Amazon Lambda Pricing

Amazon Lambda MicroVMs pricing is based on the compute resources you use, snapshot read/write operations and storage, and data transfer at standard Amazon Web Services rates.

Amazon Lambda MicroVMs eliminate the need to right-size each compute environment for peak activity. You configure a baseline by setting memory, and CPU is allocated in a 2:1 memory-to-CPU ratio — the default is 2 GB / 1 vCPU. During peak activity, your MicroVM can vertically scale up to 4x the baseline (up to 8 GB / 4 vCPU), with no action required on your part.

You pay for baseline compute resources while your MicroVM is running. When your workload consumes resources above the baseline, you are charged only for the active duration of the additional memory and vCPU consumed — not for the peak capacity. This means you can configure for your typical workload and let Lambda handle the spikes. Compute usage is billed per second. For details on sizing your MicroVMs, see creating a MicroVM image.

Graviton (ARM) compute rates — Amazon Web Services China (Beijing) Region, operated by Sinnet / Amazon Web Services China (Ningxia) Region, operated by NWCD:

Amazon Lambda MicroVMs uses snapshots to deliver near-instant startup and reduce idle costs. A MicroVM image is a snapshot of your application's pre-initialized state. Instead of cold-starting from scratch, your MicroVM launches from this image. During idle periods, you can suspend a running MicroVM for up to 8 hours, preserving its memory and disk state without paying compute charges. When traffic returns, the suspended MicroVM resumes rapidly, either automatically or programmatically. You are charged for snapshot storage, snapshot data read (on start or resume), and snapshot data written (on suspend).

Region: Amazon Web Services China (Beijing) Region, operated by Sinnet / Amazon Web Services China (Ningxia) Region, operated by NWCD

You are charged for data transfer at standard Amazon Web Services Data Transfer rates. You are also charged for data transferred between your Amazon Lambda MicroVMs and your Amazon VPC, as outlined in Data Transfer within the same Amazon Web Services China Region.

All examples below are calculated using the Amazon Web Services China (Beijing) Region, operated by Sinnet / Amazon Web Services China (Ningxia) Region, operated by NWCD, and use ARM (Graviton) rates. For each GB of memory configured, Lambda MicroVMs allocates vCPU in a 2:1 ratio (e.g., configuring 8 GB memory allocates 4 vCPU).

Let's assume you are building a platform that provides each developer with an isolated, sandboxed Linux environment for AI-assisted coding. Each environment runs in a Lambda MicroVM configured with a 2 GB memory and 1 vCPU baseline that can vertically scale to 8 GB memory and 4 vCPU for resource-intensive tasks like compilation and test execution. Your organization maintains 5 standard MicroVM images (Rust, Python, Node.js, Java, Go), each 2 GB in size. Every developer launches one environment per day from a 2 GB image. Over an 8-hour workday, each environment is used for 2.5 hours: using baseline resources provisioned for 75% of this period (2 GB & 1 vCPU) and vertically scaling to peak resources for 25% of this period (8 GB & 4 vCPU). During the remaining 5.5 idle hours of the workday the environment automatically suspends, preserving 2 GB of state i.e. working directory, process state, and file caches. Each environment experiences 6 suspend/resume cycles per day as developers context switch between coding and their other activities. With suspend/resume, you avoid paying for always-on compute during idle periods while preserving your complete working state. Developers pick up exactly where they left off without restarting their environment or re-initializing state. Your team has 100 developers, each using the development platform for 20 days per month. All amounts use ARM (Graviton) rates for the Beijing Region. Your monthly charges would be calculated as follows:

Monthly compute charges

To recap, 75% of compute time is spent at baseline resource consumption and 25% at peak resource consumption.

