Circular materials depend on efficient recycling and materials valorisation. For this, it is important to know how different elements interact with each other. Thermochemical data are the primary source to model or just estimate what materials mix and what processes can be used to separate elements. In this contribution, consistency of the different humanly curated thermochemical databases in FactSage and high-throughput calculated ab initio databases (materialsproject.org and oqmd.org) are assessed. Possibilities, challenges and limitations for machine learning-based approaches to generate thermochemical data for circular materials are discussed.