Neo4j had its first stable release in 2010 and soon became one of the defining NoSQL systems, establishing "graph databases" as a new category. These systems represent data as nodes and edges, allowing for intuitive querying and visualization, as well as performance benefits on certain types of queries (e.g. path finding).

In this talk, I give a brief overview of the past, present, and future of graph databases. I first summarize the history of graph database systems, focusing on their main categories and use cases. Then, I discuss the key challenges that continue to hinder the adoption of graph databases, including a fragmented landscape and performance limitations.

I end the talk with recent positive developments: (1) Advances in standardization that led to the ISO GQL and SQL/PGQ languages, (2) Performance increases driven by competition on the LDBC benchmark suite, (3) A new generation of open-source graph database systems such as Kùzu and DuckPGQ, (4) Opportunities in machine learning use cases such as GraphRAG.