Geospatial foundation models have arrived and are getting better. We examine how the vector embeddings they provide can be used for clustering, change detection, and natural language-search, and discuss the challenge of scaling these solutions across time and space.