How to stop a surprise DigitalOcean bill

Why a powered-off Droplet still costs money

Powering down a virtual machine in the control panel feels like it ought to pause the meter. It doesn't.

When you shut down a Droplet from the command line or click Power Off in the web console, you're only halting the guest operating system. The hypervisor still holds your RAM allocation, keeps your dedicated disk storage intact, and retains your reserved IP address. Because those physical host resources stay locked and can't be assigned to anyone else, DigitalOcean keeps billing you for them every second the Droplet exists.

That rule applies across standard bundled instances and high-end hardware. If you spin up an on-demand GPU Droplet - which runs from $0.76 per hour for an NVIDIA RTX 4000 up to $35.76 per hour for an 8x NVIDIA H200 - the GPU and CPU resources stay reserved on the hypervisor while powered off. Leaving an 8x H200 off over a three-day weekend costs over $2,500 without running a single calculation.

To actually stop the charges, you have to destroy the Droplet. If you want to keep the configuration for later without paying the full compute rate, take a snapshot first, destroy the active instance, and spin up a new Droplet from the snapshot when you need it again. Droplet Snapshots cost $0.06/GB per month, which is far cheaper than keeping compute cores reserved. Other providers handle idle state similarly; you can see our breakdown of why stopped Vultr instances still cost money for a direct comparison.

How Droplet pricing and monthly caps work

Effective January 1, 2026, DigitalOcean moved to per-second billing for Droplets, with a minimum charge of 60 seconds or $0.01, whichever is higher. If you spin up a server via a script and tear it down 15 seconds later, you still pay for the full minute or the penny minimum.

The way DigitalOcean caps your monthly invoice depends entirely on the Droplet architecture you pick. Bundled-plan CPU Droplets have a built-in monthly ceiling capped at 672 hours (exactly 28 days). If your bundled Droplet runs for 720 hours in a 30-day month, you only pay for the first 672 hours.

Newer v5 Droplets don't work like that. They do not have a 672-hour monthly usage cap, so a continuously running v5 Droplet costs whatever the actual calendar hours add up to. Instead of a flat bundle, v5 pricing is calculated per second based on the sum of configured resources:

• Shared vCPU: $0.0158219/hour per vCPU
• General Purpose vCPU: $0.028/hour per vCPU
• Memory: $0.0040411/hour per GiB
• Boot disk: $0.000137/hour per GiB

For example, an s5-8vcpu-16gb-30gb setup runs 8 shared vCPUs ($0.1265752/hour), 16 GiB of memory ($0.0646576/hour), and a 30 GiB boot disk ($0.004110/hour), adding up to roughly $0.1953 per hour. Public IPv4 addresses are included for free on bundled plans, whereas v5 Droplets show IPv4 addresses as a separate invoice line item currently priced at an introductory rate of $0.00/hour. Private Droplets don't carry extra charges beyond their baseline compute rate.

To see all available sizes and their underlying slugs directly from your terminal, run doctl:

doctl compute size list

You can query the identical dataset using the REST API endpoint /v2/sizes.

Setting up DigitalOcean spend alerts

DigitalOcean spend alerts send email notifications when spending across your team or organization crosses a specific percentage threshold of a monthly budget you define.

These alerts are not hard spending caps. With one exception for AI inference keys, an alert will not stop your running servers, restrict disk expansion, or pause network transfer.

Only users with the owner or biller role can create or modify spend alerts. Here is how to configure them depending on your workspace setup:

• For standard teams: Open the control panel, go to the Billing page, and click the Spend alerts tab.
• For organizations: Click your profile icon in the top right of the Control Panel, select View Organizations, pick your organization, click the Billing tab, and then click Spend alerts.

When configuring an alert, you choose what to track:

1. Total spend: Tracks all cumulative usage within the alert's scope.
2. Specific products: Measures combined spend across only the products you check. The interface starts with every product selected; click Reset scope to clear all checkboxes and pick specific tools like Droplets, Spaces, or Databases. If you want separate alerts for individual products, you must set up separate alerts for each one.
3. Managed Access Keys (Serverless Inference): Monitored in public preview for up to 10 model access keys (legacy keys aren't supported). This is the only spend alert on DigitalOcean that can enforce a hard stop. Turning on Pause Serverless Inference at threshold pauses each key independently the moment its spend hits 100% of the budget. Paused keys resume automatically on the first day of the next billing cycle, or immediately if you raise the budget, turn off pausing, remove the key, or delete the alert.

