Embeddings are numerical representations of machine learning features used as input to deep learning models. One-hot encoding, TF-IDF and PCA were early ways of compressing large amounts of textual data. Word2Vec was first step forward in moving on from simple statistical representations to semantic meaning of words. Transformers, transfer learning, generative methods etc. have all contributed to the explosion in use of embeddings and establishing them as a foundational ML data structure.
