €11.99 a month); the "per month" column on /pricing and the website is this cap. Every hour is still recorded; hours past the cap are charged what is left of it, then nothing. A usage record the cap reduced keeps its uncapped amount in listAmount, the invoice line says "capped at the monthly price (EUR x less)", and the cost explorer's usage lines show the cap and what it took off.
  • Block storage keeps its listed price (€0.06 per GB-month) while that stays above Hetzner's, and follows Hetzner's price up if it does not.
  • Network load balancers (lb11–lb31), elastic IPs, snapshots and traffic over the included quota are priced from Hetzner's list (lb.<type>.hour with Hetzner's monthly cap, network.floating_ipv4.month and network.floating_ipv6.month charged as an hourly share of the month, storage.snapshot.gb_month, network.traffic.gb). An application load balancer is lb.application.hour, €0.00684 an hour up to €4.99 a month, in both infrastructure modes (its fleet is run by Deutsche Cloud).
  • Object storage is €0.0063211 per GB-month of stored bytes, every version counted: Hetzner's price for storage beyond its included amount (€0.0087 per TB-hour, for a 31-day month). Requests and uploads are free. Database and file-system storage follow Hetzner's volume price, backups and snapshots its image price (backups up to one copy of the provisioned size are free). NAT gateways and file servers are machines like any other (server + IPv4, capped monthly). npm packages and CI data are at the object-storage price; monitoring logs and metrics €0.10 per GB-month.

A price change closes the old SKU row and opens a new one. Usage keeps the unit price it was recorded with, so existing invoices never change and a new price applies from the next metered hour. pnpm pricing:check fails when any SKU is below Hetzner's current price times the markup, and when the website's price list disagrees with the catalog (regenerate it with pnpm website:prices).

Unit prices carry eight decimal places, so a fraction-of-a-cent hour is not rounded to zero. Rounding to cents happens once, at the invoice line. Units in use are hour, minute, gb, gb_month, month and request.

In byo mode you pay your provider for compute directly, so compute SKUs are not charged; storage runs on platform infrastructure and is charged in both modes.

Invoice statuses

Status Meaning
draft Being assembled. It carries no invoice number yet.
issued Numbered and owed.
paid Settled.
overdue Issued, past its due date and unpaid.
void Cancelled. It keeps its number and stays readable.

Invoice numbers form an unbroken yearly sequence and are assigned only at issue.

VAT treatments

Treatment Meaning
domestic You and the platform are in the same country; domestic VAT is charged.
reverse_charge EU business with a validated VAT ID; no VAT is charged and you account for it.
oss_b2c EU consumer; VAT is charged at your country's rate.
export_outside_eu You are outside the EU; the supply falls outside EU VAT and no VAT is charged.

VAT number validation states are unchecked, valid, invalid and unavailable. unavailable means the check did not conclude; it is not the same as invalid.

Billing account statuses

active is normal, overdue means money is owed past a due date, suspended is set by hand by the platform operator, and closed follows closure of the organization.

Budget fields

Field Values Default
name 1–255 characters required
amountCents a positive whole number of EUR cents required
period monthly, quarterly, annual monthly
scope organization, project with a projectId, or tags with 1–10 clauses organization
serviceFilter service names; only their usage counts all services
alertThresholds whole percentages of the cap, 1–1000 [80,100]
notificationEmails addresses to mail when a threshold fires none
action { thresholdPercent, selector }: propose stopping non-production none

A tag clause is a key and the values that count for it. A resource carrying none of the clause keys is invisible to that budget — untagged is not the same as zero. Several budgets may cover the same resource; each is evaluated on its own.

Alerts

The sweep runs hourly, at five minutes past. A threshold fires once per occurrence of the period, the first time spend reaches it, and never again for that threshold in that period; the next period starts clean. The alert row records the spend, the cap, the number of resources the scope was selecting and a description of the scope, so what made it fire survives a later edit to the budget or the tags.

Notify status Meaning
sent The mail went to every address on the budget.
no_recipients The threshold fired and no address is set. The alert is still recorded.
failed Sending was attempted and did not succeed. The error is kept on the row.

Tags at creation

Tags are applied once, in this order, and the result is fixed:

  1. The tags you supply on the create request.
  2. The tags of the resource being created under, for every key still free — a database's tags reach the volumes made with it.
  3. For keys the organization marks as filled from the acting user, that user's username.

Inheritance is not a live link. Re-tagging a parent later does not reach down and re-tag what it already produced; change the child directly. A tag you set explicitly is never overwritten by an inherited one. The required-key check counts what the parent already carries as supplied.

What blocks closure

Closure is refused while anything is left running or storing data. Unpaid invoices — issued or overdue — block, and their outstanding total is shown. Resources block when they are live; records with nothing behind them do not. The list is ordered as a plan of work, by what depends on what:

Order Resource Note
1 DNS zones (delegated), then CDN distributions (enabled) Names resolve to machines that are about to go; the edge serves them too.
2 Apps An app owns its machine; deleting the app removes it.
3 Functions, enabled schedules, enabled uptime checks, status pages They keep running, invoking or publishing until deleted.
4 Email identities, queues A verified domain can still send; a queue holds its messages.
5 Auto scaling groups, Kubernetes clusters A group replaces deleted instances; a cluster owns its nodes and volumes.
6 Load balancers, client and site-to-site VPN, NAT gateways Running gateway machines.
7 Databases, caches, file systems, then their snapshots Snapshots outlive their source and keep billing for storage.
8 Instances, volumes, volume snapshots, elastic IPs, images Detach volumes before deleting them.
9 Buckets Every bucket blocks, empty or not.
10 Container repositories, npm packages, Git repositories Only ones holding images, versions or commits; Git last.
11 Secrets, keys, SSH key pairs Rotate anything reused elsewhere first.

