CartoDB is an open souce tool and SaaS platform that allows users to make beautiful maps quickly and easily from their own data. To complement our users needs, we launched last year our free-to-use open source OSM based basemaps Positron and Dark Matter (https://github.com/CartoDB/CartoDB-basemaps), designed in collaboration with Stamen to complement data visualization. While architecturing them, we had several compromises in mind: they had to be powered by our existing infrastructure (powered by Mapnik and PostGIS at its core), they had to be scalable, cacheable but frequently updated, customizable, match with data overlays, and, last but not least, they had to be beautiful. This talk is the tale of the development process and tools we used, how we implemented and deployed them and the technology challenges that arose during the process of adapting a dynamic mapping infrastructure as CartoDB to the data scale of OSM, including styling, caching, and scalability, and how (we think) we achieved most of those. I will also talk about the future improvements that we are exploring about mixing the combination of basemap rendering with data from other sources, and how you can replicate and tweak those maps on your own infrastructure.