Google adds Gemini integration to borderless Lakehouse

Google has introduced a Gemini integration with its borderless Lakehouse data platform. The feature is in preview.

The move is intended to give Gemini direct access to data across multiple cloud environments and software platforms, without requiring organisations to move that data into Google Cloud first.

Under the new setup, users can connect sources including Databricks Unity Catalog and Snowflake Horizon through Lakehouse catalog federation within Gemini. Business users can then query records in natural language across clouds, while AI agents can use the same data to support reasoning and code generation.

The announcement comes as companies try to move beyond experimental chatbot projects and build autonomous AI agents that can act on business data spread across on-premises systems, software-as-a-service applications and rival cloud platforms.

Data access

Google also said a series of borderless Lakehouse features launched earlier this year are now generally available. They include support for Cross-Cloud Interconnect and cross-cloud caching for Amazon Web Services and Microsoft Azure; BigQuery support for Apache Iceberg V3 specifications; read-write interoperability on Apache Iceberg tables for BigQuery and Managed Service for Apache Spark; and support for AlloyDB lakehouse federation.

Other features now generally available include bi-directional caching with AWS Glue, Databricks Unity Catalog and Snowflake Horizon Catalog through Lakehouse runtime federation; integration with Knowledge Catalog; zero-copy data integration with Salesforce, SAP, Workday and Confluent Tableflow; and synchronisation of Databricks Delta Lake assets through the Lakehouse console using Dataflow.

At the centre of the update is Google's effort to address one of the main technical limits facing AI agents in large companies: context size. An agent may only be able to take in a relatively small amount of information at one time, but enterprise data tasks often require access to much larger datasets spread across several systems.

Google said its borderless Lakehouse is designed to let agents reason over petabyte-scale data while leaving that information in place. Cross-Cloud Interconnect provides private links between Google Cloud and AWS or Azure at 10 Gbps or 100 Gbps per connection, while cross-cloud caching keeps repeat reads inside Google Cloud to reduce latency.

Open formats

BigQuery now supports Apache Iceberg V3 specifications, including binary deletion vectors and sort orders. Google said this makes row-level updates and deletions more efficient because data files do not need to be fully rewritten for every change.

Google also said it has updated BigQuery's execution engine with a FlatBuffers-based Parquet reader and parallelised input and output processes to reduce processing overhead. The changes are intended to narrow the performance gap between open table formats and data stored natively in Google systems.

Read-write interoperability on Apache Iceberg tables for both BigQuery and Managed Service for Apache Spark means data teams can work from the same Iceberg tables in Google Cloud Storage. Google also highlighted graph support for Iceberg in BigQuery, which it said can be used to analyse relationship networks such as supply chains on open table datasets.

For operational workloads, AlloyDB lakehouse federation allows users to query live database records and historical warehouse data together through a single connection. Google said this avoids copying data while giving AI systems access to both current and historical information.

Metadata layer

Another part of the strategy focuses on metadata and catalogues. Google said the Lakehouse runtime catalog, based on the Apache Iceberg REST Catalog, provides bi-directional catalog federation with AWS Glue, Databricks Unity Catalog and Snowflake Horizon Catalog.

The runtime catalog also has a Hive Metastore-compatible endpoint, allowing engines such as Apache Spark, Trino and Presto to discover metadata without significant architectural changes. Google said its integration with Knowledge Catalog lets distributed data assets, including unstructured data in Cloud Storage, be indexed with business metadata and semantic definitions that AI agents can use.

Google also pointed to broader customer adoption of open lakehouse architecture. Customer use of Apache Iceberg has more than tripled since the beginning of 2025, according to the company.

It cited several customers, including Etsy, which Google said unified three petabytes of data with borderless Lakehouse and improved pipeline query speeds by 60%. Google also pointed to Commerzbank, Hightouch and Roundel as users of the technology.

For existing users of Dataproc Metastore, Google said it now offers an automated migration path to the serverless Lakehouse runtime catalog. The process preserves schemas and lineage while moving customers to the newer data management layer, according to the company.