Apache 2.0 License

Anyone who writes a program automatically owns the copyright to it. Without explicit permission, no one else may distribute or modify that program. A license is the permission the author grants for this. The Apache 2.0 License is one such ready-made permission text, published by the Apache Software Foundation in 2004. It permits a great deal: use, modification, distribution, and even making money from it. In return, at its core it only requires that the original author be credited and that it remain clear what has been changed in the code.

Why companies prefer this license

Free software is not always released under the same terms. Some licenses are 'contagious': anyone who incorporates such code into their own product must also disclose their entire product. The best-known license of this kind is the GPL. The Apache 2.0 License does not do this. You may incorporate Apache-licensed code into a closed, commercial program and keep your own source code secret.

This is exactly why it's so popular with companies. Legal departments have known the text for twenty years and don't need to review it anew each time. A start-up can take a freely available AI model, adapt it to its use case, and sell the result. No obligation arises to share its own work with competitors.

A second point concerns patents. A patent is a monopoly on a technical idea, independent of the specific program text. In theory, someone could give away code for free and later sue users for patent infringement. The Apache 2.0 License rules this out: anyone who contributes code automatically grants the necessary patent rights along with it. Older licenses like MIT or BSD do not regulate this as explicitly.

What obligations the text actually imposes

The conditions are manageable. Anyone distributing Apache software must include the license text. They must leave the copyright notices in the code intact. And they must note in modified files that changes were made. If a file named NOTICE exists with acknowledgments, its contents must also be passed along.

There is also a protective clause for the author. The software is provided explicitly without any warranty. If a program crashes and causes damage, the developer is not liable. This rule appears in almost all free licenses and is the price paid for giving away code for free.

There is also a clause against trademark misuse. The project's name may not be used for one's own advertising. So you may modify and sell an Apache web server, but you may not continue calling it Apache. A common misconception is that the license permits everything without exception. The attribution requirements are genuine obligations, and anyone who ignores them loses the permission.

Apache 2.0 in the AI industry

In the news, this license appears above all in connection with so-called open AI models. Models from the French provider Mistral and parts of Alibaba's Qwen series were released under Apache 2.0. For developers, this is a strong signal: they can run the models on their own machines, without usage fees and without having to ask for permission.

The license is also ubiquitous outside of AI. The Android operating system, the Kotlin programming language, and many building blocks of large data centers are covered by it. Anyone using any smartphone is thereby using Apache-licensed code.

One distinction is important. Not every model you can download is under Apache 2.0. Providers like Meta use their own licenses with additional conditions, such as caps on user numbers or bans on certain use cases. Such models are often still called open, but are legally more restrictive. It's therefore always worth taking a look at the license file.