Peak usage seconds: 100 developers × 20 days × 2.5 hours × 25% = 4.5M seconds

• vCPU: 4.5M × 4 vCPU × ¥0.000209731 = ¥3,775.16
• Memory: 4.5M × 8 GB × ¥0.000027768 = ¥999.65
• Total: ¥4,774.81

Baseline usage seconds: 100 × 20 × 2.5 × 75% = 13.5M seconds

• vCPU: 13.5M × 1 vCPU × ¥0.000209731 = ¥2,831.37
• Memory: 13.5M × 2 GB × ¥0.000027768 = ¥749.74
• Total: ¥3,581.10

Monthly compute charges: ¥4,774.81 + ¥3,581.10 = ¥8,355.91

Monthly snapshot charges

Snapshot read/write:

• Monthly suspend/resume cycles: 100 × 20 × 6 = 12,000 cycles
• Snapshot write (suspend): 12,000 × 2 GB × ¥0.030998658 = ¥743.97
• Snapshot read (resume): 12,000 × 2 GB × ¥0.012293318 = ¥295.04
• Monthly launches: 100 × 1 × 20 = 2,000 launches
• Snapshot read (launch): 2,000 × 2 GB × ¥0.012293318 = ¥49.17
• Total snapshot read/write: ¥1,088.18

Snapshot storage:

• MicroVM image storage: 5 images × 2 GB × ¥0.596800000 = ¥5.97
• Suspended MicroVM: 100 × 2 GB × (5.5 idle-hrs × 20 days / 720 hrs) × ¥0.596800000 = ¥18.24
• Monthly snapshot storage: ¥24.20

Total monthly charges = ¥8,355.91 + ¥1,088.18 + ¥24.20 = ¥9,468.29

Monthly cost per developer: ¥9,468.29 / 100 = ¥94.68

Let's assume you are building a platform that runs security scans or CI/CD jobs on behalf of multiple end-users. Each job executes untrusted or semi-trusted code - dependency installation, build scripts, vulnerability scanners. For this reason, each job requires its own hardware-isolated environment. Traffic is unpredictable, so you previously maintained a large pool of idle environments on standby to meet latency requirements. With Lambda MicroVMs, environments launch fast from snapshots, so you no longer need an idle pool. Each environment is configured with an 8 GB memory & 4 vCPU baseline that can vertically scale to 32 GB & 16 vCPU for resource-intensive compilation or scanning phases. Jobs run to completion in a single shot; no suspend/resume is needed. Each job runs for an average of 10 minutes, with 10% of that time at peak (32 GB & 16 vCPU) and 90% at baseline (8 GB & 4 vCPU). You maintain a single standard MicroVM image (2 GB). With snapshot-based startup, you no longer pay for idle standby pools - each environment launches rapidly on demand and terminates when the job completes. Your platform runs 10,000 jobs per month. All amounts use ARM (Graviton) rates for the Beijing Region. Your charges would be calculated as follows:

Monthly compute charges

To recap, 90% of compute time is spent at baseline resource consumption and 10% at peak resource consumption.

Peak usage seconds: 10,000 jobs × 600s × 10% = 600K seconds

• vCPU: 600K × 16 vCPU × ¥0.000209731 = ¥2,013.42
• Memory: 600K × 32 GB × ¥0.000027768 = ¥533.15
• Total: ¥2,546.56

Baseline usage seconds: 10,000 × 600s × 90% = 5.4M seconds

• vCPU: 5.4M × 4 vCPU × ¥0.000209731 = ¥4,530.19
• Memory: 5.4M × 8 GB × ¥0.000027768 = ¥1,199.58
• Total: ¥5,729.77

Monthly compute charges: ¥2,546.56 + ¥5,729.77 = ¥8,276.33

Monthly snapshot charges

Snapshot read/write:

• Monthly launches: 10,000 jobs × 1 launch each = 10,000 launches
• Snapshot read (launch): 10,000 × 2 GB × ¥0.012293318 = ¥245.87
• Total snapshot read/write: ¥245.87

Snapshot storage:

• MicroVM image storage: 1 image × 2 GB × ¥0.596800000 = ¥1.19
• Monthly snapshot storage: ¥1.19

Total monthly charges = ¥8,276.33 + ¥245.87 + ¥1.19 = ¥8,523.39

Cost per job: ¥8,523.39 / 10,000 = ¥0.8523