11-19 更新 实时
首页/ 技术栈/ 正文

Databricks 获评 2026 Gartner 数据湖仓平台魔力象限领导者

技术栈 来源:Databricks Blog 2026-11-15 02:15 0 阅读 机器翻译
Databricks 获评 2026 Gartner 数据湖仓平台魔力象限领导者

Databricks 在 Gartner 云运营数据库管理系统魔力象限中被评为唯一的「远见者」(Visionary),这一有力的竞争定位彰显了我们对运营数据库未来的战略愿景。

Lakebase 是一种事务型数据库,它把数据以开放格式保存在数据湖上,因此智能体可以按需创建和销毁数据库,其他引擎也能使用同一份数据。Databricks Lakebase Postgres 是这种新型数据库的首个商业实现,并被用作所有应用的记录系统:包括新的智能体式工作负载、现代应用以及传统事务型系统。

AI 正在把分析数据与运营数据拉到一起

几十年来,企业一直把数据运行在两个彼此割裂的世界里。数据仓库或湖仓负责分析,传统 OLTP 数据库运行应用,而管道在两者之间搬运数据,沿途产生副本、延迟以及各自为政的治理。

在仪表盘与应用彼此分离的时代,这种架构尚能奏效。AI 改变了这一点。一个智能体要回答客户问题、批准交易或更新预测,就需要同时获得新鲜的运营数据和深度的分析上下文,而且往往还需要写回。每多一条管道都会增加延迟、成本和风险,因此这两套系统之间的边界已成为瓶颈。

这两个市场在很大程度上是各自独立评估的。Gartner 刚刚发布了数据湖仓平台魔力象限和云运营数据库管理系统魔力象限。不过在我们看来,客户越来越需要这两套系统合为一体来运作。

湖仓:一个由我们开创、如今已获得行业认可的品类

当我们推出湖仓时,理念很简单:为每一种数据负载提供统一、开放且可治理的基础底座,而不是把数据湖和数据仓库当作彼此分离的系统。如今,数千家组织在这一底座上运行数据工程、数据仓库、流处理、BI 和 AI。

我们相信,我们在 Gartner 首个数据湖仓平台魔力象限中被评为领导者、并在执行能力上获得最高分,这正反映了客户每天所依赖的东西:

AI 原生。Genie 让任何人都能用自然语言向数据提问,并以在 Unity Catalog 中一次性定义的、受治理的业务语义为依托。AI Functions 可直接在受治理的数据上运行推理,AI 还会自动优化湖仓。

开放为设计之本。数据以开放格式(Delta Lake 和 Apache Iceberg)存储,因此客户不会被锁定在专有引擎之中。

统一治理。Unity Catalog 对数据与 AI 施加同一套权限、血缘和质量控制。

性能与成本。Serverless 计算与持续的引擎改进,把数据仓库级的性价比带到了湖仓。

实时。Lakehouse//RT 把低延迟、高并发的分析能力直接带给湖仓数据。

我们已经从客户身上看到了这样的影响。Airwallex 和 Volkswagen Brasil 都把系统从 Google BigQuery 整合迁移到 Databricks Lakehouse,分别实现了 3-4 倍的洞察提速和 25% 的 TCO 降低。例如 Nubank 每月运行超过 5000 万次查询,P90 查询性能提升 7 倍,成本降低 40% 以上。

湖仓最初并不是作为一个品类出现的,而是一种不同的构建方式。客户接受了它,生态围绕它建立起来,市场随之跟进。这正是我们如今再次踏上的路径。

lakebase:重新定义构建在湖之上的运营型数据库

湖仓在开放、低成本的云存储之上重建了分析能力。lakebase 则为运营型数据库做同样的事。我们相信,它将成为所有应用的记录系统(system of record):新的智能体式工作负载、现代应用,以及传统的事务型系统。

在 2026 年 Gartner Magic Quadrant for Cloud Operational Database Management Systems 中,Databricks 是唯一入选 Visionary(远见者)的厂商,而此时距 Lakebase Postgres 正式可用还不到一年。我们认为,这是该品类走向的一个信号。

