DigitalOcean MicroVM Pricing: What You Pay While Paused

Check the preview conditions and the original launch date

DigitalOcean announced the MicroVM public preview on October 9, 2026. Built on Firecracker, it runs a container image inside a lightweight VM with its own kernel, for workloads such as AI agents started per session. This guide checks current pricing and requirements on November 11; it does not present the earlier announcement as a launch today.

The current limits page lists only MKC1 and linux/amd64 images. Do not assume an ARM64 image from your local machine will work. Some sizes depend on your team’s current plan; check your own options response for available sizes and the team count limit. The public preview is not recommended for production workloads, and its availability does not establish that every account can use every size.

Pausing stops compute charges; retained state still costs money

Compute has separate vCPU and memory components, accruing per running second: $0.022/vCPU-hour plus $0.0085/GiB-hour. The pricing page explicitly excludes time in the creating, pausing, resuming and paused states from compute charges.

Disk storage costs $0.05/GiB-month of actual data used, rather than the full disk capacity listed for a size. Pausing preserves memory, files and running processes; it does not make the retained data free.

A checkpoint is a saved state copy from which another instance can be created. Its stored memory and disk also cost $0.05/GiB-month until deleted. Creating a checkpoint has no separate fee. Destroying a MicroVM stops its disk storage charges, while retained checkpoints continue to accrue storage charges and must be cleaned up separately.

A hypothetical example separates runtime from retention

Suppose a 2-vCPU, 4-GiB MicroVM runs for a cumulative 15 minutes and remains paused otherwise. Its compute rate per running hour is 2×$0.022 + 4×$0.0085 = $0.078, so those 15 minutes cost $0.0195 in compute. This is a hypothetical calculation from published rates, not a measured invoice.

Now assume its disk averages 10 GiB of actual data retained for a full billing month. Disk storage adds 10×$0.05 = $0.50. A separate checkpoint averaging 6 GiB for the same full month adds $0.30. Neither storage charge is included in the $0.0195 compute figure.

If public networking also produces 2 GiB of outbound traffic, add $0.02 at $0.01/GiB. These assumptions yield a $0.8395 subtotal, excluding any NAT gateway, model API, other services, taxes, account discounts or final invoice rounding. Recalculate when stored data or retention changes; this example is not a fixed monthly price.

Auto-pause is mandatory, and paused instances consume quota

A MicroVM automatically pauses after its idle timeout. With auto-resume enabled, the next request triggers a resume, continuing its retained memory, files and processes. In the current preview, auto-pause cannot be disabled and the maximum idle timeout is four hours. Auto-resume can be disabled only at creation through the API or doctl; the Control Panel has no setting for it.

Your team count limit includes running and paused MicroVMs. Pausing an instance does not release that quota. An instance created from a checkpoint must use the checkpoint’s region and size and inherits its environment variables.

After creation, you cannot change the image, size, disk, idle timeout, environment variables, tags, ports, networking, region or auto-resume. Check these choices in an eligible test environment you control. This article did not create, pause or resume real resources or measure resume latency.

Distinguish public egress from the authenticated HTTPS endpoint

Public networking is the default, with outbound traffic using a shared NAT gateway. Public outbound traffic costs $0.01/GiB with no free allowance. This does not provide a dedicated or fixed public IP; MicroVMs currently do not support their own public IP address, IPv6 or Cloud Firewalls.

For VPC networking, the MicroVM must join a VPC owned by your team in the same region. It has no internet access unless that VPC has a NAT gateway, billed at standard rates. A private network does not include free public internet access.

Each MicroVM’s HTTPS endpoint reaches only the HTTP port selected at creation and supports HTTP/HTTP2. Although the API and doctl accept up to five declared ports, the others are not reachable during the preview. Every endpoint request needs a DigitalOcean API token with microvm:access. That scope reaches every MicroVM on the team and cannot be restricted to one instance. Do not include tokens or credential-bearing requests in public support posts.

Before starting, check costs against your own team options

First check current rates on the official pricing page. Then the credential owner can inspect GET /v2/microvms/options or doctl compute microvm options in their own environment. The resource_pricing and sizes fields give current resource rates and size prices; options also expose available team sizes and count limits. This article did not access a private dashboard or send an account API request.

Record expected running seconds, actual disk usage, checkpoint sizes and retention, outbound GiB, and whether a separately billed VPC NAT gateway is required. Reconcile costs against state and cleanup records: one successful request does not establish that a session has ended or that all storage has been deleted.

Network routing does not grant account eligibility, expand regions or quotas, or change DigitalOcean’s or a model provider’s pricing and rules. Check model costs and actual account offers separately. This guide does not verify any account’s free credits, final invoice or production suitability.