AWS Lambda MicroVMs Pricing

AWS launched Lambda MicroVMs on June 22, 2026. If you're running AI-generated code, untrusted user scripts, or ephemeral CI environments, AWS Lambda MicroVMs pricing works differently from standard Lambda. Unlike Lambda Functions, which charge per request and execution duration, Lambda MicroVMs are priced per instance-second. That difference matters for your bill.

A MicroVM can suspend when idle, costing zero compute, and resume in 1-2 seconds when you need it again. This flips the cost equation for bursty workloads.

TL;DR:

• Running: $0.0997/vCPU-hour + $0.0132/GB-hour (ARM/Graviton, US East)
• Suspended: $0 compute, $0.08/GB-month for snapshot storage
• Snapshot read/write and data transfer add to the base cost

What Are AWS Lambda MicroVMs?

Lambda MicroVMs are Firecracker-powered sandboxes between Lambda Functions (stateless, 15-minute max) and EC2 instances (full VM, you manage everything). Each MicroVM gets its own isolated Linux kernel, up to 8 hours of lifecycle, and a dedicated HTTPS endpoint. You launch it with state: memory, disk, running processes. It suspends when idle. Billing stops instantly. Resume it later with everything intact.

This model handles scenarios traditional serverless cannot: running untrusted code safely (AI agents, user scripts), maintaining session state across requests, or spinning up isolated environments for each customer in a multi-tenant setup. The isolation is real. A bug in one MicroVM cannot touch another because each has its own kernel.

How AWS Lambda MicroVMs Pricing Works

AWS Lambda MicroVMs pricing is metered in three dimensions: compute when running, storage when suspended, and data transfer.

Compute charges (per second when running):

• $0.0000276944/vCPU-second ($0.0997/vCPU-hour)
• $0.0000036667/GB-second ($0.0132/GB-hour)

These rates are for ARM (Graviton) in US East. x86 costs more, and pricing varies by region.

Suspend charges (zero compute, storage only):

• $0.08/GB-month for snapshot storage
• Snapshot read/write operations on resume (charged separately)

Data transfer:

• Standard AWS egress rates (varies by destination)
• Data between MicroVM and your VPC incurs additional charges

The Suspend State Economics

The real advantage of MicroVM pricing is suspension. Say you run a 2-vCPU, 4-GB MicroVM for your AI agent platform.

Running 24/7:

• vCPU: 2 × $0.0997/hr = $0.1994/hr
• RAM: 4 × $0.0132/hr = $0.0528/hr
• Total: ~$18.25/day or $547/month

Running 1 hour/day, suspended 23 hours:

• Running: 1 × ($0.1994 + $0.0528) = $0.2522/day
• Snapshot storage: 4 GB × $0.08/month ≈ $0.01/day
• Total: ~$7.63/month

The difference is extreme. Bursty workloads keep the MicroVM paused most of the time. Agents, CI/CD pipelines, and sandboxed inference endpoints benefit most.

Comparing MicroVM Pricing Across Platforms

AWS is not alone in the sandbox space. As of June 2026, Daytona, E2B, Modal, and Northflank all offer similar services. The pricing models differ significantly.

AWS Lambda MicroVMs (2 vCPU, 4 GB, US East):

• Running: ~$0.126/hour
• Suspended: $0/hour (snapshot storage only)

Daytona:

• ~$0.083/hour, billed per second, no snapshot fees

Northflank:

• $0.01667/vCPU-hr + $0.00833/GB-hr, no snapshot overhead

E2B:

• Similar to Daytona, billed per second, no separate storage tier

AWS is more expensive per running hour but competitive for workloads that suspend frequently. You pay for snapshot storage. If your agent sleeps 20+ hours per day, the net cost often undercuts always-on alternatives like EC2, which charge fixed hourly rates regardless of utilization. For always-on services (a web API, a database proxy), per-minute microVM platforms or reserved EC2 instances are more economical. For ephemeral, bursty workloads (AI agents, CI jobs, sandboxed code execution), Lambda MicroVMs can be the right choice if you're already in the AWS ecosystem.

When AWS Lambda MicroVMs Make Financial Sense

Use Lambda MicroVMs when:

• You run AI agents or untrusted code and need VM-grade isolation
• Workloads are bursty: active for minutes, idle for hours
• Sessions last up to 8 hours and must retain state
• You're building multi-tenant platforms
• You have high transaction volume needing isolated environments

Avoid Lambda MicroVMs when:

• Your service runs 24/7 (always-on compute is cheaper elsewhere)
• You need more than 8 hours of session state
• Snapshot storage becomes prohibitive
• Data transfer, VPC connectivity overhead adds significant cost

The billing model punishes idle VMs that stay warm but unused. If a MicroVM is running but receiving no requests, you're still charged for compute. Suspension, automatic when idle, is critical to the cost math.

FAQ

What's the Baseline Charge for a Lambda MicroVM?

The minimum charge is per-second billing for vCPU and RAM while running, plus $0.08/GB-month for snapshot storage. There is no minimum invocation charge or fixed platform fee.

How Much Does a Lambda MicroVM Cost per Second?

A 2-vCPU, 4-GB MicroVM costs $0.0000276944/vCPU-second and $0.0000036667/GB-second, totaling roughly $0.00015/second (~$0.54/hour) for that configuration in US East.

Does AWS Lambda MicroVMs Pricing Include API or Data Transfer Costs?

No. AWS Lambda MicroVMs pricing covers only compute and snapshot storage. You pay separately for data transfer out of AWS (egress) and data exchanged with your VPC, at standard AWS rates.

How Do I Launch a Lambda MicroVM and When Does Billing Start?

Use the run-microvm command with the AWS CLI or API, pointing to a pre-built image. Billing begins when the MicroVM starts and stops when you suspend or terminate it. Suspended MicroVMs cost zero compute but continue incurring snapshot storage charges.