如今在用的多数运营型数据库是为另一个时代设计的。存储与计算紧密耦合,因此要为峰值容量付费。数据库是沉重的资产,团队很少复制,而要把数据输送给分析或 AI,就意味着要构建和维护管道。它们并非为智能体编写和运行的软件而设计。

Lakebase Postgres 从不同的假设出发。它就是 Postgres,因此开发者可以继续使用自己已经熟悉的工具,但计算与存储分离,数据存放在云对象存储上,与湖仓基于同一基础。计算是无服务器的,并且可以缩容到零。数据库可以像代码一样即时分支,因此开发者和智能体能在几秒内创建出隔离的副本。而且由于 Lakebase Postgres 原生运行在 Databricks Data + AI Platform 之上,其数据与其他所有数据一并在 Unity Catalog 中受到治理。

客户已经在 lakebase 上运行生产工作负载。PRADA Group 用它把旗下各品牌的数千个可信 KPI 以低延迟 API 的形式对外提供,交付速度提升约 130 倍(从约 2 秒缩短到 15 毫秒),同时将基础设施成本降低约 50%。

“Lakebase 让我们从传统的基于查询的 KPI 交付方式,转向低延迟服务层,每个指标都以毫秒级 API 的形式对外暴露。”——Michele Piunti,PRADA Group 数据平台经理

LTAP:面向运营与分析的统一智能体式数据基础

把运营数据引入分析,一直意味着要复制数据。重量级的分析查询无法与对延迟敏感的事务在同一台机器上运行而不造成干扰,因此团队会加设只读副本,或搭建 CDC 管道把数据送入数据仓库。他们为额外的数据副本付费,等待脆弱的管道,最终还要治理同一份数据的两个版本,而两者的权限逐渐脱节。这正是运营数据库与分析平台长期以来被当作两个独立市场来采购、运行和评估的一个重要原因。

由 lakebase 支持的湖事务/分析处理(LTAP)架构,在存储层将两者统一起来,而不是强迫一个引擎同时承担这两项工作。Lakebase Postgres 以完整的 ACID 保证处理事务,湖仓引擎则负责分析。两者读取的都是同一份持久化数据副本,采用 Delta Lake 和 Apache Iceberg 等开放格式,并在 Unity Catalog 中统一治理。无需构建管道,也无需镜像表,分析查询绝不会拖慢应用。分析可以即时读取应用刚刚写入的最新数据,没有任何延迟。运营数据与分析数据就是这样在同一套基础上协同工作的。

我们已经看到客户从这种统一中获得了价值。Posti 已在 Databricks Lakehouse 上运行其分析工作负载,并计划使用 Lakebase Postgres 在同一个受治理的平台上支撑 GIS 和运营规划,因此运营类应用与分析不再分处不同的技术栈。

如需更深入了解 LTAP 的工作原理,请阅读我们的博客文章。

几年前,湖仓还只是一个需要我们向外界解释的概念。如今它已成为一个市场品类,我们认为 Lakebase 正处于同样转变的起点。现在写入数据库的不再只有应用程序。智能体正以传统系统在设计之初未曾考虑的速度和规模创建、分支和查询数据库,而它们需要存放在湖仓中的分析上下文。

我们正朝着一个开放、受治理的基础迈进:无论是运营类还是分析类工作负载,无论由人还是由智能体构建,都运行在同一份数据之上。湖仓是我们与客户共同打造的,Lakebase 我们也将以同样的方式构建。

Gartner,《数据湖仓平台魔力象限》,Aaron Rosenbaum、Michael Gonzales、Masud Miraz、Ramke Ramakrishan、Xingyu Gu、Robin Schumacher、Prasad Pore,2026 年 11 月 4 日。

Gartner,《云运营数据库管理系统魔力象限》,Masud Miraz、Aaron Rosenbaum、Ramke Ramakrishan、Xingyu Gu、Michael Gonzales、Prasad Pore、Robin Schumacher,2026 年 11 月 12 日。

Gartner 不为其研究出版物中涉及的任何厂商、产品或服务背书,也不建议技术用户仅选择那些评级最高或获得其他称号的厂商。Gartner 研究出版物由 Gartner 研究咨询机构的观点构成,不应被解读为事实陈述。Gartner 对本研究不作任何明示或暗示的担保,包括对适销性或特定用途适用性的任何担保。

