Databases

UKG unlocks real-time workforce intelligence at scale with the Agentic Data Cloud

Wed, 29 Apr 2026 16:00:00 +0000

At UKG, we’ve spent years building and expanding our human capital management (HCM) and workforce management (WFM) solutions with new products, capabilities, and a series of acquisitions. Our cloud platform includes a suite of connected systems that support every corner of the employee experience, including scheduling and workforce operations, HR and payroll, and culture and engagement tools.

These connected tools offer customers incredible depth, but it also means our backend reflects years of evolution. We have 126 application teams, dozens of tech stacks, and more than 12,000 database instances inherited through acquisitions and product growth. And each product carries its own schema and operational footprint.

Previously, data moved through bespoke pipelines not built for real-time use. As AI advanced, expectations did too. Customers wanted instant insights across HR, time, pay, culture, and operations, and those insights increasingly needed to drive automated workflows and intelligent applications.

Internally, teams needed consistent, high-performance access to shared data to innovate faster and modernize our architecture. We needed a unified foundation for the next generation of intelligence across our suite. That’s why we built People Fabric, our new data and intelligence platform powered by AlloyDB for PostgreSQL and the just-announced Agentic Data Cloud.

Unifying the systems behind the suite

People Fabric started with a simple need: bring the full UKG suite onto one real-time foundation. Getting there started with defining a single canonical data model for the entire suite. This would serve as the shared language for people, work, pay, and culture data — consistent no matter where the information originated.

We needed an operational database that could ingest changes quickly and scale horizontally. That’s why we chose AlloyDB as the core of People Fabric. It gives us millisecond-level read-after-write behavior, high-throughput ingestion, scalable read pools, and native vector capabilities to support AI.

With the model defined and the operational store selected, the next step was building the pipeline that feeds the platform. We created a custom change data capture (CDC) framework to extract changes from our existing operational databases inherited over the years. Those changes flow through Dataflow, where they’re transformed into the canonical structure that AlloyDB for PostgreSQL expects.

Once in AlloyDB, that data becomes the real-time backbone of the platform. Applications use it for near-instant queries. AI agents rely on it for cross-domain decisions, and vector search engines use it to power natural-language and similarity-based experience layers.

For larger analytical workloads, the same data flows into BigQuery, which gives our teams and our customers the ability to perform organization-wide reporting and analysis without straining the system. Cloud SQL holds the metadata and tenancy context that govern who can see what and how different parts of the suite interact with People Fabric. From there, the system runs continuously. Data enters through streaming ingestion and gets modeled once in AlloyDB for PostgreSQL to make it available everywhere.

aside_block: <ListValue: [StructValue([('title', 'Build smarter with Google Cloud databases!'), ('body', <wagtail.rich_text.RichText object at 0x7f8981de0b80>), ('btn_text', ''), ('href', ''), ('image', None)])]>

Bringing people intelligence to intelligent people

With the architecture in place, People Fabric gives us something we never had before: a complete and consistent view of people, work, pay, and culture data that’s updated continuously and ready for AI to use in real time.

That unified context is what powers our assistive experiences, including conversational reporting and natural-language interactions. Leaders can ask questions in plain English and get answers that reflect the full picture — not just a single system’s slice of it.

With the power of Google’s Agentic Data Cloud, our platform unifies analytical and transactional data to power real-time AI. This allows agents to reason over live workforce signals and trigger immediate actions. Because this data is governed and modeled from the start, our agents can reliably handle multi-step workflows across HR, payroll, and timekeeping.

Whether they're identifying pay discrepancies, adjusting schedules, or flagging compliance risks, they operate with the same shared semantics and security model that guides our applications. It’s the difference between AI that reacts and AI that can truly assist.

Driving impact across every layer

For engineering teams, People Fabric acts as a database-as-a-service that removes the need for each microservice to manage its own datastore or pipelines. This accelerates development and supports modernization without customer disruption.

AlloyDB for PostgreSQL delivers millisecond read-after-write behavior, zero replication lag, and near-real time ingestion latency, enabling real-time workloads with far less complexity. Migrating core person and employment data off our on-prem monolith has generated cost savings significant enough to fund half of People Fabric.

Real-time operational data now gives managers a live view of staffing, pay, and workforce activity. More than 1,000 organizations are already on the platform, with another 1,000 in progress.

As we continue expanding People Fabric, we’re laying the groundwork for deeper agentic automation, more responsive analytics, and a growing set of AI-driven capabilities — all on a trusted, scalable foundation built for what’s next.

Learn more

UKG’s success illustrates how leveraging AlloyDB for PostgreSQL and the Agentic Data Cloud allows organizations to unify operational and analytical data, creating the essential foundation for real-time, agentic AI.

Learn more about AlloyDB for PostgreSQL and get started with a free trial today!
Sign up for a strategy workshop today on how to get your data ready for the agentic era!

What’s new with Google Data Cloud

Thu, 23 Apr 2026 16:00:00 +0000

April 20 - April 24

Google-built ODBC Driver for BigQuery is now available in Preview
We are excited to announce the launch of the new, Google-built ODBC driver for BigQuery. This new open-source driver provides a direct, high-performance connection for applications to BigQuery and is developed entirely in-house by Google. Download a new driver and connect your application to BigQuery.

April 13 - April 17

We announced we are reintroducing Data Studio to play a significant role in the AI era, expanding from data visualizations and reports to host BigQuery conversational agents and data apps built in Colab notebooks.
We announced BigQuery Graph is now available in preview, offering an easy-to-use, highly scalable graph analytics solution, empowering data professionals to model, analyze and visualize massive-scale relationships in an entirely new way.

April 6 - April 10

We introduced Conversational Analytics for Looker Embedded environments, enabling users to add natural language experiences to their own custom data-driven applications, powered by Gemini.
We expanded Looker’s capabilities for faster ad-hoc analysis, with the introduction of self-service Explores, enabling you to bring your own data to Looker’s semantic layer and gain instant access to insights in a governed data environment.

March 23 - March 27

We showed you how you can scale your reads with Cloud SQL autoscaling read pools. This feature allows you to provision multiple read replicas that are accessible via a single read endpoint and to dynamically adjust your read capability based on real-time application needs.
Our customers are leveraging the full power of Conversational Analytics and Looker to drive major business and technical breakthroughs in the AI era. Companies like Telenor, Pet Circle, Fluent Commerce, Lighthouse Intelligence, Wego, and ROLLER are turning data into insights and actions, grounded by Looker’s semantic layer.

March 16 - March 20

We introduced an enhanced Gemini assistant in BigQuery Studio, transforming the agent from a code assistant into a fully context-aware analytics partner.

February 23 - February 27

We introduced managed and remote MCP support for Google Cloud databases, including AlloyDB, Spanner, Cloud SQL, Bigtable and Firestore, to power the next generation of agents. This announcement extends the ability for AI models to plan, build, and solve complex problems, connecting to the database tools our customers leverage daily as the backbone of their work environment.
We outlined how you can build a conversational agent in BigQuery using the Conversational Analytics API to help you build context-aware agents that can understand natural language, query your BigQuery data, and deliver answers in text, tables, and visual charts.

February 16 - February 20

Our customers are leveraging the full power of Looker to drive major business and technical breakthroughs. Companies like Arrive, Audika, Carousell, Framebridge, GumGum, Intel, Overdose Digital, Ocean Network Express, Subskribe and Promevo are leveraging Looker’s newest AI-driven capabilities, including Conversational Analytics, to transform data to insights and actions, and empower their entire organization with a single source of truth, powered by Looker’s semantic layer.

February 2 - February 6

Join us on March 4 for our webinar, Win Your AI Strategy with Cloud SQL Enterprise Plus, to learn how to power your generative AI workloads with 3x higher performance and 99.99% availability. Register today to discover how to build a scalable, enterprise-grade foundation for your most demanding AI applications.

January 26 - January 30

We introduced Conversational Analytics in BigQuery, which allows users to analyze data using natural language. Conversational Analytics in BigQuery is an intelligent agent that generates, executes and visualizes answers grounded in your business context directly in BigQuery Studio, making data insights for data professionals more conversational.
We outlined how data products have become the foundation for AI agents, providing the context needed to make autonomous agents reliable and trusted for real business use, backed by organized business logic and semantic understanding.
We highlighted how you can supercharge data analytics workflows, and outlined Google Cloud’s AI agent offerings for data engineering, data science, and development tools, so you can integrate agentic workflows in your applications, empower your teams and speed discovery.

January 19 - January 23

We have fundamentally reimagined Firestore with pipeline operations for Enterprise edition. Experience a powerful new engine featuring over a hundred new query features, index-less queries, new index types, and observability tooling to improve query performance. Seamlessly migrate using built-in tools and leverage Firestore’s existing differentiated serverless foundation, virtually unlimited scale, and industry-leading SLA. Join a community of 600K developers to craft expressive applications that maximize the benefits of rich queryability, real-time listen queries, robust offline caching, and cutting-edge AI-assistive coding integrations.
Introducing Google Cloud SQL on MSSQLTips: We are highlighting a new technical guide published on MSSQLTips titled "Introducing Google Cloud SQL." This article serves as an essential resource for SQL Server administrators and developers exploring Google Cloud's fully managed database service. It provides a detailed overview of Cloud SQL capabilities, including high availability, security integration, and the seamless transition of on-premises SQL Server workloads to the cloud, making it an ideal resource for those planning their migration strategy.
We are excited to announce the Public Preview of Microsoft Entra ID (formerly Azure Active Directory) integration with Cloud SQL for SQL Server. Designed to tackle the challenge of identity sprawl in multi-cloud environments, this integration allows organizations to govern database access using their existing Microsoft identity infrastructure. Key benefits include centralized identity management, enhanced security features like Multi-Factor Authentication (MFA), and simplified user administration through direct group mapping. This feature is available for SQL Server 2022 and supports both public and private IP configurations.

