BigQuery Graph GA 2026 Explained: Native Warehouse Graphs
1 Sep: BigQuery Graph leaves preview for GA. GQL runs in the warehouse. Some agent-build and memory pieces are still rolling out.
To read the 1 September 2026 cloud-data story, split two facts: BigQuery Graph moved from preview to general availability; this is not a new graph-database company. Some agent features in the same post are still preview or rolling out over coming weeks.
The Google Cloud blog, by Bei Li and Candice Chen, announced BigQuery Graph as generally available. ISO-standard Graph Query Language (GQL) sits beside SQL on one engine; traversals run in the warehouse. The vendor line is no separate ETL for the graph.
What the announcement actually ships
Questions such as how two accounts link or which path a payment took used to mean extracting tables into a standalone graph store. Google’s story is graph capability inside the warehouse, inheriting row- and column-level security plus BigQuery ML and AI functions.
One engine, two jobs
Official framing: large-scale graph analytics, and connected context for AI agents. The same query can call ML/AI functions. Do not treat “agents can build and chat a graph” as fully GA.
Speed numbers are vendor-vs-preview
Against public benchmarks, Google says GQL is 2× faster than preview and undirected traversal 100×, with cheaper cycle detection in ACYCLIC and TRAIL. Those are blog figures, not a third-party audit.
Borderless Lakehouse is the cross-cloud pitch
One graph can span native BigQuery tables and Iceberg tables in other clouds via Databricks Unity Catalog, AWS Glue, or Snowflake—traversed in place, no copy. The example is a support agent asking who supplies a delayed order across GCP and AWS.
How to read preview to GA
A timeline is calmer than “the warehouse finally learned graphs.”
-
1
Preview first unified graph and relational analytics
Preview sold GQL beside SQL, native traversal, no ETL. Pete Rubio, VP of Global Engineering at Thales Cybersecurity Products, is quoted on threat detection: security signals as a property graph, multi-hop walks in seconds.
-
2
GA adds expressiveness and agent doors
A new CALL statement and richer subqueries let you run a graph subquery per entity or invoke a named function. Conversational analytics chats the graph in natural language; Gemini Enterprise can attach via an MCP server.
-
3
Agent-authored graphs and audit memory are not all GA
A Google Cloud Data Agent Kit skill works in Antigravity, VS Code, Claude Code, and Codex. The blog says the skill will soon draft nodes and edges itself. Agent Analytics’ context graph stores actions as a queryable trace.
Do not flatten maturity into one headline
| Capability | Blog status (2026-09-01) | Still missing |
|---|---|---|
| In-warehouse engine + GQL | Generally available | Speed claims are vendor benches, not an SLA |
| Borderless Iceberg graph | Written into the GA narrative | Cross-cloud catalog IAM and latency follow docs |
| Agent-built graphs / authoring skill | Preview or coming soon | Do not put on this week’s must-ship PO |
Why enterprise data and agent stacks care
Named jobs: multi-hop threat and fraud, supply-chain digital twins, identity stitch / Customer 360, knowledge graphs and GraphRAG, network lineage. Yahoo’s Mikul Bhatt says agents traverse campaigns, audiences, and outcomes with GQL where monetization data already lives, then write conclusions back as edges. Workerbee’s Heiko Roth talks workforce capability graphs.
Cloud-to-cloud private links are a different product line; see the AWS–Azure Interconnect preview. Checking a graph model does not need a full meeting suite: in figmeet a room ID is enough for voice plus a board; see the guide and features.
# Public markers (Google Cloud blog, 2026-09-01)
Preview: BigQuery Graph (GQL + SQL, no ETL)
GA: native warehouse graph + faster GQL
Also cited: Iceberg via Unity Catalog / Glue / Snowflake (borderless)
Agent path: conversational analytics, MCP, Agent Analytics context graphGoogle compresses the job to one engine for two tasks: warehouse-scale graph analytics, and connected context for agents. Customer quotes are theirs, not your benchmark.
- BigQuery Graph
- Property-graph capability inside BigQuery. GA announced 1 September 2026; preview already offered GQL beside SQL.
- GQL
- ISO graph query language. Google cites path-finding work on acyclic and undirected traversals.
- Borderless Lakehouse
- A virtual graph over BigQuery and Iceberg in other clouds via Unity Catalog, Glue, or Snowflake; the vendor says data does not move.
Questions worth checking
Does BigQuery Graph replace standalone stores such as Neo4j?
Google writes about dropping extract-to-graph-silo overhead, not about another vendor shutting down. Buys still compare workload, latency, and ops model.
Does 1 September mean every agent feature is GA?
No. The post separates engine GA from the agent ecosystem; authoring skills and some pieces remain preview or coming soon.
Are 2× and 100× an SLA?
Those are vendor numbers versus preview on public benchmarks. Contract terms follow console legal text.
Must we copy data into BigQuery before we can have a graph?
Native tables live in-warehouse. The borderless story allows in-place Iceberg traversal; cross-account auth and billing follow official docs.