GARTNER 和 MAGIC QUADRANT 是 Gartner, Inc. 和/或其关联公司在美国及国际上的注册商标,本文经许可使用。保留所有权利。

本图表由 Gartner, Inc. 发布,是其一份更大研究报告的一部分,应结合整份文档的语境进行解读。该 Gartner 文档可向 Databricks 索取。

在您的收件箱中获取最新文章

译文由英文原文机器翻译,术语按行业惯例统一,可在文末展开原文核对。

查看英文原文(Databricks is a Leader and positioned highest in execution in the 2026 Gartner® Magic Quadrant™ for Data Lakehouse Platforms)

Databricks is named as the only Visionary in the Gartner Magic Quadrant for Cloud Operational Database Management Systems, a strong competitive placement that highlights our strategic vision for the future of operational databases.

A lakebase is a transactional database that keeps its data on the lake, in an open format, so agents can create and discard databases on demand and other engines can work with the same data. Databricks Lakebase Postgres is the first commercial implementation of this new database, and is used as the system of record for all applications: new agentic workloads, modern applications, as well as traditional transactional systems.

AI is pulling analytical and operational data together

For decades, enterprises have run their data in two separate worlds. A warehouse or lakehouse handles analytics. Traditional OLTP databases run applications. Pipelines move data between them, creating copies, delay, and separate governance along the way.

That setup worked while dashboards and applications lived apart. AI changes that. An agent that answers a customer question, approves a transaction, or updates a forecast needs fresh operational data and deep analytical context at the same time, and it often needs to write back. Each extra pipeline adds latency, cost, and risk, so the boundary between the two systems has become a bottleneck.

These markets are evaluated largely independently. Gartner just published a Magic Quadrant for Data Lakehouse Platforms and a Magic Quadrant for Cloud Operational Database Management Systems. We see that customers, though, increasingly need the two systems to work as one.

The Lakehouse: a category we pioneered, now recognized by the industry

When we introduced the lakehouse, the idea was simple. Give every data workload one open, governed foundation, instead of keeping data lakes and data warehouses as separate systems. Today thousands of organizations run data engineering, data warehousing, streaming, BI, and AI on that foundation.

We believe our position as a Leader in Gartner's first Magic Quadrant for Data Lakehouse Platforms, with the highest scores in execution, reflects what customers rely on every day:

AI-native. Genie lets anyone ask questions of their data in natural language, grounded in governed business semantics defined once in Unity Catalog. AI Functions run inference directly on governed data, and AI optimizes the lakehouse automatically.

Open by design. Data lives in open formats (Delta Lake and Apache Iceberg), so customers aren't locked into a proprietary engine.

Unified governance. Unity Catalog applies one set of permissions, lineage, and quality controls across data and AI.

Performance and cost. Serverless compute and continued engine improvements bring warehouse-class price-performance to the lakehouse.

Real time. Lakehouse//RT brings low-latency, high-concurrency analytics directly to lakehouse data.

We’ve seen the impact with our customers. Airwallex and Volkswagen Brasil both consolidated from Google BigQuery to Databricks Lakehouse, and have seen 3-4x faster time to insight and 25% reduced TCO respectively. Nubank, for example, runs more than 50 million queries a month with 7x better P90 query performance and more than 40% lower costs.

image1.png

The lakehouse didn't begin as a category, but as a different way to build. Customers adopted it, the ecosystem built around it, and the market followed. That's the path we're on again.

The lakebase: Redefining operational databases on the lake

The lakehouse rebuilt analytics on open, low-cost cloud storage. A lakebase does the same for operational databases. We believe it will serve as the system of record for all applications: new agentic workloads, modern applications, as well as traditional transactional systems.

In the 2026 Gartner Magic Quadrant for Cloud Operational Database Management Systems, Databricks is the only vendor named a Visionary, less than a year after Lakebase Postgres became generally available. We see that as a signal of where the category is going.

Most operational databases in use today were designed for a different era. Storage and compute are tightly coupled, so capacity is paid for at peak. Databases are heavy assets that teams seldom duplicate, and getting data to analytics or AI means building and maintaining pipelines. They weren't designed for software that agents write and run.