If you had an older billing alert before the spend alert overhaul, DigitalOcean automatically migrated it under the name Account Spend Alert with a single threshold at 100% of your previous value.

Alert timing, invoice thresholds, and auto-charges

Spend alert emails are sent to team or organization owners and billers within an hour of your actual spend crossing a threshold. DigitalOcean doesn't trigger alerts based on future projections. Because of this, projected outbound Droplet transfer overages won't fire an alert until the actual excess charges hit your invoice at the end of the month.

Each threshold triggers an email notification once per billing cycle. If a traffic burst causes your usage to leap across multiple thresholds between checks, DigitalOcean only sends an email for the highest threshold crossed. If you modify an alert's budget or adjust its percentage thresholds mid-month, the tracking resets, which allows notifications to fire again if spend crosses those updated targets.

Keep these billing nuances in mind:

• Credits and discounts: Team spend alerts and organization-wide alerts measure spend after promotional credits and discounts are deducted. However, an organization alert scoped down to a single team measures spend before credits, because credits are applied to the parent invoice.
• Mid-month auto-charges: DigitalOcean charges your payment method automatically on the first of each month for the prior month. But if your accrued usage hits your account tier's auto-charge amount mid-cycle, DigitalOcean attempts to charge your card immediately. If that charge fails, your account's provisioning permissions are frozen until it's resolved. For newer accounts spinning up GPU Droplets, DigitalOcean may charge the card partway through the month as an abuse prevention check.
• Data transfer pooling: Basic regular plans include free outbound transfer allowances starting at 500 GiB on the $4/month box. All bandwidth pools at the team level across your Droplets, inbound data is always free, and traffic dropped by cloud firewalls or routed over private VPC interfaces doesn't consume your quota. Excess outbound transfer costs $0.01 per GiB.
• Backups: Percentage-based backups add 20% to your Droplet cost for weekly runs or 30% for daily backups. Usage-based backup options start at $0.01/GiB per month.
• Refund policy: DigitalOcean does not offer refunds. If you leave a test server running by accident, you have to contact support to plead extenuating circumstances.

To dig into what is driving current spend, the Insights tab on the Billing page provides a "Your usage" graph that lets you filter and group line items by product, project, tag, model, token type, and model access key. For Serverless Inference, Live Usage refreshes every 10 minutes.

Catching runaway spend across all your infrastructure with Costbrake

DigitalOcean's built-in alerts help, but email notifications that take up to an hour to arrive won't help you catch a fast-moving runaway loop before it racks up significant charges. It also leaves you manually checking different dashboards if your stack spans multiple providers.

Costbrake monitors what you spend across cloud, AI and SaaS platforms like DigitalOcean, AWS, Google Cloud, Azure, Cloudflare, Vercel, OpenAI, and Anthropic. It connects to DigitalOcean using a read-only personal access token with billing:read and account:read scopes (your team role must be Owner or Biller) to pull month-to-date balance data directly.

Because DigitalOcean continues billing for powered-off instances and only destroying resources stops charges, Costbrake does not offer brakes on DigitalOcean; our killswitches never delete customer resources. But it gives you faster visibility:

• Rapid checks: Costbrake checks spend hourly on the Free plan and every 5 minutes on Pro and Team.
• Spike detection: Fires an alert if an hour costs more than 3x your 7-day baseline and adds at least $5 of unexpected spend.
• Month-end forecasting: Default budget alerts ping you at 50%, 80%, and 100%, and a forecast alert triggers when your month-to-date spend plus your last 72 hours of burn would exceed your budget before the month ends. You can read more about tracking burn rates in our guide on how to forecast your cloud bill before the month ends.
• Multi-channel alerts: Sends notifications to Slack, Discord, and custom webhooks on Pro, or PagerDuty on Team, alongside standard email.

Plans start with Free at $0 (2 connections, 1 seat, hourly checks), Pro at $12 a month (10 connections, 3 seats, 5-minute checks), and Team at $39 a month. You can test Pro features on a 14-day trial without entering a credit card. To connect your environment, tell your AI coding assistant to handle the agent setup or hook it up manually in the console.