The full review
What is AWS Lambda?
AWS Lambda is Amazon's serverless compute service. You supply code, declare what triggers it, and AWS handles provisioning, scaling, patching, and availability across multiple Availability Zones. Billing follows execution rather than uptime, and functions scale to zero when idle, so an unused endpoint costs nothing.
Lambda is no longer a single function runner. The product page now frames it as one service with several ways to run: Lambda Functions for event-driven code, Lambda MicroVMs for isolated sandboxes, Lambda Durable Functions for long-running workflows, and Lambda Managed Instances for steady high-volume compute.
Four compute shapes, one programming model
Lambda Functions remain the default path. AWS states each function scales up to 1,000 concurrent executions every 10 seconds until it hits the account concurrency limit. Memory runs from 128 MB to 10,240 MB in 1 MB increments and CPU scales proportionally, making memory sizing the primary performance dial.
The newer shapes fill real gaps. Lambda MicroVMs run user-generated or AI-generated code in sandboxes with full state retention, so a session can launch, suspend, and resume on demand. Lambda Durable Functions add steps and waits as checkpoints that let an execution resume past completed work after an interruption. Lambda Managed Instances put the same model on dedicated EC2 capacity reaching 32 GB memory and 16 vCPUs.
Event-driven reach across AWS
Reach is the real pull. AWS reports native integration with over 220 AWS services and 50 SaaS applications: an object lands in Amazon S3, a record hits Amazon DynamoDB, an API call arrives, and the matching function runs. For streaming, Lambda sits directly on Amazon Kinesis, Amazon SQS, Amazon DynamoDB Streams, and Apache Kafka, with Provisioned mode for ESM setting throughput on event source connectors.
Managed runtimes cover Python, Node.js, Java, C#, Go, and Ruby, plus custom runtimes. Code ships as a ZIP archive or a container image, deployed through the AWS Console, infrastructure as code, or a CI/CD pipeline. Editor toolkits, remote debugging, and local testing keep iteration off the console.
Observability is built in. Logs, metrics, and traces flow to Amazon CloudWatch and AWS X-Ray automatically, and Application Signals adds APM-style tracking of throughput, availability, latency, faults, and errors. Lambda extensions integrate third-party monitoring tools, and the AWS Serverless MCP Server equips coding agents like Claude Code, Kiro, and Cursor to build and manage Lambda applications.
Cold starts and cost levers
Cold starts are the standard objection, and Lambda answers with two levers. SnapStart caches an initialized snapshot of a function; the vendor reports up to 10x faster startup for Java and sub-second startup for Python and .NET, with minimal or no code changes. Provisioned concurrency keeps environments warm on latency-sensitive paths.
Cost control is largely architectural. Functions run on x86 or Arm-based Graviton processors, and AWS says you can switch between architectures without code changes to optimize for cost or performance. The free tier includes one million requests and 400,000 GB-seconds per month, which keeps low-traffic internal tools effectively free.
Isolation and compliance posture
Isolation is stronger than the shared-tenancy reputation of serverless suggests. Lambda executes functions on dedicated compute powered by the AWS Nitro System, never shared between functions, customers, or accounts, inside network-isolated VPCs that Lambda manages. Tenant isolation mode binds execution environments to a single tenant, and certifications span SOC, HIPAA, PCI, and ISO.
Who should choose AWS Lambda
AWS Lambda is the obvious pick for teams already on AWS with spiky or event-driven traffic: webhook handlers, S3 and DynamoDB triggers, scheduled jobs, stream processors, and multi-tenant back ends where per-tenant isolation is a hard requirement. Durable Functions and MicroVMs also make it a credible home for agentic AI workloads that pause, wait, and resume.
It fits poorly for teams outside the AWS ecosystem, or for anyone wanting a flat predictable bill: the meter has many dimensions (requests, duration, provisioned concurrency, ephemeral storage, durable operations) and forecasting takes real modeling. Steady always-on services often cost less on a plain instance, though Lambda Managed Instances narrows that gap.
