cloudparts

TigerGraph

A distributed graph database built for parallel traversals over very large graphs, queried with its own SQL-like language GSQL (and openCypher), self-managed or as the Savanna cloud service.

What it is
Graph databases
Type
Commercial product
Made by
TigerGraph, Inc.
License
Proprietary; Community Edition is free, including for production use within its limits
Facts checked
24 October 2026
Website

What it is

TigerGraph calls its design a native parallel graph: each machine stores part of the graph and computes on it locally, and the parts coordinate by passing messages. The storage engine (GSE) and the processing engine (GPE) are written in C++, and a REST server, RESTPP, takes requests and manages tasks. Graphs are schema-first: you declare vertex and edge types, then run loading jobs. Queries are written in GSQL, a SQL-like, Turing-complete language designed in 2015, so whole algorithms can be written as queries. Most of openCypher is also accepted, embedded in GSQL, and TigerGraph says it supports ISO GQL too. Each GSQL query or REST call is one ACID transaction; isolation is read committed, implemented with MVCC, a write is complete only when every replica has applied it, and committed data goes to a write-ahead log on disk. The engine also stores vectors for hybrid graph and vector search.

Community Edition is free for production use on a single server, with up to 16 CPUs, 200 GB of graph data and 100 GB of vectors. Enterprise, priced through sales, adds clustering, high availability and disaster recovery, role-based access control and rolling upgrades. Savanna, the managed service that replaced TigerGraph Cloud Classic, keeps storage and compute separate so each scales on its own; workspaces are billed by the hour, from $1 for TG-00 (2 vCPUs, 16 GB of RAM) to $256 for TG-64 (512 vCPUs, 4 TB) in US regions, with a free trial. Data from a lakehouse is copied in: Iceberg tables through Kafka Connect or the Spark connector, Delta Lake and Hudi through Spark. A March 2026 paper by TigerGraph and Purdue researchers describes GraphLake, a research engine built on TigerGraph that queries Iceberg tables in place.

Why teams pick it

  • Scales out: the graph is partitioned across machines and queries run in parallel on each part
  • GSQL is Turing complete, so PageRank-style algorithms and multi-step analytics run inside the database, and openCypher and GQL are accepted too
  • A free Community Edition that allows production use, up to 16 CPUs and 200 GB of graph data
  • Savanna separates storage from compute and bills workspaces by the hour

Watch out for

  • Closed source, and Enterprise pricing is not published
  • Schema-first: vertex and edge types must be declared before loading, and openCypher queries silently skip vertices that lack an attribute instead of returning NULL
  • Isolation is read committed, not serializable
  • Lakehouse data has to be loaded through Spark or Kafka Connect, which means a second copy and a pipeline to keep it fresh; the smallest Savanna workspace costs about $730 a month if left running

Good for

Specs

Compare with alternatives →
Data model
Property graph with declared vertex and edge types; vectors stored alongside
Query language
GSQL, plus openCypher and GQL
Client protocol
REST requests through RESTPP
Transactions
Each GSQL query or REST call is one ACID transaction
Default isolation
Read committed (MVCC)
How it scales
Graph partitioned across machines, processed in parallel (clustering in Enterprise)
Replication
A write completes when every replica has applied it
Storage engine
GSE, its own C++ graph storage engine, with a write-ahead log
Written in
C++
Ways to run it
Self-managed Community or Enterprise Edition, or the Savanna managed service
Pricing model
Savanna: per workspace-hour, $1 (TG-00) to $256 (TG-64) in US regions

Alternatives

What all graph databases have in common →Compare side by side →
Neo4jOpen sourceAlso good for many-hop relationship queriesThe best-known native graph database: nodes and relationships stored as linked records, queried with Cypher, self-hosted or as AuraDB.MemgraphCommercialAlso good for many-hop relationship queriesAn in-memory graph database written in C++ that speaks Cypher over the Bolt protocol, with a library of graph algorithms and streaming ingestion from Kafka, Pulsar and Redpanda.PuppyGraphCommercialA graph query engine that reads tables where they already live, in Iceberg, Delta Lake, Hudi, warehouses and SQL databases, and answers openCypher and Gremlin queries without loading the data into a graph database.ArangoDBCommercialAlso good for many-hop relationship queriesA multi-model database: JSON documents, key-value lookups, graphs, full-text search and vectors in one engine, queried with one language, AQL.

How it fits with other technologies

See it in a real stack

No architecture uses TigerGraph yet. Drop it into the playground and wire it up yourself.Open playground →

Worth exploring

Paper· March 2026GraphLake: A Purpose-Built Graph Compute Engine for LakehouseShige Liu, Songting Chen, Chengjie Qin, Mingxi Wu and Jianguo Wang (Purdue University, TigerGraph), arXivExplains why running graph analytics over Iceberg tables is hard and how a TigerGraph-based engine does it: load only the key columns that form the topology, build one edge list per Parquet file, encode file and row in each vertex ID so properties can be fetched on demand, and cache column chunks with graph-aware priorities. On LDBC benchmarks it reports up to 26× faster startup and up to 60× faster queries than PuppyGraph, and it is candid that edge-list scans lose to TigerGraph’s own storage when few edges match. Half the authors work at TigerGraph and the baseline is a closed-source competitor measured from outside, so read the comparison with that in mind.
All resources →
Where it came from2012–2026

TigerGraph began in 2012 as GraphSQL, founded by Ruoming Jin, Li Chen, Mingxi Wu, Like Gao and Yu Xu. Its query language, GSQL, was designed in 2015 by Mingxi Wu and Alin Deutsch as a SQL-like but Turing-complete language for graph analytics.

The company came out of stealth in September 2017 under the name TigerGraph, with $33 million in funding, and raised $32 million in 2019 and a $105 million Series C in February 2021. It pitched a single system for transactions and analytics that could scale to hundreds of terabytes and trillions of edges, for uses such as fraud detection and customer 360.

Savanna, a rebuilt cloud service with storage and compute separated, entered beta on 29 April 2024. In March 2025 it announced a Community Edition that is free for production use on one server and includes vector search and Iceberg loading. In March 2026 researchers from TigerGraph and Purdue University published GraphLake, a design for running GSQL over Iceberg tables without loading them first.

Why it’s built this way

Storage and computation on the same machines
The native parallel graph keeps each partition of the graph next to the code that processes it and coordinates by message passing, so a traversal over billions of edges can run on many machines at once.
Its own language instead of only Cypher or Gremlin
GSQL was designed to feel like SQL while being Turing complete, so analysts can write full graph algorithms as queries rather than calling external libraries.
Separate storage and compute in the cloud service
Savanna lets storage and compute scale independently, so a workspace can be resized or stopped without moving the data.

Timeline

  1. 2012

    Founded as GraphSQL

  2. 2015

    GSQL designed by Mingxi Wu and Alin Deutsch

  3. 2017

    Leaves stealth as TigerGraph with $33 million

  4. 2021

    $105 million Series C

  5. 2024

    Savanna cloud service enters beta on 29 April

  6. 2025

    Community Edition free for production use, with vectors and Iceberg loading

  7. 2026

    GraphLake paper on querying Iceberg tables in place

Sources10 · facts checked 24 October 2026

Improve this page on GitHub