January 12 - January 16

Google-built JDBC Driver for BigQuery is now available in Preview
We are excited to announce the launch of the new, Google-built JDBC driver for BigQuery. This new open-source driver provides a direct, high-performance connection for Java applications to BigQuery and is developed entirely in-house by Google. Download a new driver and connect your Java application to BigQuery.
Troubleshoot Airflow tasks instantly with Gemini Cloud Assist investigations: Cloud Composer just got smarter. We are excited to announce that Gemini Cloud Assist investigations are now available directly within Cloud Composer 3. Instead of manually sifting through raw logs, you can now simply click "Investigate" on a failed Airflow task. Gemini analyzes logs and task metadata to identify failure patterns—such as resource exhaustion or timeouts—and provides actionable recommendations driven by Gemini Cloud Assist to resolve the issue. This integration shifts the debugging experience from manual toil to automated root cause analysis, significantly reducing the time required to restore your pipelines. Learn more about AI-assisted troubleshooting.

Announcing Spanner Omni: Your infrastructure, Google’s innovation

Wed, 22 Apr 2026 12:00:00 +0000

Today, we announced the preview of Spanner Omni, a downloadable version of Spanner, that expands its industry-leading distributed database capabilities beyond Google Cloud. This enables enterprises to run Spanner in their own data centers, across clouds, or on a laptop — bringing its virtually unlimited scalability, high availability, strong consistency, enterprise-grade security, and multi-model capabilities to AI-enabled applications wherever they operate.

Why Spanner Omni is a big deal

Over a decade ago, Google pioneered the distributed SQL market with Spanner. It offered the "holy grail" combination: the horizontal scalability of NoSQL with the ACID (atomicity, consistency, isolation, durability) compliance and strong consistency of a traditional relational database. Since then, Spanner has evolved into an interoperable multimodal database, incorporating SQL, graph, key-value, full-text search, vector search, and analytical processing with columnar engine. This evolution offers the capabilities needed for the AI era and promises an opportunity to simplify and consolidate customers’ workloads — an advancement that has delivered significant value to our customers.

Customers often navigate across a range of IT environments, from public cloud and hybrid cloud to air-gapped deployments. Spanner Omni meets them where they operate to address the evolving needs of modern enterprises, including:

Business continuity: Mission-critical workloads often need truly resilient, high-availability architectures that extend beyond the edge of the cloud. These workloads need to keep running even when business conditions or world events compel changes in their cloud posture.
Regulatory compliance: Customers in highly-regulated industries, like financial services, must meet regulatory requirements like data sovereignty. While many still maintain large on-premises footprints to stay compliant, they’re eager to modernize for the agentic era.
Application portability: For SaaS providers and other independent software vendors (ISVs), growth depends on meeting customers wherever they are — whether in different clouds or a private data center. They want a consistent technology stack across those environments to minimize development and operational overhead without having to compromise on advanced features for innovation.

Introducing Spanner Omni

Spanner Omni delivers the same core capabilities as our fully-managed Spanner service, with the added freedom to deploy it wherever needed. It provides flexible configuration options ranging from virtual machines (VMs) and Linux containers to Kubernetes clusters. Customers can also run Spanner Omni in a wide range of configurations — on-premises, multicloud, multi-region and hybrid (within reasonable latency topologies), air-gapped, or connected — scaling from a single machine to clusters of thousands of servers. Our internal benchmark tests confirm that Spanner Omni processes millions of queries per second (QPS) across petabytes of data in a single regional deployment.

What makes Spanner Omni unique

Spanner Omni opens the door to new possibilities. Through our work with early adopters, we have observed three popular architectures for deploying Spanner Omni:

Hybrid and multicloud resilience: Customers are running their Spanner managed service in Google Cloud as their primary database while deploying Spanner Omni in a secondary cloud or on-premises data center as a hot-cold failover site. This primary-secondary architecture provides a critical safety net and helps organizations meet disaster recovery and business continuity requirements, especially in regulated industries. Our customers can also deploy multi-regional HA in jurisdictions where Google Cloud only has one data center, but data sovereignty is a requirement.
Unified tech stack across environments: For organizations with a multicloud or hybrid strategy, Spanner Omni provides a "write once, run anywhere" application building experience across environments. By standardizing on a single database layer, teams can drastically reduce their operational overhead and achieve true application portability across any environment, while tapping into cutting-edge database technologies.
On-premises modernization: Organizations with substantial on-premises commitment or preference to self-manage are no longer locked out of cloud-native innovation. With Spanner Omni, they can build or rewrite next-generation AI applications with Spanner’s multi-model capabilities on existing hardware infrastructure using Spanner’s innovative technology to achieve scale insurance, high availability, global consistency, and much more.

The tech behind Spanner Omni

Our vision is simple: If you have compute and a database-grade local file system, you should be able to run Spanner Omni. To make this a reality, we have made a series of innovations to remove Spanner’s dependency on Google. We took the core technologies of Spanner, such as the Paxos consensus, automatic sharding, synchronous replication, and replaced Google-dependent components like Colossus and TrueTime with new creative and innovative solutions.

For distributed storage everywhere, Spanner Omni replaces Spanner’s reliance on Colossus, Google’s distributed file system, by introducing a "Colossus-like" abstraction layer. This layer writes to attached local file systems and makes them available across the network to other nodes. The software automatically handles shard splitting and rebalancing to balance storage across all available servers to ensure peak performance. While Spanner Omni’s file layer isn’t Colossus, it’s a sufficient stand-in to allow Spanner Omni to perform comparably to the Spanner managed service for most workloads.

We also reimagined TrueTime, one of Spanner’s most well known technologies, which uses atomic clocks and GPS to synchronize time globally. For Spanner Omni, we developed a software-based TrueTime alternative, which — like TrueTime in Google Cloud — provides highly reliable, error-bounded time synchronization across servers in a Spanner Omni deployment. While Spanner depends on time synchronization to achieve strong external consistency, its ability to overlap time uncertainty waits with other database work allows Spanner to tolerate weaker uncertainty bounds than TrueTime provides in practice. TrueTime uses this flexibility to provide timekeeping across diverse and heterogeneous environments without limiting Spanner’s availability or performance.

aside_block: <ListValue: [StructValue([('title', 'What customers are saying'), ('body', <wagtail.rich_text.RichText object at 0x7f89706655e0>), ('btn_text', ''), ('href', ''), ('image', None)])]>

Where you can get access

Spanner Omni’s developer edition is available today in preview (download). This edition includes Spanner’s core capabilities — suited for developing and testing in non-commercial, non-production environments. While it excludes enterprise security features, it’s the ideal sandbox for your next project. For early access to the commercial edition with full features, please reach out to us at https://cloud.google.com/consulting/spanner-omni

Converging operational and analytical data for AI transformation

Wed, 22 Apr 2026 12:00:00 +0000

To act at the speed of business, AI agents must operate in fast and trusted reasoning loops. They need to “think” by reasoning across both your historical context and your live operational reality. Only by understanding this complete, real-time picture can they “do” — taking immediate action.

For decades, data architectures have been built with a structural wall that breaks this loop, separating the platforms that generate insights from the platforms that manage actions. This latency means insights are gleaned after the critical window for an agent to take action has closed. Achieving true AI transformation requires organizations to move from a passive system of record to a proactive System of Action, built on a closed-loop architecture that converges operational and analytical data.

At Google Cloud Next, we announced new unifying capabilities that drive our Agentic Data Cloud, eliminating silos and enabling 98% of our largest data cloud customers to run operational and analytical workloads in a unified data platform. By operating AlloyDB, BigQuery and Spanner together, we are delivering an AI-native architecture that unlocks the full potential of your data for real-time, agentic applications.

Flexible, real-time data agents

To act effectively, agents require both operational and historical signals for sound decision-making. Our Agentic Data Cloud bridges the gap between the operational "now" and analytical history by handling the complex plumbing for you. We provide diverse integration models across data federation, reverse ETL, and real-time ingestion to the lakehouse, empowering your agents to make high-stakes decisions with both live context and historical depth.

For example, sometimes an agent driving a live operational application needs to pull historical context on demand. Through Lakehouse federation for AlloyDB (Preview), agents can access Lakehouse data directly from AlloyDB itself. This allows frontline systems to instantly query extensive historical data without relying on brittle data movement pipelines.

In other scenarios, the challenge is reversed: deeply complex historical insights have already been calculated in the data warehouse, but an agent needs to deliver them to millions of users at conversational speeds. Reverse ETL for BigQuery (Preview) provides a one-click solution to push these heavy analytical insights back into AlloyDB, Bigtable, or Spanner, enabling agents to serve them with sub-millisecond latency.

One-click reverse ETL

Teams running real-time analytics on live operational data typically have to move that data into analytical systems — an error-prone process that introduces lag. With Spanner Columnar Engine (GA) users can perform analytical queries that run up to 200 times faster with zero impact on production transactional workloads.

Finally, the reasoning loop is not complete until an agent’s real-time action is captured for downstream analysis. To close this loop, Datastream for Lakehouse Apache Iceberg tables provides real-time Change Data Capture (CDC) from AlloyDB, Cloud SQL, Spanner, and Oracle directly into the open Lakehouse. This process streams every operational change as an append-only event into Lakehouse tables, making that data immediately available in BigQuery for ML model training, feature engineering, and real-time analytics.