Lakebase Postgres starts from different assumptions. It's Postgres, so developers keep the tools they already know, but compute is separated from storage and data lives on cloud object storage, the same foundation as the lakehouse. Compute is serverless and scales to zero. Databases branch instantly, like code, so developers and agents can spin up isolated copies in seconds. And because Lakebase Postgres is native to the Databricks Data + AI Platform, its data is governed in Unity Catalog alongside everything else.

Customers are already running production workloads on a lakebase. PRADA Group uses it to serve thousands of trusted KPIs across its brands as low-latency APIs, delivering them about 130x faster (from roughly 2 seconds to 15 milliseconds) while cutting infrastructure costs by about 50%.

“Lakebase allowed us to move from traditional query-based KPI delivery to a low-latency serving layer where every metric is exposed as an API in milliseconds.”- Michele Piunti, Data Platform Manager, PRADA Group

LTAP: a unified agentic data foundation for operations and analytics

Getting operational data into analytics has always meant copying it. A heavy analytical query can't run on the same machine as latency-sensitive transactions without causing disruptions, so teams add read replicas or build CDC pipelines into a warehouse. They pay for extra copies, wait on brittle pipelines, and end up governing two versions of the same data whose permissions drift apart. That's a big reason operational databases and analytics platforms have long been bought, run, and evaluated as separate markets.

The Lake Transactional/Analytical Processing (LTAP) architecture, enabled by the lakebase, unifies the two at the storage layer instead of forcing one engine to do both jobs. Lakebase Postgres handles transactions with full ACID guarantees, and lakehouse engines handle analytics. Both read a single durable copy of the data in open formats like Delta Lake and Apache Iceberg, governed in Unity Catalog. There's no pipeline to build and no tables to mirror, and analytical queries never slow down the application. Analytics reads instantly fresh data the application just wrote, with no delays. That's how operational and analytical data work on one foundation.

We’re already seeing customers achieving value from this unification. Posti already runs its analytics on Databricks Lakehouse, and plans to use Lakebase Postgres to power GIS and operational planning on the same governed platform, so operational applications and analytics no longer live in separate stacks.

For a deeper look at how LTAP works, read our blog post.

A few years ago, the lakehouse was an idea we had to explain. Today it's a market category, and we think the lakebase is at the start of the same shift. Applications aren't the only thing writing to databases anymore. Agents now create, branch, and query databases at a speed and scale traditional systems weren't designed for, and they need the analytical context that lives in the lakehouse.

We're building toward an open, governed foundation where every workload, operational or analytical, built by people or by agents, runs on the same data. We built the Lakehouse together with our customers, and we're building the Lakebase the same way.

Gartner, Magic Quadrant for Data Lakehouse Platforms, Aaron Rosenbaum, Michael Gonzales, Masud Miraz, Ramke Ramakrishan, Xingyu Gu, Robin Schumacher, Prasad Pore, November 4, 2026.

Gartner, Magic Quadrant for Cloud Operational Database Management Systems, Masud Miraz, Aaron Rosenbaum, Ramke Ramakrishan, Xingyu Gu, Michael Gonzales, Prasad Pore, Robin Schumacher, November 12 2026.

Gartner does not endorse any vendor, product or service depicted in its research publications, and does not advise technology users to select only those vendors with the highest ratings or other designation. Gartner research publications consist of the opinions of Gartner's Research & Advisory organization and should not be construed as statements of fact. Gartner disclaims all warranties, express or implied, with respect to this research, including any warranties of merchantability or fitness for a particular purpose.

GARTNER and MAGIC QUADRANT are registered trademarks of Gartner, Inc. and/or its affiliates in the U.S. and internationally and are used herein with permission. All rights reserved.

This graphic was published by Gartner, Inc. as part of a larger research document and should be evaluated in the context of the entire document. The Gartner document is available upon request from Databricks.

Get the latest posts in your inbox

原文为英文,正文由自动抓取并保留原文。

本文由采集器自动同步

原文发布于 Databricks Blog(Databricks is a Leader and positioned highest in execution in the 2026 Gartner® Magic Quadrant™ for Data Lakehouse Platforms),标题、摘要与正文由机器翻译,术语按行业惯例统一,未经人工逐句校订。

阅读原文 ↗