Google lets AI assistants work across all of a company's data

Google Cloud has announced updates that let AI assistants work across a company's data, even when that data sits on Amazon or Microsoft cloud services. Google calls the package its Agentic Data Cloud. The pitch is simple: companies should not have to copy their records into one place before an AI tool can use them.

What changed

The main piece is what Google calls a borderless Lakehouse, a way to search data wherever it is stored. It is built on Apache Iceberg, an open format for organizing large data tables. Google says business users can now ask questions in plain language through its Gemini AI and get answers from records held across clouds.

Connections to Amazon and Microsoft cloud services are now generally available. Google says they use a private, secure link and carry zero variable egress fees. Egress fees are charges for moving data out of a cloud. Google also says the system keeps retrieved data nearby in Google Cloud so repeat searches are quicker.

Making AI answers more accurate

A second piece is Knowledge Catalog, which Google describes as a tool that gives AI assistants business context. It indexes scattered files and documents so an assistant knows what the data means. A new self-learning feature, in preview, mines sources such as Google Workspace, GitHub, Jira, Confluence and SharePoint. Google says people stay in control of governance.

Google claims, based on its internal benchmarks, that Knowledge Catalog makes AI agents nearly 4x more accurate. It also cites up to 33% lower latency, meaning faster replies, and up to 75% fewer tokens. Tokens are the small units of text AI services charge for.

Other pieces

• AlloyDB can now query Iceberg tables directly, with no copies, and Google says it is up to 7x faster than Amazon's comparable service on aggregation queries.
• Spanner Omni, now generally available, brings Google's database to on-premises, edge and multi-cloud setups.
• Open connections cover Databricks, Snowflake and AWS Glue, plus SAP, Workday and Salesforce.

What customers say

Google names PayPal, Siemens, Telus, Vodafone and Walmart as organizations seeing business impact. One quoted example is Delivery Hero's talabat. Its data engineer says the company cut data-movement costs by about 40%. He also says transaction events became searchable across two clouds within minutes.

What to keep in mind

The speed, accuracy and cost numbers come from Google's own tests and a customer quote, not independent reviews. Some features, including self-learning in Knowledge Catalog, are still in preview. For a business that already uses AI tools, the practical question is whether data spread across several systems can be used without costly copying. Google says it can.