Contents 

Summary

  • In 2027, LLNL’s license for the Anaconda Python distribution will expire and will not be renewed.
  • The conda package manager is still free, but the binary packages that Anaconda, Inc. provides on conda’s defaults channel are not.
  • NOTE LLNL will begin blocking Anaconda’s paid channels site-wide in February 2027.
  • Many alternative software packages are available.
  • Read on to understand how to migrate to open alternatives.

Conda and Anaconda

To understand this change, it is important to understand the difference between Conda, the tool, and Anaconda, the service.

Historically, Python’s package management tools only supported installation of Python programs, but many data science packages require lower-level software packages written in languages like C, C++, and Fortran. The Conda package manager and the Anaconda python distribution were created in 2012 by Anaconda Inc. (then called Continuum Analytics) to address this need.

Conda is an open-source binary package manager, similar to RPM, APT, and other systems. Conda allows users to install python packages along with compatible, pre-compiled C, C++, and Fortran software. It also allows users to keep these packages up to date. Anaconda is a distribution, or a set of builds, of these software packages, focused primarily on data science and AI. It is maintained by Anaconda, Inc. and the software packages are hosted on Anaconda Inc.’s servers. Many users download these packages daily, and Anaconda pays for servers, bandwidth, and compilation (build) cycles to maintain them.

Anaconda Licensing

While originally both conda and the Anaconda python distribution were free for anyone to use, Anaconda, Inc. changed their licensing terms in 2020. The Anaconda package distribution is now free for personal use and for organizations smaller than 200 people, but it requires a license for any larger organization not using Anaconda for educational purposes. LLNL is a mission-oriented research organization, and our users do not use Anaconda solely for research, so it requires a paid license to use the Anaconda binary packages.

Anaconda updated its license terms further in 2024 to include not only regular users, but also AI agents in its paid licensing model.

While the Anaconda packages are a paid product, and conda is still free and open source, the default channel for conda is still the defaults channel from the Anaconda python distribution. Any user who downloads the tool and uses it without configuration will end up downloading non-free binaries, and Anaconda, Inc. will see the access to its packages repository and will count this towards LLNL’s license usage. This makes it extremely difficult for LLNL to control usage of Anaconda or to limit license costs. Over the years, the Anaconda license has become increasingly expensive.

Blocking Anaconda

Because of this new licensing model, and because the license for Anaconda has become prohibitively expensive, in early 2027, LLNL will let the Anaconda license expire and will block Anaconda’s servers site-wide. LLNL users will no longer be able to connect to the defaults channel to use the Anaconda python distribution.

LLNL joins other large research organizations in this decision, including:

Alternatives to Anaconda

Luckily, many free, well-maintained alternatives to Anaconda have emerged. We link to the most popular of them below, along with some brief advice for projects undertaking a migration away from Anaconda.

Conda-forge + Miniforge

Miniforge is an installer for Conda and its faster sibling, Mamba that comes preconfigured to use the conda-forge package repositories. Conda Forge is a software distribution like Anaconda, but it is community-led, free, and has many more packages than Anaconda’s defaults channel. For most users, Miniforge should be a drop-in replacement for existing Anaconda workflows.

Pixi

For those wanting a more modern package management experience for the conda-forge ecosystem, Pixi is gaining traction. Pixi offers modern package management features like lockfiles and reproducibility, as well as better performance than conda.

Pip and UV

In the years since Conda’s creation in 2012, the Python ecosystem has evolved many of the same features.  In particular, the Python Package Index (PyPI) now supports binary packages called “wheels”, which allow its users to leverage the fast installation and performance of C, C++, and Fortran libraries in a similar way to conda.

  • Pip is the traditional tool for installing python packages, and many projects like PyTorch have moved away from Conda in favor of pip because of its large number of users. Pip is managed by the Python Software Foundation, a non-profit company that stewards the Python language and software ecosystem. You may find that you can simply pip install many of the packages you need.
  • UV is rapidly gaining traction as a faster, more modern replacement for pip. It offers advanced workflow features similar to pixi, but it leverages the PyPI ecosystem rather than conda-forge packages. UV is developed by Astral, a venture-funded company dedicated to building fast, modern developer tools for the python ecosystem.

Spack

Spack is a package manager for high performance computing (HPC), developed at LLNL. Spack includes many popular python packages and AI tools like PyTorch and TensorFlow, as well as a wide range of HPC software packages, compilers, and other tools. It is not primarily a binary package manager, and you will likely need to build many of native packages from source. If you are looking to integrate your software with HPC codes, or if you need many components compiled for GPUs or for vector instructions on modern CPUs, then you should consider using spack for your project.

Getting Help

LC Users

The Livermore Computing (LC) Hotline (2-4531) is available to answer questions around migrating away from Anaconda for LC HPC users.

Institutional Users

Other users can ask questions or request assistance via the Python Teams channel. NOTE LivIT Service Desk does not have the Anaconda expertise necessary to answer such questions via ServiceNow or x4-HELP.