Neo4j is a graph database platform. What differentiates this approach to modeling, storing, and recalling data, and which business and technology applications is it useful in?
Databases have been a cornerstone of technology since the 1970s, but for decades, they were limited to a table format, organising information in rows and columns, much like an Excel spreadsheet. This worked well for simpler tasks, like payroll systems. However, as data complexity has grown with the rise of users and devices, the table format has become less effective.
At Neo4j, we rethought data organisation, taking inspiration from the human brain, which stores information through a network of connected neurons and synapses. In mathematics, this is called a graph, and we designed our database to use nodes (representing entities) and relationships between them. This system mirrors how humans understand facts and interact with information, forming connections. It is especially useful for modern applications, particularly with AI, where understanding the relationships between data points is critical to ensure accurate and contextually relevant results.
You earlier called AI a “magical fit” for Neo4j. How are graph databases currently being used in AI and machine learning?
Before AI can be effective, data must be consolidated and organised properly. Fragmented or inconsistent data will not yield meaningful AI insights.
AI thrives on context, and that is where graphs excel.
For example, a law firm used Neo4j to consolidate vast amounts of legal documentation into a knowledge graph. Their small team could now extract answers in seconds, a task that used to take a day. AI reads the data and provides direct answers, eliminating the need for humans to sift through multiple systems.
Similarly, Klarna, a company that transitioned from buying SaaS services to building internal software using Neo4j, now manages its data and AI applications much more efficiently. These examples highlight the importance of a solid data foundation before applying AI to achieve valuable results.
There is a lot of talk of Artificial General Intelligence (AGI). Do you think existing graph-based data systems like Neo4j could provide an infrastructural foundation for that kind of machine intelligence, or would database logics have to evolve further to sustain AGI?
The term AGI has been useful in the past to paint a vision for AI, but it has never been clearly defined. Technology evolves quickly, and what we once thought was magical is now just part of our everyday reality. Take the Turing test, which for 60 years was a benchmark for AI — it is no longer the definitive marker for AGI. While AGI is an ideal goal, it is hard to pinpoint exactly when or how it will manifest. What we are seeing is rapid progress, with fast model innovation and a growing need to make sense of data. Whether or not we call it AGI, the trend is exponential. In terms of Neo4j’s current role, it is not about sustaining AGI but about providing the infrastructure to help AI systems make sense of complex data, which will be critical regardless of AGI’s arrival.
In 2023, prominent figures like Elon Musk, Steve Wozniak, and AI pioneers Stuart Russell and Yoshua Bengio signed an open letter via the Future of Life Institute calling for a six-month pause on AI development, citing potentially unmanageable future risks - but we quickly moved beyond this ‘red line’. What do you consider a hard red line in AGI development?
I do not have a clear red line. There is an assumption that reaching AGI automatically signals danger, but we are still building tools and humans are in control. AI is like a hammer — it can build houses or harm people, but it is ultimately how we choose to use it. The real issue is not the software itself but its opacity. As AI systems grow in complexity, understanding their decisions becomes more challenging. If we lose the ability to understand and hold AI accountable for its decisions, that is a real danger. The opacity of AI systems needs to be addressed, and current regulatory frameworks are lagging behind, creating risks that need careful management.
The risk lies in the opacity of AI models, particularly deep learning models like neural nets. While we understand how these models are built, once they are created, they become more like a telescope — we use them to discover new things, but we do not always understand everything they reveal. The challenge arises when AI systems are made up of multiple models and components that operate as black boxes, making it harder to understand their decision-making. Neo4j’s graph-based approach helps here by providing transparency into the relationships within AI systems, which makes them more auditable. While transparency will not eliminate all risks, it is a crucial step in minimising them. Explainability and auditability are key to keeping AI systems under control.
We have spoken to several companies whose business or technology is modeled on human behavior or biology, like cybersecurity systems that mimic the human immune system, or the application of Fibonacci retracement levels in cryptocurrency market analysis. Do you think that this is key to being successful?
It is not necessarily key to success, but there is a meta-trend where software systems evolve similarly to complex biological systems. It is not a matter of simply copying biology, but drawing inspiration from it has proven valuable. Thirty or forty years ago, hardware constraints prevented computer scientists from expressing these ideas, but today, the hardware has evolved to support these kinds of complex systems.
Neo4j would not have been possible in the past because the hardware was not capable of supporting it, but today, with advanced technology, we are able to build systems inspired by complex biological processes. This is likely why we are seeing technologies like Neo4j emerge now.
The global graph database market is projected to grow from USD 2.85 billion in 2025 to USD 15.32 billion by 2032. It seems like Neo4j has cornered the market, but why are there not more players in this space? Is it becoming more pluralized?
When we founded Neo4j, we took a category-creation approach. We did not just call it a database; we coined the term "graph database" to describe this new kind of database. When you create a new category, competition is inevitable. Even Oracle, the dominant database company, only holds about 40% of the market. In the early days, we were alone in this space because the term "graph database" didn’t exist, but now all the major cloud platforms, like AWS, Microsoft, Google, and Oracle, have launched their own graph database products. Plus, younger startups are entering the market, looking to compete with us.
But competition is a good thing. It means we have succeeded in growing the market and raising awareness about the value of relationships in data. We want everyone to use relationships in data to build better applications, and that is something we’re all working towards.
