BigQuery Graph goes GA, plus Google's November data cloud update: Iceberg everywhere and Knowledge Catalog

On 4 November 2026 Google published its autumn update for what it now calls the Agentic Data Cloud: BigQuery, AlloyDB, Spanner, its lakehouse and the catalog that ties them together. Some items were announced earlier at Google's Gemini at Work event; others are new. Here is what changed, filtered to what an analyst or data engineer can use. Our BigQuery October update covers last month's SQL features.

BigQuery Graph, now GA

Some questions are about connections: which accounts sent money to which, which customers share an address, which products are bought together. In SQL these need chains of self-joins that are hard to write and read. BigQuery Graph lets you declare which tables are nodes and which are edges, then query them with graph patterns in GQL, the new ISO standard graph language. The data stays in your normal tables.

-- 1. Define the graph once, on existing tables
CREATE OR REPLACE PROPERTY GRAPH graph_db.FinGraph
  NODE TABLES (graph_db.Account, graph_db.Person)
  EDGE TABLES (
    graph_db.PersonOwnAccount
      SOURCE KEY (id) REFERENCES Person (id)
      DESTINATION KEY (account_id) REFERENCES Account (id)
      LABEL Owns,
    graph_db.AccountTransferAccount
      SOURCE KEY (id) REFERENCES Account (id)
      DESTINATION KEY (to_id) REFERENCES Account (id)
      LABEL Transfers
  );

-- 2. Who did Dana send money to?
GRAPH graph_db.FinGraph
MATCH
  (person:Person {name: "Dana"})-[own:Owns]->
  (account:Account)-[transfer:Transfers]->(account2:Account)
  <-[own2:Owns]-(person2:Person)
RETURN person.name AS owner, transfer.amount AS amount,
       person2.name AS transferred_to
ORDER BY person2.name;

The borderless lakehouse: Iceberg across clouds

Knowledge Catalog: context for AI agents

AI agents answer badly when they have to guess what "active customer" or "net revenue" means. Google's answer is Knowledge Catalog, and it gained several features:

Google says its internal tests show agents nearly 4 times more accurate with up to 75% fewer tokens when grounded this way. Treat vendor benchmarks with care, but the direction matches what Snowflake (Business Ontology) and Databricks (metric view relationships) launched in the same fortnight: business definitions are becoming the most important input for AI on data.

Forecasting and prediction in SQL

Two models from Google Research now run inside BigQuery, both in preview: TimesFM 3.0, which forecasts multiple related time series with no training, and TabFM, a foundation model for predictions on tables. Both are called from SQL, so an analyst can get a first forecast or a churn score without a machine-learning project.

Tools for developers

• Data Agent Kit (GA): Google Data Cloud tools inside VS Code, Claude Code and Antigravity through MCP, so coding agents can build and test pipelines without the console.
• Gemini for data engineers: describe the outcome and Gemini writes PySpark, opens a notebook, trains models and fixes failing pipelines.
• Operational reports: business users ask for a report, Gemini saves the query, and later runs cost no tokens.
• Managed Service for Apache Spark: faster start-up for Spark 4.0 at no extra cost.

What to try first

More on Google Cloud: BigQuery October 2026 update, BigQuery vs Snowflake and SQL window functions explained.

Sources: Google Cloud blog, "New capabilities to power agents at scale with Google's Agentic Data Cloud" (4 November 2026); BigQuery documentation, "Create and query a graph". Performance and accuracy figures are Google's own and were not independently verified.

Alessandro Ecclesie AgazziWritten by Alessandro Ecclesie Agazzi, analytics engineer in London. Updated 9 November 2026. About me