Networks, security groups, key pairs holding only public halves, disabled schedules, checks and CDN distributions, access keys (listed under All resources), IAM users, policies and settings never block. Every service of the platform is on this list or states why it holds nothing to remove (src/lib/inventory/coverage.ts); a new service cannot be added without one or the other.

Closed means the organization's status is deleted, every membership is suspended, nobody can sign in to it and its access keys stop working.

What happens to your data after closure

30 days after closure, everything the organization still has is deleted: IAM users, groups, roles and policies, memberships and invitations, settings, firewall rules, DNS records, monitoring history, CI logs, artifacts and caches, Git attachments, registry leftovers, snapshot chunks and the organization's audit log. The 30 days are there so a closure made by mistake, or by someone who took over an owner's account, can still be undone by support.

What is kept: invoices, their line items, credits and the billing account (company name, address, VAT id, billing email) for the statutory ten years (§ 257 HGB, § 147 AO), the request to close (who and when), and the organization's id and name, which the invoices refer to. The deletion is recorded in Deutsche Cloud's own audit log: which organization, when, how many rows and objects.

Deleted data stays in the encrypted backups of the platform database until they expire (14 days of point-in-time backups, 30 daily dumps), and is not restored from them except to recover the whole platform.

A dispute or an authority's request can put an organization on hold; it is then not deleted until the hold is lifted.

API

Operation Permission
GET /billing/api/summary billing:GetBillingSummary
GET /billing/api/invoices, /invoices/{id}/pdf billing:GetInvoices
GET /billing/api/account billing:GetBillingAccount
PATCH /billing/api/account billing:UpdateBillingAccount
GET /billing/api/budgets, /budgets/{id}, /budgets/status, POST /budgets/scope-preview billing:GetBudgets
POST /billing/api/budgets billing:CreateBudget
PATCH /billing/api/budgets/{id}, DELETE /billing/api/budgets/{id} billing:UpdateBudget, billing:DeleteBudget
GET /billing/api/budgets/action-proposals, /budgets/{id}/action-preview billing:GetBudgets
POST /billing/api/budgets/action-proposals/{id}/approve, …/dismiss billing:ApproveBudgetAction (and each stop's own action)
POST /billing/api/estimate billing:EstimateCost, or the estimated operation's own action
GET /billing/api/resource-cost, /cost-report/preview billing:GetCostAndUsage
GET /billing/api/cost-report billing:GetCostReportSettings
PUT /billing/api/cost-report, POST /billing/api/cost-report/send billing:UpdateCostReportSettings
GET /billing/api/payment-methods billing:GetPaymentMethods
POST, PATCH …/{id}/default, DELETE …/payment-methods/{id} billing:ManagePaymentMethods
POST, DELETE /settings/api/organization/deletion iam:RequestOrganizationDeletion, iam:CancelOrganizationDeletion
reading the closure state on the settings page iam:GetOrganizationDeletion
DELETE /settings/api/organization iam:DeleteOrganization

There is no route that writes usage: the platform meters what it runs, at the catalog price for the moment the usage occurred.

What is not supported

  • Any currency other than EUR.
  • Prepaid credit or top-ups.
  • Invoicing a single user or team separately; an invoice covers the organization.
  • Automatic suspension for non-payment. The suspended status exists and is set by hand.
  • A budget that stops anything without a person approving it, or that deletes anything. Budget actions stop instances and databases and scale autoscaling groups; caches, apps and Kubernetes node groups are not touched.
  • Adding a card or a direct debit mandate. Bank transfer is the only payment method.
  • Invoices by e-mail, and e-invoices (ZUGFeRD / XRechnung). Invoices are in the console and as PDF.
  • Re-tagging a parent to change what it already created; tags inherit once, at creation.
  • Reopening a closed organization yourself. Within 30 days of closure, support can reopen it; after that its data is gone.
  • Exporting everything an organization holds in one go; export each service's data before you remove it.

Troubleshooting

Symptom Cause Fix
a tag-scoped budget shows "guarding nothing" no resource carries that tag yet tag the resources, or check the scope with scope-preview
a budget never alerts every threshold is above the spend, or no address is set lower a threshold, or add an address to notificationEmails
a budget alert exists but no mail arrived notify status no_recipients or failed add an address; check the recorded error on the alert
a budget reports a broken scope the project it points at was deleted point it at another project, or change the scope
Missing required tags: … when creating something the organization requires those keys supply them on the create form, or create under a parent that already carries them
a cost lands in the wrong project the resource's tags were fixed when it was created set the tag directly on the resource
closure is refused with a list unpaid invoices, or resources still running or storing data pay each listed invoice and remove each listed resource, then try again
Did this page help you?