At a time when every enterprise looks to leverage generative artificial intelligence, data sites are turning their attention to graph databases and knowledge graphs. The global graph database market size was $2.12 billion in 2022 and is expected to grow to $10.3 billion in 2032, according to estimates from Adroit Research.
These databases enable data managers and analysts to model, store, and query complex and increasingly interconnected relationships between datapoints. They also are employed to construct knowledge graphs that can cut through data silos, enable semantic searches, and support chatbots and recommendation engines.
The bottom line, experts agree, is that graph databases seek out and support relationships between datapoints. In the process, graph databases and knowledge graphs offer several advantages over standard or relational databases. “By design, graph databases are efficient at storing and querying relationships between datapoints—think nodes and edges in graphs,” said Marinela Profi, strategic AI advisor at SAS. “Contrast this to relational databases which often rely on complex and costly operations to fully understand relationships.”
Graph databases “excel at handling hierarchical and interconnected data—social networks, supply chains, ride-sharing apps—especially where the relationships are a key part of the data,” Profi continued. “They can provide performance advantages as the number of relationships grow.”
In contrast, Profi said, “Relational databases typically see a degradation in performance as relationships become more complex.” Knowledge graphs “incorporate semantics and context, making them invaluable for AI and LLMs [large language models],” said Dattaraj Rao, chief data scientist at Persistent Systems. “Graph databases are focused on the efficient storage and retrieval of information in the form of graphs. Graph databases serve roles that extend beyond knowledge graphs as well.”
“Graph databases are the data containers for any kind of networked data, including but not limited to knowledge graphs,” said Dorian Selz, co-founder and CEO at Squirro. “Knowledge graphs require a graph database to provide the data container, but they augment the database with semantic schema in the form of ontologies and taxonomies. The semantic schema supports machine reasoning. Knowledge graphs are an ideal complement to AI, including LLMs and retrieval-augmented generation (RAG).”
Graph databases and knowledge graphs, alongside other technologies like vector databases, “re-emerged recently to provide better interfaces into RAG frameworks,” said Yuval Perlov, CTO at K2view. “Our recent survey indicates that 44% use vector databases for internal documents and 38% use graph databases.”
Use Cases
These databases also enable the development of data models that can address challenges such as identity resolution and fraud detection. “They focus on relationships—edges—between datapoints—vertices—enabling faster, more efficient queries of nodes—entities—such as users, products, and companies,” said Ishaan Biswas, director of product management at Aerospike.
According to Selz, optimal use cases for graph databases include “social, professional, and knowledge networks, and pattern-based queries such as recommendation engines, multi-parameter similarity matching, and fraud detection.”
“The graph data model and query languages like Gremlin or Cypher are more expressive, visual, and intuitive than standard relational tables and SQL,” Biswas pointed out. “This allows business analysts to easily model graph queries using a common business language, making graph data easier to render in a human-readable format via common visualization tools.”
In terms of industry use cases, “graph databases are particularly useful in industries like finance, ecommerce, and adtech, where analyzing entity relationships within large datasets is critical,” Biswas said.
In applications such as fraud detection or recommendation engines, “graph databases can quickly identify connections between entities such as users or transactions, enabling real-time decision making,” said Rao. “Their ability to traverse data efficiently allows for faster and more insightful discoveries.”
For example, with LLMs and RAG systems, “using graph databases to map relationships within vast datasets can uncover deeper insights—such as better recommendations or enhanced search results—making them ideal for industries like ecommerce, social networks, or healthcare, where relationships between datapoints are critical,” Rao added.
“Think of graph databases as three-dimensional representations of datasets, where the relationships between a set of entities are just as meaningful as the attributes of the data,” said Christian Ward, EVP and chief data officer at Yext. “For example, in a social network application, a graph database can quickly identify influential users or suggest friends based on shared connections, interests, or interactions. The ability to maintain and query these intricate relationships in real time makes graph databases a powerful tool.”
Knowledge graphs are also “instrumental in natural language processing, where understanding context and relationships is essential for generating accurate responses,” Ward said. “The flexible, composable, and open nature of knowledge graph-based data eases the challenge of ensuring the semantic consistency of data across the enterprise,” said Perlov. “This allows business users, software engineers, and data scientists to find, understand, and use the data they need.”
Not Ready to Replace Relational?
Both graph and relational databases have their own strengths and optimal use cases. While graph databases store data as nodes and edges, relational databases store data in tables of rows and columns joined to other tables via key columns, Selz explained. “Where the data model is relatively static, a relational database is a highly efficient method to store, index, and retrieve large-scale, grid-like datasets. For example, relational is ideally suited to the management of financial transactions or product inventories.”
At the same time, “If the data model needs to be flexible, then graph databases are a better solution than relational, because the data model—schema—is not confined to a predefined set of data tables; rather, it is rapidly and continuously extensible by the creation of new entity classes and relationship types,” Selz added.
Graph databases are not suitable for all data types, and, therefore, they are not “an end-all” for generative AI systems, Perlov agreed. “For example, they struggle with transactional data; structured enterprise data found in systems like CRM, ERP, SCM; and with high-volume, low-relationship data. Relying solely on graph databases limits AI’s potential to deliver personalized, context-sensitive customer experiences. To fully realize AI’s benefits, organizations must also leverage data systems designed for structured data.”