Google's Agentic Data Cloud now lets agents query data anywhere — without moving it
Google Cloud's Agentic Data Cloud now lets companies query data across Amazon, Microsoft and Google clouds without moving it first — and Gemini can now ask questions of live company databases in plain language.
📊 Scorecard: our Bullish call on $GOOGL is scored from $284.78, the last close before it ran. Its first result comes after the next close.
The News
Google Cloud announced on November 4 an expansion of its Agentic Data Cloud — the set of tools that lets AI agents find, understand and use a company's data. Cross-cloud interconnect and caching for Amazon Web Services and Microsoft Azure are now generally available, so companies can query data sitting in those clouds without moving it first. The expansion also adds a borderless Lakehouse and a self-learning Knowledge Catalog — the self-learning enrichment is in preview — that grounds agents in business context, the background knowledge about what a company's data actually means.
On November 6, Google launched Data Cloud connectors in preview for Gemini for business, letting users query live data in BigQuery, Looker and Spanner, Cloud SQL and AlloyDB using plain language. The connectors run under the user's own identity and the database's existing permissions, so an agent can only see what that person is already allowed to see. That matters because the most useful company data is usually the most sensitive, and the question for every AI agent project is how to give the agent access without giving it the keys to everything.
The early evidence comes from one customer. Delivery Hero's talabat used the borderless Lakehouse to federate Iceberg data from AWS S3 into BigQuery, cutting data-movement costs by roughly 40%. That is a single-customer, vendor-reported figure — it shows what the setup can do, not what every company will get. These are November 4–6 announcements being caught up on today, not news from this morning.
Companies can now query data across clouds without moving it first.
Why It Matters
AI agents are only as useful as the data they can reach. A company that stores its sales data in Amazon's cloud, its analytics in Google's and its customer records in Microsoft's has a problem: every question an agent wants to answer means copying data between clouds, which is slow, expensive and risky. Letting the agent query the data where it lives removes that friction — and the Knowledge Catalog is meant to stop the agent from misunderstanding what the data means, which is how agents make their most embarrassing mistakes.
The permissions detail is the part that matters most for adoption. Companies have been slow to let agents near live databases because an agent with too much access can leak or misuse sensitive information. Running the connectors under the user's own identity and the database's existing permissions means the agent inherits exactly the access that person already has — no more, no less. That is the difference between a feature a security team will block and one it will approve. The connectors cover BigQuery, Looker and Spanner, Cloud SQL and AlloyDB — the main places Google Cloud customers keep live data — and they run under native user identities, so the permissions a person already has are the permissions the agent gets.
Analyst's Take
Our call is Bullish. The boring-sounding problem this solves — data stuck in the wrong cloud — is the thing that actually decides whether AI agents work in a real company. An agent that can only see one database is a demo; an agent that can query live data across clouds, under the same permissions as the person asking, is a tool a business can deploy. Google is betting that the company which makes its agents easiest to ground in real data wins the enterprise, and this is a credible step in that direction. Cross-cloud caching is the practical half of the announcement: instead of copying data between clouds, the system keeps frequently used results close to where they are needed.
What to watch: whether the Gemini connectors move from preview to general availability and how quickly other big customers follow talabat's example; whether Amazon and Microsoft answer with their own cross-cloud agent tooling; and whether the 40% cost figure holds up beyond a single showcase customer. The risk is the usual one for agent infrastructure: the demo is impressive, the rollout is slow, and the value only shows up when agents are trusted with real work. On the evidence so far, Google is moving faster than most.
Are you bullish or bearish on this story?
By The Numbers
FAQ
What is a data lakehouse?
A storage setup that combines two older ideas: a data lake, which stores huge amounts of raw data cheaply, and a data warehouse, which stores data in neat, structured tables ready for analysis. A lakehouse tries to give you both at once — cheap storage and fast, structured querying.
What does 'federate' mean here?
To query data wherever it lives as if it were in one place. Delivery Hero's talabat federated data stored in Amazon's cloud into Google's BigQuery, so analysts could ask questions across both without copying the data over first.
What is an AI agent?
Software that does multi-step work on its own — not just answering one question, but planning a task, calling tools and databases, and carrying it through, with a person stepping in when a decision is needed. An agent is only as good as the data it can reach, which is what this announcement is about.
What is the difference between preview and generally available?
Generally available means the feature is fully released and supported for all customers. Preview means it is an early release — usable, but still being tested and possibly changing. The cross-cloud features are generally available; the Gemini data connectors are in preview.
Is the 40% cost saving proven?
Not independently. It is a single customer's result — Delivery Hero's talabat — reported by the vendor, Google. It shows what is possible, not what every company should expect.
👁
Source: SiliconANGLE / Google Cloud Blog