谷歌云更新Agentic Data Cloud：无边界Lakehouse打通AWS与Azure跨云查询，Knowledge Catalog默认接入Gemini
谷歌云11月4日在其官方博客公布 Agentic Data Cloud 的一轮更新，核心是让人工智能智能体可以直接使用分散在多云、本地与软件服务中的企业数据。谷歌称，PayPal、西门子、Telus、沃达丰与沃尔玛等机构已在使用该平台。
本次更新的第一项是「无边界 Lakehouse」（borderless Lakehouse）的增强。该湖仓基于开放的 Apache Iceberg 格式，谷歌称其纳管的 Iceberg 数据量在过去一年已增至三倍以上；Gemini 现可直接在湖仓上工作，业务用户可用自然语言查询跨云存放的记录；面向 AWS 与 Azure 的跨云互联与缓存已正式商用，用户可通过私密安全连接跨云查询数据、无需支付可变出口费用，查询远程数据时系统会自动把取回的数据块缓存在谷歌云本地。湖仓还新增 AlloyDB 集成：AlloyDB 可直接查询云存储中的 Iceberg 表，无需抽取转换加载管线或复制数据，在一条 PostgreSQL 查询中把实时业务数据与历史湖仓数据连接起来，谷歌称其聚合查询速度最高可达亚马逊同类 PostgreSQL 兼容服务的7倍。Spanner Omni 已正式商用，把一致性与多模型能力带到本地、边缘与多云环境。
第二项是企业上下文层 Knowledge Catalog 的升级：新增自学习富化功能（预览版），可从 Google Workspace、GitHub、Jira、Confluence 与 SharePoint 等非结构化来源持续挖掘上下文；Knowledge Catalog 现已在 Gemini 中默认开启（预览版）；来自 BigQuery、Looker、Databricks、Snowflake、dbt 与 SAP 的可信语义模型接入已正式商用。谷歌称，按其内部基准，Knowledge Catalog 可使智能体准确率接近4倍、延迟最高降低33%、词元用量最高减少75%。客户案例方面，Delivery Hero 旗下 talabat 的工程师 Harish Ramesh 在博文中表示，采用无边界 Lakehouse 把 Amazon S3 表中的单份 Iceberg 数据联邦查询到 BigQuery 后，数据移动成本下降了约40%，流式交易事件在产生后数分钟内即可跨两朵云查询。
其他更新还包括：Data Agent Kit 已正式商用，把谷歌数据云工具接入 VS Code、Claude Code 与 Antigravity 等开发环境；AlloyDB Express 以预览版推出，面向智能体开发测试，支持2秒以内快速开通与缩容至零；BigQuery Graph 已正式商用，谷歌研究院的 TimesFM 3.0 与 TabFM 两个模型以预览版原生进入 BigQuery；谷歌称过去一年 BigQuery 查询速度提升逾40%、查询处理成本下降逾50%。
Google Cloud Updates Agentic Data Cloud: Borderless Lakehouse Opens Cross-Cloud Queries Across AWS and Azure as Knowledge Catalog Turns On by Default in Gemini
Google Cloud detailed a round of updates to its Agentic Data Cloud in a November 4 post on its official blog, centered on letting AI agents work directly with enterprise data spread across multi-cloud, on-premises and software-as-a-service environments. Google said organizations including PayPal, Siemens, Telus, Vodafone and Walmart are already using the platform.
The first set of updates enhances the borderless Lakehouse, built on the open Apache Iceberg format; Google said Iceberg data under management has more than tripled over the past year. Gemini now works directly on the Lakehouse, letting business users query records that live across clouds in natural language. Cross-cloud interconnect and caching for AWS and Azure are now generally available, allowing queries across clouds over a private, secure connection with zero variable egress fees, with retrieved data blocks automatically cached locally within Google Cloud. New AlloyDB integrations let AlloyDB query Iceberg tables stored in cloud storage directly, without ETL pipelines or copies, joining live operational data with historical lakehouse data in a single PostgreSQL query; Google said direct Iceberg access in AlloyDB is up to 7x faster on aggregation queries than Amazon's comparable PostgreSQL-compatible service. Spanner Omni is generally available, bringing consistency, scale and multi-model capabilities to on-premises, edge and multi-cloud environments.
The second set upgrades Knowledge Catalog, Google Cloud's context engine: self-learning enrichment, in preview, mines unstructured sources including Google Workspace, GitHub, Jira, Confluence and SharePoint to close context gaps; Knowledge Catalog is now on by default in Gemini, in preview; and grounding in trusted semantic models from BigQuery, Looker, Databricks, Snowflake, dbt and SAP is generally available. Based on its internal benchmarks, Google said Knowledge Catalog makes AI agents nearly 4x more accurate, with up to 33% lower latency and up to 75% fewer tokens. Harish Ramesh, a staff data engineer at Delivery Hero's talabat, said in the post that federating single-copy Apache Iceberg data from Amazon S3 tables to BigQuery reduced data-movement costs by approximately 40% while making streaming transactional events queryable across both clouds within minutes of production.
Other updates include general availability of the Data Agent Kit, which brings Google Data Cloud tools into development environments including VS Code, Claude Code and Antigravity; AlloyDB Express in preview, offering sub-2-second provisioning and scale-to-zero pause and resume for agent development and testing; BigQuery Graph in general availability; and Google Research's TimesFM 3.0 and TabFM models running natively in BigQuery in preview. Google said BigQuery queries have become over 40% faster in the past year while query processing costs have fallen by more than 50%.
—-
来源：谷歌云、T-Systems（公司
发布：GSM·美国（发布链接）
图片来源：数据中心机房（资料图）。（Wikimedia Commons）|A data center (file photo). (Wikimedia Commons)
GSM Global · 本篇英文版 English edition of this story
