/
.webp)
Generated by Master Biographer | Source for LinkedIn Content
A developer is on a flight to Mumbai. The year is 2000.
He's not drawing a product spec. He's not sketching a go-to-market slide.
He's drawing a data structure. Nodes. Relationships. Things connecting to other things — not in a flat table, but in a web, exactly the way the real world actually works.
He stares at the napkin and thinks something that will take him seven more years to act on: why doesn't this exist as a database?
His name is Emil Eifrem. The napkin is, technically, the founding document of graph databases.
Emil Eifrem, Johan Svensson, and Peter Neubauer met in Sweden in the late 1990s. They were engineers, not academics. They were building real software — specifically an enterprise content management system (CMS) on top of a relational database (Informix).
The system worked. Barely. The data they needed to model was deeply interconnected: documents linked to authors, authors linked to organizations, organizations linked to projects, projects linked back to documents. Every query required joining tables across tables across tables. The SQL got incomprehensible. The performance got terrible. And every time they added a new kind of relationship, the schema broke.
On a flight to Mumbai in 2000, Eifrem grabbed a napkin and sketched out what the data actually looked like — not as tables, but as a graph. Nodes. Edges. Properties on both.
He could not have possibly imagined the worldwide impact that napkin sketch would create.
But he also didn't start a company. Not yet.
For seven years, the three of them kept working. They kept running into the same wall. And they kept refining the model. In 2007, they formally incorporated Neo Technology in Malmö, Sweden, and open-sourced the first version of Neo4j under the GPL license.
When they launched, there was no category for what they were selling. The term "graph database" barely existed. They weren't competing with Oracle or MySQL — they were trying to explain to developers that relationships between data were just as important as the data itself. That was a harder sell than it sounds.
Version 1.0 shipped in February 2010. When it launched, Neo4j was not just the leading graph database — it was the only graph database. There was no market. There was no category. There were developers who stumbled across it and had a sudden, visceral recognition: this is how I should have been modeling this data all along.
Early adoption came from developers solving problems that made relational databases cry:
- Social networks (who knows whom? who influences whom?)
- Recommendation engines (what should I buy next?)
- Fraud detection (this account shares an address with that account, which shares a phone number with a known bad actor)
- Content management (the original use case)
The company nearly didn't make it.
In 2009, at the height of the global financial crisis, Neo4j had a term sheet from a top European VC. Due diligence was complete. The deal was done — in theory.
The VC pulled out mid-process.
Neo4j had $2,000 left in the bank.
Payroll was in days.
What happened next is a case study in operational improvisation: on Tuesday Eifrem told the team to go talk to customers. On Wednesday they were consulting in the field. On Thursday they sent invoices in advance. On Friday they sold those invoices for immediate cash. By the following Tuesday, payroll cleared.
They survived. Raised $2.5 million in seed funding from Sunstone and Conor that year. Signed their first Global 2000 customer in 2009.
In April 2016, the world's largest data leak dropped.
The Panama Papers: 11.5 million documents from Panamanian law firm Mossack Fonseca, detailing offshore tax structures, shell companies, and illicit wealth flows for politicians, oligarchs, and public figures across 200 countries.
The International Consortium of Investigative Journalists (ICIJ) had been sitting on this data for a year. They had 380 journalists across 76 countries working on it. And they had a fundamental data problem: the information wasn't a spreadsheet. It was a web. Who owned what company. Which company owned which other company. Which person shared which address with which entity. It was connections all the way down.
A relational database could hold the data. It could not answer the questions.
The ICIJ used Neo4j and visualization tool Linkurious to map the entire network. They loaded 320,000 offshore accounts into a graph. Journalists could then traverse it: click on a politician's name, and the graph would show every shell company they touched, every address they shared, every intermediary that connected them to a known sanctioned entity.
The connections that would have taken years of manual research were visible in seconds.
One of the first things the graph revealed was a direct link between the Icelandic Prime Minister and an offshore company holding claims on his country's failed banks — a conflict of interest he hadn't disclosed. He resigned within 48 hours of the story publishing.