Strengths
- Always-free tier never expires: 1M requests + 400,000 GB-seconds every month
- True pay-per-use billing metered per millisecond — zero idle/server cost
- Arm/Graviton2 cuts duration ~20% ($0.0000133334 vs $0.0000166667 per GB-second)
- Auto-scales from zero to thousands of concurrent executions with deep AWS integration
Watch out for
- Cold starts add latency (100ms to several seconds) for infrequently hit functions
- Can cost more than always-on containers under high, sustained traffic
- 15-minute max runtime and 10,240 MB memory ceiling limit heavy/long jobs
- Vendor lock-in plus multi-service billing makes cost forecasting hard
Key features
| Feature | What it does |
|---|
| Pay-per-request + GB-second | $0.20 per 1M requests plus $0.0000166667/GB-second on x86, billed per millisecond of runtime |
| Always-free tier | 1M requests and 400,000 GB-seconds monthly, forever — not limited to the first 12 months |
| Arm/Graviton2 architecture | ~20% cheaper duration and up to 34% better price-performance via a one-line architecture switch |
| Configurable memory 128 MB–10,240 MB | vCPU scales with memory; 512 MB ephemeral /tmp free, extra at $0.0000000309/GB-second |
| Provisioned Concurrency | $0.0000041667/GB-second to keep functions warm and eliminate cold-start latency |
| 200+ event sources | Triggers from API Gateway, S3, DynamoDB, SQS, EventBridge and the wider AWS ecosystem |
How AWS Lambda compares
| Alternative | How it differs |
|---|
| Google Cloud Run | Container serverless billed per vCPU-second + GiB-second; often cheaper for concurrent, sustained traffic |
| Azure Functions | Comparable consumption plan and free grant; best fit for Azure/.NET-centric teams |
| Cloudflare Workers | Edge serverless with near-zero cold starts; $5/mo paid plan, cheaper for lightweight global requests |
Frequently asked questions
How much does AWS Lambda cost?
AWS Lambda charges $0.20 per 1 million requests plus $0.0000166667 per GB-second of compute on x86 (memory in GB times runtime). A 512 MB function running one second costs about $0.0000083 in duration. Arm/Graviton2 is roughly 20% cheaper at $0.0000133334 per GB-second. Total cost scales with invocations, memory allocated, and execution time.
Is AWS Lambda free?
Yes. Lambda includes an always-free tier of 1 million requests and 400,000 GB-seconds of compute every month, and unlike most AWS free tiers it never expires after 12 months. Small apps and side projects often stay entirely free. Beyond those limits you pay $0.20 per million requests plus per-GB-second duration charges.
AWS Lambda vs Google Cloud Run — which is cheaper?
Both bill per request and per resource-second. Lambda charges $0.20 per million requests and $0.0000166667 per GB-second; Cloud Run bills vCPU-seconds and GiB-seconds separately and can win for sustained, concurrent traffic since one instance serves many requests at once. Lambda usually wins for spiky, event-driven workloads and its non-expiring always-free tier.
How is AWS Lambda pricing calculated?
Duration cost equals allocated memory (GB) multiplied by execution time (seconds) to get GB-seconds, then multiplied by the rate — $0.0000166667 on x86 or $0.0000133334 on Arm. Add $0.20 per million requests. Memory is configurable from 128 MB to 10,240 MB, and more memory also grants more vCPU. Billing is per millisecond.
Is AWS Lambda cheaper on Arm (Graviton2)?
Yes. Arm/Graviton2 duration costs $0.0000133334 per GB-second versus $0.0000166667 on x86 — about 20% less — and AWS cites up to 34% better price-performance. Requests cost the same $0.20 per million on either architecture. Switching is a one-line config change, provided your dependencies and runtime support Arm64.
AWS Lambda ratings on other platforms
Independent user ratings from third-party review sites, linked here for transparency.
These are not our editorial score, are captured on the date shown, and may have changed since.
