Serverless computing lets you run code without provisioning or patching servers. You upload a function, the cloud provider runs it when a request or event arrives, and you pay for the time it runs. Servers still exist. The provider operates them, which removes capacity planning from your list of jobs.
How it works
AWS Lambda, Google Cloud Functions, Azure Functions, Vercel Functions and Cloudflare Workers all follow the same model. An event triggers a function: an HTTP request, a file upload, a queue message or a schedule. The platform starts a runtime, runs your handler and shuts the runtime down after a period of inactivity. If a thousand requests arrive together, the platform starts many copies in parallel.
export const handler = async (event) => {
const name = event.queryStringParameters?.name ?? "world";
return { statusCode: 200, body: JSON.stringify({ message: "Hello " + name }) };
};Where serverless saves money
Traditional servers bill by the hour whether or not they serve traffic. Functions bill per request and per millisecond of compute. A workload with idle gaps or sharp spikes costs less on functions:
- A contact form that gets 50 submissions a day.
- A nightly job that builds a report.
- An image thumbnail generator triggered by uploads.
- A webhook receiver for payment or messaging events.
Steady, high-volume traffic reverses the math. At a constant load, a reserved container or virtual machine is often cheaper. Run the numbers with your real request counts and memory settings before you commit.
Cold starts
When no warm instance is available, the platform must start a new one. This cold start adds latency, typically from tens of milliseconds to over a second depending on runtime, package size and network setup. You can reduce it:
- Keep deployment packages small and import only the modules you use.
- Choose lightweight runtimes such as Node.js, Python or Go over heavy JVM setups, or use features like Lambda SnapStart for Java.
- Use provisioned concurrency for latency-critical endpoints.
- Avoid attaching functions to a private network unless the database requires it.
Limits to check
Every platform caps execution time, memory, payload size and concurrency. AWS Lambda, for example, stops a function after 15 minutes. Long video processing or big batch jobs need a container service or a workflow tool such as Step Functions. Functions are also stateless: keep sessions and files in a database, cache or object storage, not on local disk.
Databases and connections
Hundreds of simultaneous functions can open hundreds of database connections and exhaust a traditional PostgreSQL or MySQL server. Put a connection proxy such as RDS Proxy or PgBouncer in front, or choose databases built for this pattern, such as DynamoDB, Firestore or HTTP-based serverless Postgres.
Vendor lock-in and local development
Event formats, permission models and deployment tools differ by provider. Reduce the coupling by keeping business logic in plain modules and writing a thin handler around them. Use the Serverless Framework, AWS SAM or SST for repeatable deployment, and run tests against the plain modules on your laptop.
Observability
Debugging a distributed set of short-lived functions is harder than reading one server log. Log structured JSON with a correlation ID, enable distributed tracing, and set alarms on error rate, throttles and duration. Set a budget alert on your cloud account too, because a bug that loops a function can create a large bill .
A sensible way to start
- Pick one low-risk workload, such as form handling or file processing.
- Build it as a function with infrastructure defined in code.
- Measure cost and latency for a month.
- Expand to more workloads once the numbers hold up.
Common questions
Is serverless always cheaper?
No. Functions cost less for spiky or low-volume workloads and cost more for steady, high-volume traffic that a reserved server would handle at a flat rate. Estimate with your real numbers: monthly requests, average duration and memory size. Most cloud providers publish pricing calculators that make the comparison quick.
Can I run a whole website on serverless?
Yes. Platforms such as Vercel and Netlify run server-rendered pages and API routes as functions, and static assets come from a CDN. The pattern suits marketing sites, e-commerce storefronts and SaaS front ends. Long-running processes, persistent WebSocket connections and heavy batch jobs fit better on containers.
How do I handle secrets and configuration?
Store secrets in the platform's encrypted environment variables or a secrets manager such as AWS Secrets Manager, and load them at startup. Never commit them to Git. Give each function its own role with only the permissions it needs, so a compromised function cannot read unrelated data.
How do I test functions locally?
Keep business logic in plain modules and test those with ordinary unit tests. For handler-level testing, tools like SAM CLI, Serverless Offline, the Vercel CLI and Wrangler emulate the platform on your laptop. Add a small set of integration tests that run against a real cloud staging environment before each release.
Many of our clients run a hybrid setup: a core application on containers and event-driven tasks on functions. To see where serverless would cut your hosting bill, contact us for an architecture review, or read about our cloud and backend services.