"AlloyDB, along with other Google Cloud products like BigQuery, provides the agility and performance needed to continually enhance our platform's capabilities and help us anticipate emerging trends rather than merely reacting.” - Javi Fernández, CTO, Loyal Guru

Grounding agents in a unified governance foundation

Inconsistent definitions and unclear data ownership across operational and analytical systems can cause agents to hallucinate. To address this, we are extending Knowledge Catalog (Preview), formerly Dataplex, with new integrations for AlloyDB, BigQuery, Bigtable, Cloud SQL, and Spanner to provide a unified map of your data landscape. Integrations with Oracle AI Database@Google Cloud and Firestore are coming soon. The Knowledge Catalog works by aggregating native context across your Google and partner data platforms, semantic models, and third-party catalogs, unifying them into a single, governed source of truth needed to build and scale reliable agents.

"Seven-Eleven Japan created “Seven Central,” a scalable data platform that uses Spanner and BigQuery to provide real-time insights and support the company’s digital innovation strategies. We collect data from all 21,000+ stores, and in anticipation of a future expansion in business operations, we have designed a system that can scale up and run without issue, even if we were to have 30,000 stores, with 1,000 customers per store per day."
-Izuru Nishimura, Executive Officer and Head of ICT Department, Seven-Eleven Japan

Unified engines for deep reasoning

To move beyond simple Q&A chatbots to autonomous agents, AI must reason across every dimension of your data estate. Historically, combining keyword search, semantic understanding, and relationship mapping meant moving data out of operational databases and into specialized, siloed search engines — introducing latency and complexity.

Google’s Agentic Data Cloud eliminates these silos. By embedding native vector and full-text search directly into operational databases like AlloyDB, Bigtable, Cloud SQL, Firestore, and Spanner, agents can execute highly accurate hybrid searches combining keyword relevance and semantic intent.

We’re also bringing together graph and vector support across BigQuery and Spanner. With graph federation, an agent can match live user intent in Spanner and immediately trace that intent through historical graph relationships in BigQuery Graph — accelerating autonomous decision-making without moving the data. This multi-model approach powers advanced GraphRAG patterns, equipping agents with the rich, interconnected context required to accelerate autonomous decision-making.

“To deliver AI that actually works across HR, payroll, and workforce operations, you need a consistent, real-time data layer. With the power of Google’s Agentic Data Cloud, People Fabric is the backbone of UKG’s Workforce Operating Platform — turning fragmented systems into a single source of truth that powers intelligent, agent-driven experiences.”
-Radhi Chagarlamudi, Group Vice President, Product Engineering, UKG

Built for performance at agent scale

Our Agentic Data Cloud delivers the closed-loop architecture required for the AI era without compromising operational performance. Built on open standards like Iceberg and PostgreSQL, and governed by universal semantics, Google Cloud provides the speed, throughput, and trusted context needed to build the next generation of conversational and autonomous applications.

Build: Explore the AlloyDB AI documentation to start grounding your agents.
Connect: Visit the BigQuery Console to set up your first federated query to Spanner.
Govern: Opt-in to the Knowledge Catalog for unified visibility.

Oracle AI Database@Google Cloud: The foundation for the Agentic Enterprise

Wed, 22 Apr 2026 12:00:00 +0000

For decades, many of the world’s most critical enterprise datasets have relied on the performance of Oracle databases. Today, we are bringing that reliability even closer to the cutting edge. By enabling customers to run Oracle AI Database services natively within Google Cloud, we’ve bridged the gap between foundational data and the modern AI stack.

With the latest wave of upcoming launches for Oracle AI Database@Google Cloud, we aren't just making it easier to migrate; we are building a direct pipeline from your Oracle systems of record to the insight layer of Google Cloud. By bringing mission-critical data easily and securely to Gemini models and BigQuery, customers can transform static records into autonomous, agentic workflows.

New capabilities announced at Next ‘26

Here is a breakdown of the key new features designed to strengthen your Oracle-to-agentic- AI strategy:

New regions launched: We have significantly expanded the availability of Oracle AI Database@Google Cloud, across 15 regions (and 20 sites) globally. The recent rollout included key global hubs such as Milan, Iowa, São Paulo, Tokyo, Sydney, and Mumbai, among others. With additional regions like Mexico and Turin coming soon, this expansion ensures higher availability and lower latency for your mission-critical workloads across the globe for our Google Cloud customers.
Enhanced AI capabilities: This is the foundation for agentic AI. We are introducing the preview of Managed MCP Server for Oracle workloads, which allows agents like Gemini to interact directly and seamlessly with your Oracle infrastructure. Building on this, the new Oracle AI Database Agent, available in the Google Cloud AI Agent Marketplace, lets you talk to your Oracle data directly from Gemini Enterprise — no custom chatbot or NL-to-SQL solution required.
Database Center integration (Generally Available): To move at the speed of AI, your infrastructure must be healthy and visible. Database Center now supports Oracle AI Database@Google Cloud, providing a "single pane of glass" for your entire data estate. Whether you are running Exadata or Autonomous Database, you can now monitor your inventory and streamline operations through a unified experience within the Google Cloud console.
Knowledge Catalog integration (Preview): Data discovery is the first step toward intelligence. By extending the Knowledge Catalog to Oracle AI Database@Google Cloud, we are breaking down the walls between your Oracle systems and the Google Data Cloud. This allows for a unified governance and metadata layer, making it easier for customers to find, trust, and use Oracle data and provide context to AI agents.
OCI GoldenGate Service integration (Preview): Real-time data is the lifeblood of AI. This integration enables low-impact, continuous data movement, allowing you to streamline migrations from on-premises environments to Oracle AI Database@Google Cloud. In addition, it provides a live link to BigQuery, enabling operational data analytics that reflect the "here and now" of your business.
VPC Service Controls: Oracle AI Database@Google Cloud administrators can use VPC Service Controls to restrict access to the admin API and create databases within a service perimeter. VPC Service Controls protect businesses from unauthorized access outside the security perimeter, even if credentials have been compromised.

The agentic future

The goal of these integrations is simple: To make your data active. When your Oracle data resides natively in Google Cloud, Gemini doesn't just “talk about” your data — it can work with it. Whether it's an AI agent forecasting supply chain shifts in BigQuery based on live Oracle ERP feeds, or a customer service bot with real-time access to legacy account history, the data vault is more open, accessible, and valuable than ever (while remaining just as secure).

Hear directly from our customer, Banco Actinver, regarding the transformative impact of relocating their Oracle data to Google Cloud.

Get started

You can access Oracle AI Database@Google Cloud through the Google Cloud Marketplace using your existing Google Cloud account and billing.

For more information, visit:

What’s new with Databases: Powering the agentic future

Wed, 22 Apr 2026 12:00:00 +0000

In the agentic era, the role of enterprise data is shifting from a passive resource to a dynamic System of Action. That’s why we announced the Agentic Data Cloud — a unified architecture that integrates models, analytics, and operational databases into a single, AI-native system. This allows organizations to unlock the full potential of their data, ensuring every AI application and agent is grounded in truth and is capable of taking real-time action without costs spiraling.

Today, we’re moving the needle by:

Embedding AI across the entire data stack for an unmatched, optimized developer experience.
Unifying transactional and analytical worlds, to eliminate the friction of manual integration.
Simplifying enterprise-grade deployments with industry-leading, open databases that are easier to manage and faster to deploy.

Embed AI into every layer of the data stack

We continue to enable smooth connections between AI agents and enterprise databases, making your database experience effortless and agent-driven. From vibe coding tools for citizen developers, to high-scale vector search for the enterprise, we know that databases are one of the primary components of agentic workflows.

Key launches include:

Vibe coding integrations for databases with AI Studio: To enable creators, we are launching an integration with our leading vibe coding platform Google AI Studio. This agent-led, automated workflow allows you to create a live application from a simple text prompt in seconds, that connects the application to trusted database services like Firestore, with support for Cloud SQL for PostgreSQL coming soon.
Tools for Data Agents (Preview): These are the modular building blocks developers use to give their custom AI agents direct, secure access to database capabilities. Available for AlloyDB, Cloud SQL, and Spanner, these tools provide ready-to-use functions — like the QueryData tool for high-accuracy text-to-SQL— that allow any agent to query, understand, and interact with your data reliably. These tools for Data Agents in AlloyDB, Spanner, and Cloud SQL provide near-100% text-to-SQL accuracy.
Database onboarding and observability agents (Preview): The new Database Onboarding agent takes the guesswork out of database selection and deployment. By evaluating your stated requirements — from simple use case descriptions, to complex enterprise needs — it recommends the ideal Google Cloud database and guides you through provisioning. Additionally, the new AI-powered Database Observability Agent proactively monitors performance and health across AlloyDB, Bigtable, Cloud SQL, and Spanner. It identifies root causes of potential issues and delivers accurate remediations for troubleshooting.
AlloyDB AI-powered search at scale (Preview): We are introducing unprecedented scale and performance for enterprise search. AlloyDB can now scale to 10 billion vectors using Google’s ScaNN index, and offers up to 6 times faster vector queries than the HNSW index in standard PostgreSQL. Additionally, our HNSW index, accelerated by the Columnar Engine, now outperforms standard PostgreSQL by 4 times. We're empowering developers to build the ultimate hybrid search engine by pairing high-speed vector retrieval with industry-standard full-text search, optimized by native BM25 support coming soon.

AlloyDB vector search innovations

New and optimized AI functions for AlloyDB (Preview): AlloyDB AI functions allow you to integrate enterprise data with the power of LLMs like Gemini. For example, the ai.if function can replace custom logic to determine if user transactions are fraudulent. Joining our existing suite, ai.if, ai.rank, ai.generate, and ai.forecast, are two new additions: ai.analyze_sentiment and ai.summarize. We also now have an optimized mode for ai.if, which delivers superior cost and performance boosts.
Managed, remote MCP servers for databases (GA): We’re announcing managed, remote Model Context Protocol (MCP) servers for AlloyDB, Bigtable, Cloud SQL, Firestore, and Spanner, alongside a preview for Memorystore, Database Migration Service, Datastream, Database Center, and Oracle AI Database@Google Cloud to achieve MCP coverage across our portfolio. By fully managing the infrastructure required to connect AI models securely to your data, we are eliminating the operational burden of hosting, securing, and scaling MCP servers yourself. This ensures your AI models can reason and act upon your most up-to-date enterprise data with production-grade reliability.
Open-source MCP Toolbox for Databases (1.0): Supporting over 40 databases with contributions from 10 vendors, the open source, industry-standard MCP Toolbox for Databases has achieved a major stability milestone with this release. By guaranteeing the API will not break without a major version bump, we are giving you the confidence to build production applications, delivering a highly reliable foundation for both developers and autonomous AI agents.

Break down walled gardens with lakehouse integrations

To build real-time AI applications, developers need to combine live transactional context with massive historical insights at sub-millisecond latencies. Today, 98% of our largest data cloud customers are running their operational and analytical workloads on Google’s Agentic Data Cloud.

We’re announcing:

Lakehouse federation for AlloyDB (Preview): AlloyDB customers can now access live data from Iceberg and BigQuery from the PostgreSQL data plane. Users can discover and search for any BigQuery or Lakehouse for Apache Iceberg table directly from the AlloyDB Studio UI and begin querying immediately, with pushdown to BigQuery for filters and aggregations. This allows live joins between AlloyDB's transactional data and historical insights in BigQuery or Iceberg without any data movement.
Reverse ETL for BigQuery (Preview): Many customers use BigQuery and Lakehouse to combine structured warehouse data with rich context from the data lake using tools like Apache Spark or BigQuery ML. Now, with our new one-click Reverse ETL, customers can activate these insights and sync data from the lakehouse to AlloyDB with a single click. AlloyDB acts as a high-concurrency, low-latency serving layer, optimized to meet the needs of real-time and agentic applications. Excellent serving performance is powered by AlloyDB’s unique columnar engine and ultra-fast cache, available at no additional cost.
Seamless ongoing replication with Datastream (GA): To close the data loop, Datastream now allows users to replicate data continuously to BigQuery and, importantly, to Iceberg tables directly from AlloyDB. This ensures operational changes are quickly reflected in the analytical environment, crucial for real-time ML feature engineering. Datastream is simple to set up, fully serverless, and offers a free tier for AlloyDB to BigQuery streams.
Knowledge Catalog, formerly Dataplex (Preview): The Knowledge Catalog is a universal context engine which maps and infers business meaning across your entire data estate, using a rigorous framework of aggregation, continuous enrichment, and search. It works by aggregating native context across your Google and partner data platforms, semantic models, and third-party catalogs, unifying them into a single, governed source of truth. Governance is consistently applied, access controls are managed centrally, and full visibility and security are maintained across the entire data lifecycle.
Spanner Columnar Engine (GA): This capability accelerates analytical queries by speeding up scans up to 200 times on live operational data. By storing data in a columnar format alongside traditional row-based storage, Spanner can execute complex queries automatically using vectorized execution — processing batches of data at once rather than row-by-row. Spanner also now supports Iceberg tables, continuous reverse ETL from BigQuery, and accelerated federated queries — all using the Spanner columnar engine.
Database Center with BigQuery (Preview): Database Center now supports BigQuery alongside operational databases and self-managed databases on Google Compute Engine. You can now monitor your entire data estate — from the databases running your critical applications to the warehouses analyzing your business — from a single, intelligent management plane. Additionally, Gemini-powered fleet analytics now proactively surfaces performance optimization opportunities across your environment. And with our new API and managed MCP support, you gain the freedom to stream these rich fleet metrics directly into your preferred third-party tools and custom dashboards.

“Our partnership with Google Cloud is focused on building a smarter, faster exchange. Google’s Agentic Data Cloud allows us to dismantle the legacy silos and technical debt that once slowed us down. By integrating the operational reliability of Cloud SQL with the deep reasoning of BigQuery, we’ve created a data ecosystem where our developers and AI agents can validate, optimize, and innovate in real time.”
—Kristofer Shane Sikora, Executive Director, Cloud Data Engineering, CME Group

Commitment to open data and multi-cloud flexibility

We recognize how critical databases are for the day-to-day operations of all your applications. We continue to innovate on industry-leading reliability, price/performance, and scale, while maintaining a focus on open source and open data formats like Iceberg, MySQL, PostgreSQL, and Valkey.

To simplify enterprise-grade deployments, we’re announcing:

Spanner Omni (Preview): A new downloadable edition of Spanner that extends the industry’s leading distributed database beyond Google Cloud. Spanner Omni empowers organizations to use Spanner’s unparalleled scalability, high availability, strong consistency, enterprise-grade security and fully interoperable, multi-model capabilities for AI-enabled applications in their own data centers, across clouds or even at the edge.

What makes Spanner Omni unique

Bigtable in-memory (Preview): Bigtable is evolving its architecture and now delivers sub-millisecond read latency with the new memory tier. Available as part of Bigtable’s new Enterprise Plus edition, the new nodes with hybrid storage architecture spanning RAM, SSD and HDD, integrate frequently accessed data in-memory with long-term archival storage — all within a single, fully-managed service. Complete this form to express interest.
Memorystore for Valkey 9.0 (GA): Demonstrating our strong commitment to open source, this milestone anchors our focus on simplifying adoption and delivering enterprise-grade performance. We are introducing a managed migration path from self-managed Redis and Valkey to Memorystore, making it easier than ever to modernize. We’re also adding new smaller and larger node sizes for better price-performance, support for the Valkey-bloom and Valkey-json modules, and enterprise-grade security featuring Access Control Lists (ACLs), token-based authentication, and Flexible Certificate Authorities.
Oracle AI Database@Google Cloud: We are expanding to 20 regions, and are adding Oracle GoldenGate Service support, near real-time data replication to BigQuery, and integration with Knowledge Catalog (formerly Dataplex) and Database Center.
Compute Engine-to-managed migrations (Preview): This is a new experience designed to eliminate the complexity of moving databases. Natively integrated into Cloud SQL, AlloyDB, and Database Center, it provides automated networking and replication for PostgreSQL, allowing you to move your Compute Engine workloads to our powerful managed services with minimal effort and downtime.
Firestore full-text and geospatial search (Preview): Google’s industry-leading search technology is now natively integrated into Firestore, creating a differentiated, unified serverless database and search offering. This offering allows developers to execute high-relevance keyword, phrase, and geospatial queries, while ensuring that search results remain consistent with the underlying database data to significantly reduce operational overhead.

Mercado Libre: “Mercado Libre built Fury, our in-house developer gateway. Through Fury, we offer a NewSQL service powered by Spanner, leveraging its unmatched scalability, resilience, and consistency. While we are a cloud-native company, we have long remained vigilant regarding risks like insider threats, ransomware, and cloud outages. Spanner Omni enables true cross-cloud resilience, providing a robust and differentiated strategy that would be significantly more complex to implement with other cloud providers."
– Diego Oscar Narducci, Sr Technical Manager, Mercado Libre

Get started today

To learn more about our full portfolio please visit https://cloud.google.com/products/databases. To get started with migration and modernization initiatives, get started with our RAMP offer.

What’s new in the Agentic Data Cloud: Powering the System of Action

Wed, 22 Apr 2026 12:00:00 +0000

Companies are shifting from gen AI that simply answers questions to autonomous agents that perceive, reason, and act on their behalf. Attempting to scale these agents on legacy stacks exposes structural failures that can lead to fractured governance, a persistent trust gap, and broken reasoning loops, all while causing costs to spiral.

To solve this, we’re introducing the Agentic Data Cloud: an AI-native architecture that evolves the enterprise data platform from a static repository into a dynamic reasoning engine. It closes the gap between thinking and doing, allowing AI agents to act on your business data and context. While last-generation systems of intelligence were built only for human scale, the Agentic Data Cloud is a System of Action, built for agent scale.

Leading organizations are already using the Agentic Data Cloud to deliver tangible value:

Vodafone has launched hundreds of agents to deliver uninterrupted service to their customers, which is expected to save them millions of euros every year.
American Express is moving a core on-premises data warehouse and hundreds of production applications to BigQuery to power trusted agentic commerce at scale.
Virgin Voyages is using over 1,000 specialized AI agents, including one that slashes mass itinerary rebooking from six hours to just 11 minutes.

Today, we’re announcing three new innovation areas powering our Agentic Data Cloud:

A universal context engine that provides agents with trusted business context to drive higher accuracy.
Agentic-first practitioner experiences to evolve the role of data practitioners and developers as orchestrators of agents.
An AI-native, cross-cloud lakehouse that eliminates data silos by connecting your entire data estate.

Enabling agents with a universal context engine

AI is only as smart as its context. If an agent doesn't understand your definition of "margin" or the intricate relationships in your supply chain, it’s forced to guess. In the age of the agentic enterprise, data alone is not enough, and the old model of governance is insufficient. This is why we evolved the Dataplex Universal Catalog into the Knowledge Catalog, which maps and infers business meaning across your entire data estate, using a rigorous framework of aggregation, continuous enrichment, and search.

Here’s how it works:

Aggregation: To build true context, you must bring it together from everywhere. We are aggregating native context across your Google Cloud and partner data platforms. This includes third-party catalogs, applications, operating systems, and AI platforms like Palantir, Salesforce Data360, SAP, ServiceNow, and Workday (Preview). By using our Lakehouse, your third-party data assets are automatically mapped to the Knowledge Catalog. For Google Cloud sources, we are automating business logic with the new LookML Agent (Preview) and BigQuery measures (Preview), which embeds that business logic natively into the platform.
Continuous enrichment: The Knowledge Catalog delivers continuous enrichment by analyzing usage logs across your organization and profiling data behind the scenes. It learns how your enterprise actually uses data, not just what it is. This extends to unstructured data. The moment a file lands in Google Cloud Storage, our Smart Storage (Preview) instantly tags and enriches images and, soon, PDF objects. The Knowledge Catalog also identifies useful collections of unstructured data and uses Gemini to automatically generate missing schemas, mapping complex relationships so your AI is no longer flying blind.
Search and retrieval: Creating a massive context layer is great, but in the agentic era, search has evolved to be the new query path. The hardest problems at enterprise scale are speed, relevance, global reach, and security. To solve this, the Knowledge Catalog uses a sophisticated hybrid search stack built on Google Search innovation. To deliver relevance, it combines semantic and lexical matching with intelligent, machine-learning-based re-ranking. To deliver trust, we are enforcing your security permissions natively with access-control-aware search, so agents can only retrieve and act on the assets they are authorized to see. This high-precision infrastructure instantly identifies trusted context and feeds it to specialized agents.

The Knowledge Catalog now powers the Deep Research Agent (Preview). Part of the suite of Google-made agents available in Gemini Enterprise, this agent can perform multi-step reasoning across Google Cloud data platforms, such as BigQuery, as well as internal documents and web assets to answer complex questions with citations and precision that previously required weeks of manual effort.

Agentic-first practitioner experiences

As we move to this new architecture, the data practitioner role shifts from writing manual pipelines to orchestrating intent-driven engineering.

We’re accelerating this transition with the Google Cloud Data Agent Kit (Preview). Rather than introducing a new interface, we are launching a portable suite of skills, tools, environment-specific extensions, and built-in plugins, that drop into the environments developers love. By meeting practitioners where they already build — including VS Code, Gemini CLI, Codex, and Claude Code — we turn your IDE, notebook, or agentic terminal into a native data environment. This enables your environment to autonomously orchestrate a wide range of business outcomes, automatically selecting the right frameworks (e.g., dbt, Apache Spark, or Apache Airflow) and generating production-ready code based on Google’s gold standards.

This kit doesn't just connect tools — it injects high-performance capabilities directly into the developer's flow, scaling to petabytes without moving data. In fact, the Data Agent Kit features the same skills and tools that power our own out-of-the-box agents, including:

Data Engineering Agent (GA): Builds complex pipeline transformations from scratch and enforces governance rules to keep bad data out of production.
Data Science Agent (GA): Automates the model lifecycle — from wrangling to training — scaling across BigQuery Dataframes and Serverless Apache Spark.
Database Observability Agent (Preview): Acts as a 24/7 guardian for your infrastructure, diagnosing root causes and executing database remediations.

To help ensure the smooth execution of agents, Google Cloud has fully embraced Model Context Protocol (MCP), which provides a secure, universal interface that allows any agent to discover and use your data assets across our core engines, including: BigQuery, Spanner (Preview), AlloyDB, Cloud SQL (GA), and Looker MCP (Preview). MCP for Google Cloud uses our security stack, governing agent interactions based on your existing IAM policies, VPC Service Controls, and data residency requirements.

We’re also reimagining the business user experience with Conversational Analytics, now supported across BigQuery (GA), Cloud SQL, Spanner, AlloyDB (Preview), and Looker (GA). Organizations can simply publish these custom analytical agents in Gemini Enterprise, enabling employees to chat with live data in a familiar interface. By removing the technical barriers, we’ve eliminated the weeks spent waiting for manual reports, allowing businesses to move at the speed of thought.

A cross-cloud foundation built for agentic scale

For an agent to act, it must have a fundamentally open foundation. If an agent is blocked by cross-cloud latency or trapped in a proprietary walled garden, its autonomy is broken. That’s why we are introducing a truly borderless, cross-cloud Lakehouse that liberates your data wherever it resides by:

Connecting analytical estates: We’re integrating Cross-Cloud Interconnect (CCI) directly into our data plane. By combining CCI’s dedicated, high-speed private networking with Apache Iceberg REST Catalog, we’re enabling connectivity across clouds that is low latency and eliminates massive egress fees. As a result, agents can use data across AWS and Azure as if it were local to Google Cloud with seamless cross-cloud access.
Ending proprietary silos: We’re championing open federation to end the era of proprietary catalogs by launching bi-directional federation (Preview). Powered by the Iceberg REST Catalog, engines can now read directly from Databricks Unity Catalog on Amazon S3 (Preview), Snowflake Polaris (Preview), and the AWS Glue Data Catalog on Amazon S3 (Preview). This is reinforced by enhanced Lakehouse Governance (Preview), which ensures your security policies and access controls apply instantly across this borderless environment.
Unlocking operational data: We’re announcing Spanner Omni (Preview), unchaining the most scalable, globally consistent database on the planet. For the first time, you can run the Spanner engine anywhere — across clouds, on-premises, or on your laptop — with the same capabilities used to run Google.
Bridging the insight to action gap: We’re also closing the gap between insight and action. Most "unified" data platforms force the creation of complex ETL pipelines that block agents from accessing your real-time data. With Lakehouse federation for AlloyDB (Preview), we’re removing these pipelines by providing protocol-level, zero-ETL synchronization to give agents access to deep analytical history with low latency in operational transactions.

Automating the future

Moving to agent scale generates orders of magnitude more workloads. To support this, we are announcing four major performance breakthroughs:

Lightning Engine for Apache Spark delivers up to 2x the price-performance over the proprietary market alternative.
Managed Lustre delivers up to 10 terabytes-per-second of throughput to make sure data moves quickly enough for demanding models.
Bigtable now supports an in-memory tier that delivers sub-millisecond read latency for real-time applications. This means you can finally eliminate separate, side-by-side caching layers.
BigQuery fluid scaling helps lower costs by up to 34% on average for autoscaling workloads, scaling up resources instantly when agents act, and scaling back when they don't.

Build your success on a System of Action

The era of passive observation is over. The future belongs to the System of Action, made possible by Google’s Agentic Data Cloud.

Ready to see what an Agentic Data Cloud can do for your business?
Sign up for a strategy workshop today on how to get your data ready to fuel autonomous agents through a System of Action.

Want to learn more about an Agentic Data Cloud?
Read our blueprint for turning passive data into proactive action.

Introducing BigQuery Graph: Unlock hidden relationships in your data

Tue, 14 Apr 2026 16:00:00 +0000

Today, we're thrilled to announce that BigQuery Graph is now available in preview. With BigQuery Graph, we’ve built an easy-to-use, highly scalable graph analytics solution for data engineers, data analysts, data scientists, and AI developers, empowering them to model, analyze and visualize massive-scale relationships in an entirely new way.

As data changes and grows, it’s important to understand how different entities such as people, places, and products relate to one another. After all, data is more meaningful when we know how entities are interconnected. With traditional SQL, to find a "friend of a friend of a friend" requires multiple nested JOIN operations, which are usually challenging to read and write, and the performance degrades exponentially at scale. Finding the "blast radius" of a supply chain disruption during a storm requires multi-hop traversals, a full-scale graph analysis. To better solve this challenge, data is often modeled as a graph representation of the physical world around us, which can be better at finding complex and hidden relationships than traditional relational data structures.

Graph challenges faced by enterprises across industries

Graph technology has been broadly used across industries for fraud detection, recommendation engines, supply chain management, knowledge graph applications, and many others. However, users face some major challenges in adopting graphs:

Data silos and maintenance overhead: Having to store and maintain graph data in a standalone graph database creates data silos, data inconsistency, additional cost — not to mention extra operational overhead.
Lack of graph expertise: Adopting graph technologies often requires learning a new language, paradigm, and potentially a new database. At the same time, organizations’ investment in SQL expertise are less relevant.
Performance and scalability concerns: Many standalone graph databases work well when traversing the graph from a handful of nodes, but struggle to scale to billions of entities as business demands grow.

BigQuery Graph addresses many of these challenges by supporting:

Built-in graph query experience: A more intuitive graph query language (GQL) allows you to find patterns and traverse relationships among disparate data sets, based on the newest ISO GQL standard.
Unified relational and graph data models: Tight integration between graph and relational data models allows you to choose the best tool to model the data on a single source of truth without data duplication or data movement. Full interoperability between graph queries and SQL allows you to continue to leverage existing SQL skills, while taking advantage of the expressiveness of graph queries.
Graph over structured and unstructured data: Rich AI functions, vector and full-text search capabilities are supported with BigQuery Graph, allowing you to use semantic meaning, keyword search on graphs, bridging the gaps of structured and unstructured data.
Graph visualization: You can easily explore, investigate, and explain how your data is connected in an intuitive graph format using BigQuery Studio notebook and Jupyter Notebook.
Industry-leading ease of use, performance and scalability: BigQuery Graph is built upon BigQuery's serverless, scalable, cost-effective and distributed analytics engine that can scale to billions of nodes and edges.
Integration with Spanner Graph: This provides a unified graph schema and graph query language that serve a full spectrum of real-time (Spanner Graph) and batch graph needs (BigQuery Graph). You can also build a virtual graph by combining the latest data from Spanner and historical data from BigQuery without data movement using federated queries.
Chat with your graph: Very soon, you will be able to chat directly with graphs with the Conversational Analytics Agent (stay tuned).

Common BigQuery Graph use cases

BigQuery Graph opens up a realm of possibilities across industries for building intelligent applications:

Financial fraud detection: Analyze complex relationships among users, accounts, and transactions to identify suspicious patterns and anomalies, such as money laundering and irregular connections between entities, which can be difficult to detect using relational databases.
Customer 360: Track customer relationships, preferences, and purchase histories. Gain a holistic understanding of each customer, enable personalized recommendations, targeted marketing campaigns, and improved customer service experiences.
Social networks: Capture user activities and interactions and use graph pattern matching for friend recommendations and content discovery.
Manufacturing and supply chain management: Use graph patterns for efficient stockout analysis, cost rollups, and compliance checks by modeling parts, suppliers, orders, availability, and defects in the graph.
Healthcare: Capture patient relationships, conditions, diagnosis, and treatments to facilitate patient similarity analysis and treatment planning.
Transportation optimization: Model places, connections, distances, and costs in the graph, and then use graph queries to find the optimal route.

BigQuery Graph in the real-world

Many customers across industries have leveraged BigQuery Graph to solve real-world business challenges. Here are a few examples of how they are putting these capabilities into practice:

BioCorteX: drug discovery

"Understanding disease isn't about collecting more data; it’s about understanding the relationships within that data. By leveraging pathway search in BigQuery Graph at a massive scale, reaching depths of 7+ hops, we are finally able to see more of the human metabolism map. This level of scale is what allows us to move beyond trial and error, identifying the precise biological levers that need to be pulled to cure complex diseases. We aren't just guessing anymore, we’re emulating life at the speed of compute." - Nik Sharma, CEO and Cofounder, BioCorteX

Curve: fraud detection

"By implementing BigQuery Graph, we have successfully moved away from the previous limited sql-based approach to a more scalable solution for fraud detection network analysis. This has allowed us to detect sophisticated fraud networks by uncovering hidden connections between seemingly unrelated accounts and transactions. This transition from traditional relational queries to graph-based analytics has showcased measurable business impact with ~£9.1M of savings. This shift has not only improved the precision of fraud detection but has also provided a scalable foundation for protecting the ecosystem without adding significant operational overhead." - Francis Darby, VP Data & ML, Curve

Virgin Media 02: fraud detection

"At Virgin Media O2, we are constantly evolving our defenses to stay ahead of increasingly sophisticated fraud networks. We’ve added a powerful new layer to our already robust fraud alerting system. Using BigQuery Graph, we can now execute complex 4-hop queries that map the hidden relationships between accounts, devices, and activities. This deeper visibility allowed us to identify networks of suspicious addresses. This doesn't just catch fraud; it acts as an early warning system, flagging new connections to known risk networks before they can cause damage." -- Jonathon Ford, Director Data Applications, Virgin Media O2

How to use BigQuery Graph

BigQuery Graph is more than just a new feature; it's a new way of thinking about data, empowering you to ask bigger questions, uncover deeper insights, and solve your most challenging problems.

You can get started in three simple steps:

Step 1: Create graph schemas on top of the relationship tables using DDL with a single copy of data.

Create a finance graph by mapping relational tables into “Account”, “Person”, "Loan" nodes and their relationships “Transfers”, “Owns”, "Repays" via edges.

code_block: <ListValue: [StructValue([('code', 'CREATE PROPERTY GRAPH graph_db.FinGraph\r\nNODE TABLES (\r\n graph_db.Account KEY(id),\r\n graph_db.Person KEY(id),\r\n graph_db.Loan KEY(id)\r\n) \r\nEDGE TABLES (\r\n graph_db.Transfers \r\n KEY (id, to_id, timestamp) \r\n SOURCE KEY (id) REFERENCES Account (id)\r\n DESTINATION KEY (to_id) REFERENCES Account (id), \r\n graph_db.Owns\r\n KEY (id, account_id, timestamp) \r\n SOURCE KEY (id) REFERENCES Person (id)\r\n DESTINATION KEY (account_id) REFERENCES Account(id),\r\n graph_db.Repays\r\n KEY (id, loan_id, timestamp) \r\n SOURCE KEY (id) REFERENCES Person (id)\r\n DESTINATION KEY (loan_id) REFERENCES Loan(id)'), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7f8982ea0d30>)])]>

Step 2: Use intuitive SQL/GQL to traverse data relationships and find hidden connectivities.

Find the accounts owned by Jacob and the loans he repays from those accounts:

code_block: <ListValue: [StructValue([('code', 'GRAPH graph_db.FinGraph\r\nMATCH\r\n (person:Person {name: "Jacob"}) \r\n -[own:Owns]->(account:Account)\r\n -[repay:Repays]->(loan:Loan)\r\nRETURN\r\n account.id AS account_id,\r\n loan.id AS loan_id'), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7f8982ea0e80>)])]>

Combine vector search with graph traversals to find fraudster-like accounts and their transfer activities within 1-6 hops:

code_block: <ListValue: [StructValue([('code', "DECLARE similar_account_to_fraudster DEFAULT ((\r\n SELECT array_agg(base.id)\r\n FROM VECTOR_SEARCH(TABLE graph_db.Account, 'embedding',\r\n (SELECT * FROM graph_db.Account WHERE id=102), 'embedding', \r\n top_k => 6)\r\n));\r\nGRAPH graph_db.FinGraph\r\nMATCH\r\n (person:Person)-[own:Owns]->\r\n (account:Account)-[transfer:Transfers]->{1,6}\r\n (to_account:Account)\r\nWHERE to_account.id IN \r\n UNNEST(similar_account_to_fraudster)\r\nRETURN\r\n person.id AS person_id,\r\n account.id AS src_account,\r\n to_account.id AS to_account"), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7f8985705730>)])]>

Step 3: Visualize graph results to detect connectivity of disparate data in a more intuitive way in BigQuery Studio notebook.

If you are looking for a specialized graph visualization tool, BigQuery Graph has integrated with industry leading partners including G.V(), Graphistry, Kineviz, Linkurious. They allow you to see a visualization of BigQuery Graph query results outside the Google Cloud console.

Ready to get started?

The future of data analysis is connected. With BigQuery Graph, you have the power to unlock that connectivity and transform your business into actionable insights grounded with your enterprise knowledge. Start exploring today and unleash the power of your data's interconnected relationships!

Visit the BigQuery documentation: find overview and quickstart guide.
Explore tutorials: get hands-on experience with BigQuery Graph through tutorials.
Share your feedback: join our community, and get your questions answered via bq-graph-preview-support@google.com.
Related blogs:

From operational to analytical: The unified Spanner Graph and BigQuery Graph solution

Tue, 14 Apr 2026 16:00:00 +0000

Understanding the relationships within your data is crucial for uncovering hidden insights and building intelligent applications. However, managing operational (OLTP) and analytical (OLAP) graph workloads usually means wrestling with disconnected databases, building brittle data pipelines, and managing complex integrations. This fragmentation creates data silos, increases operational overhead, and limits scalability.

Today, we're thrilled to introduce a unified graph database and analytics solution powered by Spanner Graph and BigQuery Graph. The solution consists of the two platforms, recommended blueprints for how to deploy them, and getting started guides for the most prominent use cases. In this blog, we review the solution’s components, provide an overview of the most common use cases, and hear from customers who have deployed the solution in the real world.

Spanner Graph for operational workloads

Spanner Graph reimagines graph data management, bringing together graph, relational, search, and generative AI capabilities into a single database. It is backed by Spanner’s signature unlimited scalability, high availability, and strong consistency.

With Spanner Graph, you get:

Integrated table-to-graph mapping: Define graphs directly over your existing Spanner relational tables, allowing you to view and query operational data as a graph without data duplication.
Interoperable graph and relational querying: Leverage an ISO-standard Graph Query Language (GQL) interface for intuitive pattern matching, and mix GQL with SQL in a single query to traverse both graph and tabular data together.
Advanced search and AI integration: Utilize built-in vector search, full-text search, and Vertex AI integration to retrieve data by semantic meaning and power intelligent applications directly within your database.

Customers are already using Spanner Graph to power high-throughput, low-latency applications - for identity resolution across millions of entities, identifying dependencies across vast complex environments, data lineage, customer 360 use-cases, and enhancing real-time fraud detection.

"Open Intelligence is our foundational intelligence layer that securely connects trillions of live data points from clients, partners and WPP in a privacy-first way and is now integrated and powers WPP’s agentic marketing platform, WPP Open. Enabled by Google Cloud's Spanner Graph, Open Intelligence is a significant advancement in AI-driven marketing and we are excited about extending the use case for analytical graph workloads on BigQuery Graph." - Rob Marshall, Head of Strategy, Data & Intelligence, WPP

BigQuery Graph for analytical workloads

While Spanner Graph handles your active operations, true large-scale analysis requires exploring relationships across billions of nodes and edges to identify patterns and query historical data. Just as SQL relies on distinct tools for databases and data warehouses, the graph landscape requires specialized tools for different workloads. That's why we built BigQuery Graph.

BigQuery Graph brings connected data analytics directly into your data warehouse. You can map existing BigQuery data to a graph schema and query it with SQL or GQL to uncover hidden relationships in massive datasets - without moving any data.

Key capabilities include:

Integrated table-to-graph mapping: Map your existing BigQuery tables to graphs instantly, uncovering hidden relationships in your data warehouse without building ETL pipelines or moving a single byte of data.
Interoperable graph and relational querying: Apply the same expressive pattern matching of GQL to massive historical datasets, and mix SQL with GQL in a single query to combine the familiarity of your data warehouse with powerful graph traversal.
Advanced search and AI integration: Leverage native integration with BigQuery AI for predictive analytics, alongside built-in vector search, full-text search, and geospatial functions to locate connected information across billions of records.

Spanner Graph and BigQuery Graph as a unified solution

While each platform is powerful on its own, their true value emerges when they are deployed together. By connecting your operational and analytical environments, you eliminate data silos and accelerate your time-to-insight without compromising database performance.

"Spanner Graph enables Yahoo to unify our data into a connected foundation at a global scale, powering real-time, intelligent decision-making across our agentic advertising platform. This enhances our AI-driven approaches that drive one of the largest digital advertising ecosystems, and we look forward to building on it with BigQuery Graph to unlock deeper analytics and predictive capabilities to power future innovation." - Gabriel DeWitt, Head of Consumer Monetization, Yahoo

Take financial fraud detection as an example: your application can use Spanner Graph to instantly identify a suspicious connection and block a transaction at checkout. Meanwhile, BigQuery Graph can analyze petabytes of historical transaction data to expose the complex, long-term fraud ring that initiated it.

Here is how these two engines integrate to create an end-to-end graph workflow:

1) A unified graph query and schema experience

A core advantage of this solution is the consistent schema and GQL shared across both platforms. This shared language reduces development time and minimizes the friction of context-switching.

For example, to find potential fraud rings originating from a specific account in real-time, you would use this Spanner Graph query:

code_block: <ListValue: [StructValue([('code', 'GRAPH FinGraph\r\nMATCH p=(:Account {id: @accountId})-[:Transfers]->{2,5}(:Account)\r\nRETURN PATH_LENGTH(p) AS path_length, TO_JSON(p) AS full_path;'), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7f8982a54d90>)])]>

To run that same analysis to find all accounts involved in historical fraud rings, the BigQuery Graph query is nearly identical:

code_block: <ListValue: [StructValue([('code', 'GRAPH bigquery.FinGraph\r\nMATCH p=(:Account)-[:Transfers]->{2,5}(:Account)\r\nRETURN PATH_LENGTH(p) AS path_length, TO_JSON(p) AS full_path;'), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7f8982a54e20>)])]>

2) Query Spanner data in BigQuery Graph through Data Boost

With Data Boost, you can query Spanner Graph data directly from BigQuery without impacting performance of your transactional workloads. This allows you to build a "virtual graph" combining real-time operational data from Spanner with historical analytics in BigQuery - without moving any data.

For instance, you can combine real-time Account and User nodes from Spanner Graph with historical LogIn edges from BigQuery to identify suspicious login patterns across different devices.

To do this, you first connect BigQuery to Spanner using the CREATE EXTERNAL SCHEMA statement:

code_block: <ListValue: [StructValue([('code', "CREATE EXTERNAL SCHEMA spanner\r\nOPTIONS (\r\n external_source = 'google-cloudspanner:/projects/PROJECT_ID/instances/INSTANCE/databases/DATABASE',\r\n location = 'LOCATION'\r\n);"), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7f8982a54df0>)])]>

Next, define a BigQuery Graph, incorporating tables from both Spanner and BigQuery:

code_block: <ListValue: [StructValue([('code', 'CREATE OR REPLACE PROPERTY GRAPH bigquery.FinGraph\r\n NODE TABLES (\r\n -- Account and Person are stored in Spanner,\r\n -- made available in BigQuery through the `CREATE EXTERNAL SCHEMA` statement.\r\n spanner.Account KEY (account_id),\r\n spanner.Person KEY (person_id),\r\n -- Media is stored in BigQuery.\r\n bigquery.Media KEY (media_id)\r\n )\r\n EDGE TABLES (\r\n -- Transfers and Owns are stored in Spanner.\r\n spanner.Transfers AS Transfers\r\n KEY (transfer_id)\r\n SOURCE KEY(account_id) REFERENCES Account\r\n DESTINATION KEY(target_account_id) REFERENCES Account,\r\n spanner.Owns AS Owns\r\n KEY (person_id, account_id)\r\n SOURCE KEY(person_id) REFERENCES Person\r\n DESTINATION KEY(account_id) REFERENCES Account,\r\n -- LogIn is stored in BigQuery.\r\n bigquery.LogIn AS LogIn\r\n KEY (login_id)\r\n SOURCE KEY(media_id) REFERENCES Media\r\n DESTINATION KEY(account_id) REFERENCES Account,\r\n );'), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7f8982a54c40>)])]>

Finally, execute a query on BigQuery Graph to access data across both Spanner (accounts, users, transfers, owns) and BigQuery (logins, devices), identifying potentially suspicious login patterns:

code_block: <ListValue: [StructValue([('code', 'GRAPH bigquery.FinGraph\r\nMATCH p=(owner:Person)-[:Owns]->\r\n (:Account)<-[login:LogIn]-\r\n (media:Media {blocked: true})\r\nRETURN TO_JSON(p) AS full_path\r\nORDER BY login.time\r\nLIMIT 20;'), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7f897073d700>)])]>

3) Export BigQuery data into Spanner Graph through reverse ETL

When you need to bring analytical data back into Spanner for low-latency, real-time querying, you can use reverse ETL without additional pipelines. For example, you can import historical device data (IP addresses, device IDs) from BigQuery into Spanner Graph to enhance your real-time fraud detection operations:

code_block: <ListValue: [StructValue([('code', 'EXPORT DATA\r\n OPTIONS (\r\n uri = \'https://spanner.googleapis.com/projects/PROJECT_ID/instances/INSTANCE/databases/DATABASE\',\r\n format=\'CLOUD_SPANNER\',\r\n spanner_options="""{ "table": "Media" }"""\r\n ) AS \r\nSELECT * FROM Media;'), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7f8985710280>)])]>

4) Visualize your graph data

Visualizing connected data is core to analysis, explorations and investigations. With Spanner Studio and BigQuery Studio (coming soon), you can instantly visualize your graph data without leaving your familiar environment or setting up external tools. For deeper programmatic exploration, you can also leverage Spanner Graph notebook and BigQuery notebooks to render query results directly within your existing data science workflows.

5) Graph visualization partner integrations

Spanner Graph and BigQuery Graph also integrate with leading graph visualization partners to provide a comprehensive suite of exploration tools:

Kineviz: Combines cutting-edge visualization with advanced analytics via GraphXR.
Graphistry: Extracts meaningful insights from large datasets using a GPU-accelerated visual graph intelligence platform.
G.V(): Offers a quick-to-install client for high-performance visualization and no-code data exploration.
Linkurious: Detects and analyzes threats in large volumes of connected data via the Linkurious Enterprise platform.

One unified solution for all your graph needs

Together, Spanner Graph and BigQuery Graph provide a unified solution for operational and analytical needs across various use cases:

Domains	Spanner Graph	BigQuery Graph
Financial Services	Instantly blocks anomalous, suspicious transactions.	Uncovers complex, long-term fraud rings.
Retail & E-commerce	Serves personalized product recommendations on the fly.	Analyzes vast purchasing histories to predict demand.
Cybersecurity	Isolates active threats and traces attack origins instantly.	Models historical vulnerabilities to strengthen defenses.
Healthcare	Powers clinical decision support systems at the point of care.	Analyzes population health trends and disease risk factors.
Supply Chain	Tracks goods globally and alerts teams to immediate disruptions.	Identifies systemic bottlenecks to optimize future routing.
Telecommunications	Creates a network digital twin for detecting anomalies, and root cause analysis in real-time.	Analyzes traffic patterns at scale to plan future infrastructure upgrades.

Get started with Spanner Graph and BigQuery Graph today

With Spanner Graph and BigQuery Graph, we’re excited to offer a unified graph data management experience across your operational and analytical needs.

Explore Spanner Graph's use cases and setup guide for your operational workloads, and the BigQuery Graph overview and creation guide for your analytical needs. To experience the full power of this combination, check out our unified solution guide and try the codelab.

QueryData helps agents turn natural language into queries for AlloyDB, Cloud SQL and Spanner

Fri, 10 Apr 2026 16:00:00 +0000

QueryData launches in preview today. It is a tool for translating natural language into database queries with near-100% accuracy. With QueryData, you can build agentic experiences across AlloyDB, Cloud SQL (for MySQL and PostgreSQL), and Spanner (for GoogleSQL). It builds upon Google Cloud’s #1 spot in the BiRD benchmark, one of the world's most competitive benchmarks for natural-language-to-SQL – as well as upon Gemini-assisted context engineering.

Developers are already seeing the benefits from QueryData, including Hughes Network Systems, a leader in telecommunications, that deployed QueryData in production. “We have transformed user support operations with Google Cloud’s data agents. At the heart of our solution is QueryData, enabling near-100% accuracy in production. We are excited about the future of agentic systems!" - Amarender Singh Sardar, Director of AI, Hughes Network Systems

The opportunity for agentic systems: from intent to action

Agentic systems are evolving from human-advisory roles into active decision-makers. To execute business actions accurately, agents require precise information from operational databases (such as pricing, inventory, or transaction records).

With requests expressed in natural language, bridging the gap between conversational input and database records is essential. High-quality natural language-to-query capability is a critical requirement for enabling agents to take actions.

The developer’s dilemma: why natural language for agents with databases is hard

Hurdles for agents querying enterprise data are threefold: accuracy, security and ease of use. QueryData addresses all three of them:

Accuracy – Inaccurate answers carry a risk of poor business decisions, disappointed end-users or financial losses. In many industries, translating text into SQL with 90% accuracy is simply insufficient for taking action.
Security – how to make sure that each person (or agent) only queries the data they are allowed to see? Enterprises need auditable, deterministic access controls. Relying on the LLM's judgement (aka “probabilistic” access controls) falls short of that. Even a low risk of security breaches means disproportionately high losses
Ease of use – Achieving high accuracy requires developers to provide extensive contextual information about their data. This can be a laborious task. Another example of developer friction is integration and maintenance of agentic tools

Understanding the accuracy gap

LLMs are really good at writing query code. However, to write accurate queries for a given database – it takes more than coding skills, and more than just parsing the schema:

Schemas can be unclear – developers often use shorthands or abbreviated names. For example: what does a column named “product” mean? A product category? A particular model…? It gets even worse with column names like “prod” or simply “p”
Values can be ambiguous – let’s take a column named “order return status”... where values are expressed as integers: “1”, “2” and “3”. Which of these represents “returned” or “return initiated”?
Schemas cover data structure, but not the business logic – Your business may define “monthly active users” as those who have posted at least once, not just logged in (but database may lack this nuance).
Underspecified queries – Natural language questions can be ambiguous, like “latest sales”.

How QueryData solves for near-100% accuracy

QueryData leverages the Gemini LLM, as well as context which describes your unique database.

Database context, which is essentially the code fueling QueryData, is a set of descriptions and instructions including:

Schema ontology – information about the meaning of the data. Descriptions of columns, tables and values. It helps QueryData overcome ambiguity by figuring out what data is needed to answer the question
Query blueprints – guidelines and explicit instructions for how to write database queries to answer specific types of questions. Templates and facets specify the exact SQL to write for a given type of question.

As a last resort, QueryData will detect when a clarifying question needs to be asked.

Deterministic security for your queries

Agentic applications require deterministic, auditable security. Developers can use Parameterized Secure Views (PSVs) to define agent access via fixed parameters, like user ID or region. By passing these security-critical parameters separately from queries, the application ensures agents can only access the authorized data. This prevents agents from querying restricted information, even if they attempt to do so.

Support for PSVs is available today in AlloyDB, and coming soon to Cloud SQL and Spanner.

Ease of use for quality hill-climbing and tool integration

Integration of QueryData into your agentic workflows is easy. The QueryData API can be used directly or exposed as a Model Context Protocol (MCP) tool via our popular open source MCP Server: MCP Toolbox for Databases. QueryData automatically works across different database dialects – no need for database-specific code, just one API to query them all.

Another area where QueryData makes things easier for developers – is context engineering. It is the process of iteratively evaluating and optimizing context. It is critical to QueryData’s ability to accurately query your database. Developers using QueryData enjoy support from a robust suite of tools:

Out-of-the-box context generation – upon configuring QueryData, the Context Engineering Assistant, a dedicated agent in Gemini CLI, will help you create the very first context set for your database.
Evals: Developers can use the bundled Evalbench framework to measure accuracy against a set of tests specific to your use case
Context optimization: the Context Engineering Assistant reviews eval results, recommends changes and then helps run evals again. Through this iterative process, you can reach near-100% accuracy.

What you can build with QueryData today

Developers are already building with QueryData. Examples include:

Customer-facing applications: a real estate search engine, where QueryData translates user prompts into database queries, and then schedules viewing appointments
Internal tools: an AI-powered staffing app querying human resources data and then enabling managers to assign workers to shifts
Multi-agent architectures: a trade compliance workflow where a top level agent asks a sub-agent to verify that an entity has appropriate KYC (“Know Your Customer”) status. The KYC agent queries a database to confirm the customer’s identity.

Next steps

You can have your agent start using QueryData as a tool for near-100% accurate database calls today. For more details, explore our technical documentation:

Check out the "Swiss property search" high-fidelity demo, pictured below (video walkthrough here). Note: This is an independent project (not maintained by Google Cloud) and is for illustrative purposes only: GitHub link

Near-100% Accurate Data for your Agent with Comprehensive Context Engineering

Fri, 10 Apr 2026 16:00:00 +0000

Agentic workflows are already used for initiating action. To be successful, agents typically need to combine multiple steps and execute business logic reflective of real-life decisions. But, as developers rush to deploy these autonomous agents, they are slamming into a wall: the compounding error problem of accuracy.

To understand why agentic workflows require near-100% accuracy on questions that are answerable by your database data, let’s look at the numbers: Assume an accuracy of 90% in a single-step AI process. You ask a question; you get a correct answer 90% of the time. But in an agentic workflow, the AI takes multiple dependent steps – and errors compound exponentially.

Let’s run the numbers on a 90% accurate agent:

One step: 90% success rate.
Two steps: 0.90 × 0.90 = 81% success rate.
Five steps: 0.90^5 = 59% success rate.

Now, imagine that same five-step workflow running on an 80% accurate agent. The success rate plummets to just 33%.

In a business context, even 90% accuracy is often insufficient. And 59% or 33% success rate is downright catastrophic. Indeed, in many industries near-100% accuracy is needed, because the agentic application is customer-facing and inaccuracies lead to loss of trust and loss of revenue. Furthermore, in many industries there are legal, safety and compliance requirements. In such industries, near-100% accuracy must be combined with explainability so that the human-in-the-loop can understand and verify the answers.

Example: consider a real estate agency using an AI workflow to handle new tenant onboarding in a five-step flow. The agentic flow must:

extract data from an application
run a background check via an API
query the database for available units
draft a lease, and
email the tenant.

If step three fails because the AI makes a mistake in the database query and pulls a unit for the wrong city – then, steps four and five will generate a legally binding lease for a property that doesn't exist, and then send it to the client. The cost of manual remediation, lost trust, and legal liability makes anything less than near-perfect execution completely unviable.

Agentic Tools: A Path to Accuracy and Explainability

To achieve the required accuracy and explainability when agents interact with enterprise databases, developers are turning to specialized tools. QueryData is such a tool for agents, designed specifically to offer near-100% accuracy for natural language-to-query. By enabling agents to retrieve correct data, QueryData ensures that agents are well-equipped to take action.

The Key Ingredient: Comprehensive Database Context

A Large Language Model (LLM) inherently knows many dialects of SQL, but it doesn't know your business logic and your database. Agentic tools use context to bridge that gap. Context is essentially the code which a tool like QueryData uses to guide the LLM towards correct answers. Crucially for achieving near-100% accuracy and explainability, the QueryData works with a comprehensive database context, organized into three main pillars: Schema Ontology, Query Blueprints and Value Searches.

1. Schema Ontology

Schema ontology is about understanding your database structure and semantics. This includes natural language descriptions of tables and columns. The QueryData LLM has a greater chance to translate the natural language question into the correct query using these instructions. You can think of schema ontology as a set of “cues” or “hints” – meant to steer the LLM into picking the right tables and columns and synthesizing them correctly into a database query. A couple of examples:

Here is what a database-level description could look like for a search engine of real estate listings:

“Listings, real estate agents and information about communities where listings are located – schools, amenities and hazards: fire, flood and noise”

The table description for property could look like this:

“Current real estate listing, including houses, townhomes, condos and land”

An example of column description that explains that the proximity_miles means

“property distance from the district’s school in miles”

For ease of use, you can autogenerate rich descriptions, which will typically include sample values of the column.

2. Query Blueprints

If ontology is the vocabulary, query blueprints are the way to introduce fine control of the generated SQL for important questions that must absolutely receive accurate and business-relevant answers. For example, consider the question “Riverside houses close to good schools”. The interpretation of “close” and “good” provided by Gemini is impressive- in a demo application it translated to

…WHERE city_name = 'Riverside' AND school_ranking <= 5ORDER BY proximity_miles ASC

But this interpretation still leaves much to be desired: Wouldn’t you drive one more mile for a school whose school_ranking is much higher than the Gemini-chosen cutoff? Of course you would! Both proximity and school ranking should affect the overall ranking. A no-cut-corners developer will take control of the interpretation of “close to good school” by introducing a sophisticated ranking function, which may be the result of continuous A/B experiments, along with sensible cutoffs.

Templates
In particular, she will use a template: A pair of natural language intent with its respective parameterized SQL translation.

parameterized_intent : “$1 houses close to good schools”,parameterized_SQL : “SELECT … FROM … WHERE city_name = $1 AND "school_ranking" <= 5 AND "proximity_miles" <= 2ORDER BY school_score("school_ranking", "proximity_miles")”– the school_score stored procedure combines school ranking and proximity into a single ranking

Such info can be given in a JSON file but, even more user-friendly, you can use Gemini CLI, prompt it with an example natural language question and your ideal respective SQL and it will produce the JSON for you.

Furthermore, templates enable the agent to explain how the question was interpreted. This mitigates the effect of the occasional remaining inaccuracies, allowing a human-in-the-loop or agent to understand what the answer of QueryData means.

Facets
While plain query templates provide highly accurate and explainable answers, they have low flexibility: they can only answer the specific critical question patterns that they were designed for. What if you wanted to combine the “close to good schools” with price conditions, square footage, bedroom conditions and more. The facets generalize templates to combine the best of both worlds: highly-accurate, explainable answers to large numbers of questions.

"parameterized_intent": "Property price between $1 and $2", "parameterized_sql_snippet": "T.\"price\" BETWEEN $1 AND $2"

Value searches
Some ambiguities in the NL question are rooted deep in the private data of your database and need a collaboration of the LLM with the database to disambiguate. Value searches solve the hard problem of correctly associating data values in the database with the “entities” that the question talks about.

For example, consider the question “Westwod''s sold properties in the last 1 month.” The first problem is that there is no “Westwod”; it is a misspelling of “Westwood”. Apart from the misspelling, there is a second problem - a deeper ambiguity in our sample database: “Westwood” appears as both the name of a real estate brokerage and as the name of a city. Value searches can utilize the built-in powerful vector+text search capabilities of Google Cloud’s AI-native databases. Here, value searches will enable QueryData to respond to the agent that this is likely a misspelling of ‘“westwood, which appears as both a real estate brokerage and a city name.

Accuracy As Foundation for Agentic Actions

Agentic workflows are poised to revolutionize operations, but they are unforgiving when it comes to accuracy. Through context engineering, businesses can mitigate compounding failures and start trusting their autonomous agents to deliver.

As a next step, you can explore how to create context sets across these databases:

And here – your “cheat sheet” for building blocks of context (courtesy by Nanobanana):

Spanner's multi-model advantage for the era of agentic AI

Tue, 31 Mar 2026 17:00:00 +0000

In the age of agentic AI, the role of databases is fundamentally changing. It’s evolving from a passive data repository into an intelligent, active context hub designed to ground generative AI foundation models and a reasoning engine that drives proactive actions. To power sophisticated AI and agentic workflows, a database must do more than store and query your data — it must facilitate reasoning, provide deep contextualization, and turn static intelligence into proactive action – all in a multiple data model environment.

Rather than manually orchestrating data across disparate silos, a multi-model database approach coalesces relational, vector, and graph data into a rich, unified knowledge base. This allows AI to leverage situational, semantic, and relationship context simultaneously. