Keras supports multiple backends, allowing users to choose the one that best suits their needs. As of my last update in January 2022, the main supported backends for Keras are :

* TensorFlow : TensorFlow is the default and most widely used backend for Keras. It provides high performance, scalability, and supports both CPU and GPU computation. Starting from TensorFlow 2.0, Keras has been integrated into TensorFlow as its official high-level API, making TensorFlow the recommended backend for new projects.

* Microsoft Cognitive Toolkit (CNTK) : CNTK is another backend supported by Keras. It is developed by Microsoft and is known for its scalability and efficiency, particularly for training deep learning models on large datasets. However, its usage with Keras is less common compared to TensorFlow.

* Theano : Theano was one of the original backends supported by Keras. It is a symbolic mathematical library that allows for efficient computation of mathematical expressions, making it suitable for deep learning research. However, development and support for Theano have been discontinued, and it is no longer actively maintained.

While TensorFlow is the recommended backend for new projects due to its integration with Keras and ongoing development, the choice of backend ultimately depends on factors such as performance requirements, compatibility with existing infrastructure, and personal preference.