What is serverless and why are there still servers behind it?

Published on 2026-10-20 by frfrand

Serverless explicado: qué son las funciones bajo demanda, cuándo se ejecutan, límites, cold starts, costos y diferencias con un VPS.

Serverless is a way to run applications or functions in which the provider manages most of the servers needed. Its name is a bit misleading: physical machines still exist, but you don't have to log in to each one to manage processes, patches, and capacity in the same way you would with a VPS.

It's similar to calling a taxi. You don't need to buy and maintain a car to get around, but of course someone is still using a real vehicle.

What is a serverless function?

Imagine that a person uploads a photo and your application needs to convert it to a small size. You can run a server permanently turned on waiting for that job, or use a function that the platform executes when needed.

A function as a service (FaaS) receives an event or request, processes information, and returns a result. The platform takes care of starting and scaling execution environments within its limits.

There are other serverless product modalities, including databases and managed services. Not everything that carries the serverless label works exactly like a function.

When is the code executed?

It can be triggered when an HTTP request arrives, a file is uploaded, a message appears in a queue, or a configured event is met.

For example, a store can use a function to process a payment notification. The function must verify where the notice comes from, validate its authenticity, and handle event repetitions before modifying the order.

The platform manages the runtime environment, but the business logic and code security remain the responsibility of the developer.

What is a cold start?

On certain platforms, if there is no runtime environment prepared for a function, the first call must wait for one to be prepared. That delay is known as a cold start.

Characteristics depend on the provider, language, and configuration. It is not advisable to assume that all calls take the same time or that every provider uses the same scaling system.

For an occasional administrative action, a few extra milliseconds may be irrelevant. For an extremely latency-sensitive experience, the difference may matter.

How much does serverless cost?

Many products bill based on invocations, duration, resources consumed, or transfer, with different plans and minimums.

This can be convenient for sporadic jobs: you do not need to maintain a server exclusively to run a function a few times a day.

But a constant or very intensive load can produce a bill different from expected. There is no universal rule that serverless is always cheaper than a VPS. It is advisable to measure consumption and compare alternatives.

What limitations may appear?

Depending on the service, there may be limits on duration, memory, packet size, connections, or available systems. It may also be necessary to store persistent data outside the function.

Thinking that the function has a permanent local disk where all client images can be saved is usually a bad idea. Functions must be designed with the provider's actual execution model.

If your application needs processes that remain active for days or very specific system software, a container service, a VPS, or a hybrid architecture may be more suitable.

Is serverless the same as microservices?

No. Microservices describe a way of separating software responsibilities. Serverless describes an operational model of execution and administration.

You can build a monolith with serverless parts or several microservices deployed on traditional servers. The decision about architecture and the decision about infrastructure are not identical.

What is vendor lock-in?

If your code depends deeply on events, APIs, and particular services from a provider, moving it to another may require quite a bit of work.

That doesn't make the platform a bad choice, but it's worth keeping in mind which pieces are portable and which depend on the chosen ecosystem.

How does it relate to ArduMaker?

In ArduMaker you can manage VPSs and application processes on servers. It's a different approach from fully delegating execution to a serverless platform. For real projects, both options can complement each other: an event-driven function can call an API hosted on a VPS.

In summary: serverless doesn't eliminate servers, it eliminates part of their direct administration. It's convenient when the work pattern and provider limits fit your application.

Sources: AWS Serverless · AWS Lambda · What is Vercel.