Neo4j Unveils Infinigraph: Revolutionizing Graph Database Technology with Unified Workloads at 100TB+ Scale

Neo4j unifies real-time transactions and graph analytics at scale - SiliconANGLE

Neo4j unifies real-time transactions and graph analytics at scale Graph database maker Neo4j Inc. today launched Infinigraph, calling it a significant advancement in distributed graph technology. The company said the architecture allows users to run both operational and analytical workloads on a single graph database platform at over 100 terabytes in scale without fragmenting the graph, duplicating infrastructure or compromising performance. The product of more than two years of development, Infinigraph addresses the problem of harmonizing transactional systems and analytical workloads. Previous versions of Neo4j required running on a single physical computer, meaning that organizations had to use extract/transfer/load pipelines, synchronization, or multiple databases to handle high volumes. Infinigraph addresses this limitation by using sharding -- a database technique that splits large datasets into smaller, more manageable pieces -- to support billions of relationships and thousands of concurrent queries across multiple processors while maintaining the atomicity, consistency, isolation and durability, known as ACID, that's needed in transactional scenarios. The company can shard graph data across multiple machines while preserving its logical consistency, allowing for automatic distribution and scaling of data without the need for application rewrites or manual intervention. Neo4j said the new architecture lets customers embed tens of millions of documents as vectors directly into the graph. This enables use cases such as fraud detection, product knowledge graphs, long-term compliance monitoring and semantic search to be conducted on much larger and richer data volumes. "We're now able to support billions of vectors in the Neo4j database," said Sudhir Hasbe, president of technology at Neo4j. "This is particularly useful in life sciences, where companies are processing tens of millions of scientific documents for drug discovery. In the past, these documents would be orphaned. Now, they can be embedded directly into the graph." Neo4j laid the groundwork for the new architecture with the introduction of Fabric four years ago. That enabled federated graph queries across machines, but customers had to manage sharding themselves. Infinigraph automates this process while retaining full ACID compliance, a feature Hasbe said is critical for transactional reliability. "Graph sharding is a difficult problem due to traversal queries," which are a type of database query used to navigate relationships between connected data points in graph databases, Hasbe said. "We solved it by maintaining a global index in one environment for fast path queries, while distributing the actual data across machines for horizontal scalability. Even distributed transactions remain consistent and reliable." Neo4j quoted early-access customers Intuit Inc. and Dun & Bradstreet Corp. expressing their enthusiasm for the new features. "Running real-time queries while also analyzing broader patterns is critical," said Moheesh Raj, D&B's director of engineering. "That requires a graph to scale both." "Some of the biggest banks are now able to run fraud detection systems using Infinigraph, working with hundreds of terabytes of interconnected transactional data," Hasbe said. Neo4j also trumpeted the value of graphs as vector databases used in generative artificial intelligence. AI training requires both structured and unstructured data. The company first added vector support in 2023, allowing documents to be stored as vector embeddings. Infinigraph enables storage at a much larger scale. "Gen AI has made unstructured data more valuable than ever," Hasbe said. "We've seen customers go from using [Elasticsearch BV's] Elastic Store for vectors to managing everything within Neo4j. That's a huge simplification of their stack." Infinigraph is available on an early access basis now in Neo4j's self-managed Enterprise Edition, with broader availability set for October. The company said the feature will soon be available within its AuraDB cloud-native graph platform. Pricing for Infinigraph will follow a decoupled model, separating compute and storage to provide greater flexibility. "We're aligning our pricing model with how modern distributed systems operate," said Hasbe. "It allows customers to scale their workloads without unexpected costs." He said customers with smaller workloads will probably see costs decline from what they are currently paying.

Neo4j Launches Infinigraph for 100TB+ Unified Graph Workloads

Neo4j has launched a graph database built to unify workloads at 100TB+ scale for generative AI. Neo4j has launched Infinigraph, a new distributed graph database architecture designed to run both transactional and analytical workloads in one system at 100TB+ scale. The platform enables enterprises to store and analyse billions of relationships and run thousands of concurrent queries in real-time. It supports use cases such as embedding tens of millions of documents as vectors for context-aware assistants, global fraud detection, large product catalogues, and compliance analysis. By merging operational and analytical workloads, Infinigraph addresses the long-standing challenge of data silos. Enterprises often run separate transactional and analytical systems, leading to cost overheads and delays. Neo4j claims its approach removes ETL pipelines, sync delays, and redundancy.