Pierre Romera, CTO at the ICIJ, said it plainly: "The true value of these leaked files was in the connections we could draw between data points, but finding those manually would've taken years and would've probably been impossible. Neo4j is critical — despite the leak's massive size, it allows us to understand connections between all of our data — people, entities, transactions — and it helps us find the promising leads to investigate."
The world had just seen what a graph database could do that nothing else could.
A year later, the Paradise Papers — 13.4 million documents, 1.4 terabytes — went through the same pipeline. Same ICIJ. Same Neo4j. Same result: connections made visible that would otherwise have stayed hidden.
Neo4j went from Malmö garage project to infrastructure for the global financial transparency movement.
Enterprise adoption followed: all 20 of the largest U.S. banks. Nine of the ten largest pharmaceutical companies. Eight of the ten largest retailers. Adobe, AstraZeneca, eBay, UBS, Walmart.
The U.S. Army uses Neo4j to track equipment — one tank alone has ~10 million parts records and 15 million possible relationships between components. Relational queries don't work. Graphs do.
In June 2021, Neo4j raised $325 million in a Series F led by Eurazeo — at the time the largest single investment in a private database company in history — valuing the company at over $2 billion. Total funding across 10 rounds: $581 million.
Then the AI wave arrived.
LLMs hallucinate. They hallucinate because they don't know how things connect — they know facts in isolation, not relationships in context. The solution is a knowledge graph: a structured map of entities and their relationships that an AI can query before generating a response.
Gartner put knowledge graphs at the center of its 2024 Generative AI Impact Radar. The prediction: graph technologies will be used in 80% of data and analytics innovations by 2025, up from 10% in 2021.
Neo4j calls their approach GraphRAG — retrieval-augmented generation using a knowledge graph instead of a flat vector store. A study showed GraphRAG improved LLM response accuracy by 3x on average across 43 business questions.
The category Emil Eifrem invented on a napkin in 2000, the one he had to explain from scratch for years, the one that nearly ran out of money in 2009 — is now considered foundational infrastructure for artificial intelligence.
(Post-ready. Drop any of these as standalone content.)
1. The company had $2,000 in the bank and made payroll anyway.
In 2009, their lead investor pulled out mid-deal. They had days. They went from "we're done" to "consulting on the street, invoicing in advance, selling invoices for cash" in one week. That's not a pivot. That's survival instinct.
2. A napkin on a flight to Mumbai in 2000 is the founding document of graph databases.
Emil Eifrem wasn't pitching investors. He was building a content management system that kept breaking. He sketched the model that became Neo4j seven years before anyone would fund it.
3. Neo4j toppled a sitting head of government.
The Panama Papers used Neo4j to map 320,000 offshore accounts. The graph revealed Iceland's Prime Minister held undisclosed stakes in offshore entities with claims on Iceland's failed banks. He resigned within 48 hours of publication. That's a graph query with geopolitical consequences.
4. SQL takes 18 minutes to do what Neo4j does in under a second.
A developer tested "find the shortest connection path between two Spotify artists." SQL join query: ~18 minutes. Neo4j graph traversal: sub-second. The difference isn't the hardware. It's that relational databases scan indexes; graph databases chase pointers.
5. The Cypher query language was invented in 2011 because SQL is the wrong tool for connected data — and it became an ISO standard.
Andrés Taylor designed Cypher specifically for graphs — declarative like SQL, but built around patterns of nodes and relationships. Neo4j open-sourced it in 2015. In 2023, the ISO published GQL (Graph Query Language) — a new international standard, with Cypher as its primary influence. A Swedish startup's internal query language is now an international ISO standard.
Sources used: Hackernoon/Neo4j founder interview, Alejandro Cremades profile, PRNewswire Series F announcement, ICIJ Panama Papers coverage, Computer Weekly, inapps.net Neo4j Panama Papers, Neo4j government use cases page, Wikipedia Cypher query language, Neo4j GraphRAG manifesto, forttknox.com Emil Eifrem AI era profile
