ボーダーレス Lakehouse：AWS・Databricks・Snowflake のデータを AI エージェントへThe borderless Lakehouse: Bring AWS, Databricks and Snowflake data to your AI agents

Google Cloud has introduced what it calls the borderless lakehouse, a concept aimed at letting autonomous AI agents query and act on data that resides across competing platforms, including AWS, Databricks, and Snowflake, without first moving or copying it. The announcement matters because it reframes the data lakehouse from a passive storage layer into a foundation for continuous, agent-driven automation that spans multiple clouds, addressing a persistent problem for enterprises whose information is scattered across different vendors and regions.

At the center of the pitch is a shift in what a lakehouse is meant to do. Traditionally, a lakehouse combined the low-cost, flexible storage of a data lake with the structure and query performance of a data warehouse. Google Cloud frames the borderless lakehouse as a "system of action" rather than merely a system of record. Instead of producing static reports that people read after the fact, the company describes always-on, autonomous AI agents that run continuous, real-time reasoning loops, monitoring processes such as supply chains and executing tasks as conditions change.

The distinguishing technical claim is that these agents can operate on data where it already lives. Rather than building fragile pipelines that duplicate datasets into a single warehouse, the approach appears to rely on federated access, letting queries reach into external systems like Snowflake tables, Databricks environments, or storage on AWS. Avoiding data movement is significant for several reasons: it reduces duplication costs, lowers latency, limits the governance risks that come from scattering copies of sensitive records, and respects what practitioners call "data gravity," the tendency for large datasets to stay put because moving them is expensive and slow. It can also help organizations keep data within specific regions to meet compliance requirements.

Much of this is likely enabled by open table formats, which have become a shared foundation across the industry. Apache Iceberg in particular has emerged as a common standard that Google Cloud, Snowflake, Databricks, and AWS all support to varying degrees, making it possible for multiple engines to read the same underlying files without proprietary lock-in. Google Cloud's existing components, such as BigQuery and BigLake, have progressively added support for querying open formats and external sources, and the borderless lakehouse appears to build on that direction rather than introducing an entirely separate stack.

The agent layer connects to Google's broader artificial intelligence strategy. The category tag points to Gemini, the company's family of large language models, which supplies the reasoning capability that turns a query engine into something that can interpret goals, plan steps, and decide when to act. For agents to behave reliably in production, they generally need governed access to trustworthy data, clear permissions, and observability so their actions can be audited. Those prerequisites are why a unified, cross-cloud data foundation is being positioned as the missing piece for agentic systems, though the practical maturity of such deployments will depend on how well governance and security controls extend across vendor boundaries.

The move fits a wider industry contest over where enterprise data and AI workloads sit. Microsoft has pushed a similar consolidation story with Fabric and its OneLake concept, while Snowflake and Databricks have each expanded from their original niches toward full lakehouse and AI platforms, and all major providers are racing to attach agent frameworks to their data offerings. Google Cloud's angle is notably cloud&tag=multi-cloud&entry=33939f16c6f8220c">multi-cloud in framing, emphasizing interoperability with rivals rather than requiring customers to migrate everything onto its platform first. That positioning is pragmatic, since most large organizations already run workloads across more than one provider and are unlikely to abandon existing investments in Snowflake or Databricks.

For readers evaluating the announcement, a few caveats are worth keeping in mind. Cross-cloud querying can introduce egress charges, performance variability, and additional security surface area, and vendor descriptions of seamless access often understate the engineering required to make it dependable at scale. The concept of autonomous agents acting on live data also raises questions about guardrails, human oversight, and error handling that remain active areas of development. Even so, the borderless lakehouse reflects a clear and consistent trend: data platforms are converging on open formats and increasingly treating AI agents, rather than dashboards, as the primary consumers of enterprise data.