Digital twins constitute virtual representations of
physically existing systems, synchronized at a specified frequency and fidelity. One way to connect physical and digital twins is through data lakes, which are efficient storage and query processing systems to manage the data exchanged between the twins. Existing digital twin systems make use of NoSQL or time databases for realizing their data lakes. Although very efficient, these proposals present some limitations for implementing non-trivial queries over highly connected data. In this paper we explore the use of graph databases for implementing data lakes, compare them with similar NoSQL proposals, and discuss the situations where one solution outperforms the other.