Who Wins The AI Pivot: What Graph Database Vendors Can Learn From NVIDIA, Palantir And IREN

NVIDIA built its chips for gamers, Palantir built its software for intelligence agencies and IREN built its sites to mine bitcoin. None of them planned for AI, but each found that the asset it already owned was exactly what AI needed. Graph databases are in the same position. Vendors built them to catch fraud and power recommendations, and AI agents now need what graph databases already do well: connected context, multi-hop reasoning and explainable answers. However, having a solution that fits the AI puzzle only triggers the pivot. The firms that fully pivoted – by changing their operating and business models – are the ones that reaped the most from the AI boom.

Successful pivots need more than the original asset

Every successful pivot needed more than the original asset or product. NVIDIA went from gaming chips to AI data centres, but the CUDA software around its chips is what locked in developers. CUDA launched in November 2006, yet data centres only overtook gaming as NVIDIA's biggest business in FY2023. By FY2026 it brought in $193.7 billion – approximately 90% of the firm’s revenue.

IREN went from mining bitcoin to selling AI cloud. It holds $4 billion of contracted AI ARR but recognized only $128.8 million of AI cloud revenue in FY2026, with bitcoin still making up about 82% of revenue. Palantir went from government intelligence work to commercial customers, and in Q3 2026 its US commercial revenue passed US government revenue for the first time. Its Ontology (a graph in all but name) sold as a business platform with actions and security on top. NVIDIA and Palantir show that a software layer on top decided who won, and IREN shows that demand alone doesn’t make a pivot pay.

Three major pivots in graph database history, with a fourth on the way …

The graph technology market has shifted three times since 2023, each one adding a layer on top of the last. Before 2023, graph was a specialist analytics tool for fraud detection, recommendations, network and identity graphs, and knowledge hubs. When LLMs arrived in 2023, vendors added vector indexes so graph could ground models alongside vector search. In 2024 and 2025, GraphRAG used multi-hop reasoning and explainable answers to make retrieval more accurate than vectors alone, and it is now standard across vendors. In 2026, vendors began building agent infrastructure: MCP tool layers, agent memory and governed write-back. A fourth shift – to operational context – is next: agents that watch the business and act on it. The agent layer, the part that matters most for the pivot, is less than two years old.

Graph vendors have reached the IREN stage, not the NVIDIA outcome

Graph vendors are where IREN is today: the demand is real, but the revenue to prove it hasn't fully landed. Neo4j reported 6x growth in GenAI customers and 58% growth in cloud consumption revenue in October 2025, but no graph vendor yet discloses AI-specific revenue. Buyers are at an early stage, too. In our 2026 survey of 306 firms, 92% reported plans to use graph technologies to some extent in 2027, but only 8% expect a firmwide rollout, compared with 25% for RAG.

Investors, meanwhile, are backing a new layer above the database: memory and context tools that sit on top of a graph and decide what an agent remembers and retrieves. The sums are still small next to what graph vendors raise, but the growth is fast. For example, Mem0 has raised $24 million and grew API calls from 35 million to 186 million between Q1 and Q3 2025. Apple's October 2025 deal to acquire Kùzu, an embedded graph database, treats graph as a component inside someone else's product.

How graph technology vendors can successfully navigate the pivot

Winning the pivot takes three things, and many graph vendors are strongest at the first.

1. A strong engine: agents depend on fast multi-hop queries, security, scale and governed writes, so the engine is the price of entry.
2. The software layer above it: the part people and agents actually use. One pharmaceutical firm chose it first, because it hid the graph's complexity from the rest of the business, and only then picked the database underneath. Yet many graph vendors leave this layer to others, connecting to tools such as LangChain, LlamaIndex, Graphiti and Mem0 rather than building their own.
3. Selling to executives: Most graph vendors grew through developers, which gets graph into pilots, but firmwide rollouts are funded higher up. In our survey, 40% of firms say the CEO makes the final call on large firmwide AI investments, and 95% favour vendors that send solution architects or forward-deployed engineers to build alongside them. Palantir already sells its graph-like Ontology to that audience, as an operational layer for the business rather than a database.

The vendors that win will be the ones building the layer agents call and selling it to the people who sign off firmwide rollouts. For more insights on graph database technologies, visit our website, and look out for our upcoming benchmark of 13 graph vendors in our Green Quadrant, set to publish on January 7.