Google Colab

Google Colab is a service that lets you code in a web browser. You don’t need to install any program on your own computer for this. The code isn’t actually run on your own laptop, but on a computer in one of Google’s data centers. So you type commands into a page in the browser, Google does the computing, and the result appears right below. The programming language used is almost always Python, the standard language for data analysis and artificial intelligence. The name comes from “Collaboratory,” because several people can work on the same file at the same time, similar to Google Docs.

Why Colab lowers the barrier to entry for AI

Modern AI programs require a great deal of computing power. Graphics cards are especially important for this — special chips that can perform many calculations simultaneously. A good graphics card for AI easily costs several thousand euros. Colab provides such a graphics card for free for a few hours a day. This allows a student on an old laptop to try things out that used to require a university lab.

A second advantage is that no installation is needed. Anyone who sets up AI software on their own computer often spends hours dealing with error messages about incorrect versions. In Colab, the most important software is already available. You open a link and can start right away.

That’s why Colab has become the standard format for tutorials. When researchers publish a new model, a Colab link is often included. Readers can then run the code themselves with a single click instead of just reading it. This speeds up how quickly new methods spread.

Notebooks, cells, and borrowed computing time

Colab works with what are called notebooks. A notebook is a file that alternates between text and code. The code is contained in individual boxes called cells. Each cell can be run individually, and you see the result directly below it, for example a number or a chart. The whole thing resembles a lab notebook: explanation, experiment, result, alternating throughout.

In the background, Google assigns each session a virtual machine, i.e. a borrowed computer. This machine only runs temporarily. After a few hours, or after a longer period of inactivity, it is shut down and all files stored there are gone. The text of the notebook itself is preserved because it is stored in Google Drive.

This is the most common beginner’s mistake: Colab is not permanent storage and not a server that runs an application around the clock. It is a workbench for a limited time. Anyone who wants to keep results must explicitly save them to Google Drive or their own computer. For permanent operation, one instead rents actual servers, for example from a cloud provider.

Colab in teaching, research, and cloud competition

Colab is most commonly encountered in courses and tutorials. Computer science courses at schools and universities often distribute their assignments as a Colab link, because that way all participants have the same environment. Many YouTube tutorials on image generators or language models also essentially consist of a Colab notebook.

In research, Colab serves as a showcase. A team publishes a model, and the accompanying notebook shows in just a few cells how it is used. Companies like Hugging Face, a platform for freely available AI models, regularly link their examples there.

In business news, Colab usually comes up in the context of competition for developers. Google offers the service cheaply or for free so that programmers get used to its own cloud. Anyone who later wants to do serious computing then pays for Colab Pro or for Google Cloud. Competitors like Amazon and Microsoft run similar notebook services for the same reason.