Google builds Agentic Data Cloud

Google is trying to move the AI conversation one layer down — from flashy agents to the plumbing those agents need to not break things. That is the point of its new “Agentic Data Cloud,” unveiled at Google Cloud Next on April 22 in Las Vegas. The pitch is simple enough: if AI agents are going to do real work inside big companies, they need access to the right data, the right definitions, and the right guardrails. Otherwise they are just confident interns with API keys. (cloud.google.com) ### What is Google actually launching? It is not one product. It is a bundle. Google wrapped together BigQuery, its databases, metadata tools, governance features, and cross-cloud data access under one label — Agentic Data Cloud. The idea is to turn enterprise data from something people query into something software agents can use to reason, decide, (cloud.google.com) stack should feed workflows, not just dashboards. (cloud.google.com) ### Why does the data layer matter so much? Because agents are only useful if they understand context. A sales agent needs to know what “margin” means inside that company. A support agent needs customer history, product status, and policy rules. An operations agent needs current inventory, not last night’s export. If those definitions live in ten sys(cloud.google.com)neage, business glossaries, permissions — is what keeps agentic AI from turning into expensive chaos. (cloud.google.com) ### What are the concrete new pieces? The most important ones are cross-cloud lakehouse access and a stronger metadata layer. Google says BigQuery Omni can now analyze data across Google Cloud, AWS, and Azure storage, including Amazon S3 and Azure Blob. It also introduced Knowledge Catalog, which pulls together technical metadata, business definition(cloud.google.com)s also pitching tools across AlloyDB, Cloud SQL, Spanner, and Bigtable as part of the same agent-ready foundation. (cloud.google.com) ### Why is cross-cloud such a big deal? Because almost no large company keeps all its useful data in one place. Some of it sits in old warehouses. Some of it sits in AWS. Some of it is trapped in operational databases. Some of it is still on-prem. Traditional analytics tools could tolerate that mess because humans stitched the answers together. Agent(cloud.google.com)hat the winning AI platform will be the one that spans the mess instead of demanding a full migration first. (cloud.google.com) ### Is this just branding? Partly, yes. Google already had many of these products. What changed is the way they are being packaged and prioritized. At Next ’26, Google also rolled out a Gemini Enterprise Agent Platform and pushed an “agentic enterprise” story across infrastructure, security, and productivity software. Agentic Data Cloud is the data p(cloud.google.com)r buying more agents. (cloud.google.com) ### Who is this really for? Big enterprises with fragmented data estates and serious compliance worries. Google highlighted customers like Vodafone and talked up auditability, lineage, and security by design. That tells you where the pressure is coming from. The blocker is no longer “can an LLM answer a question?” The blocker is “can an agent take action across finance, operations, and customer systems without making something untraceable?” (cloud.google.com) ### What is the catch? The catch is that this only works if companies do the boring cleanup. Metadata has to be accurate. Permissions have to be mapped. Definitions have to be standardized. Cross-cloud access sounds elegant, but it does not erase messy source systems. In other words, Google is selling a faster path to agentic AI — but also reminding customers that the hard part is still their data estate. (cloud.google.com) ### So what is the bottom line? Google is betting that enterprise AI will be won in the data layer, not just the model layer. That feels right. The next fight is not over who can build an agent demo. It is over who can make agents reliable enough to touch real business processes. (cloud.google.com)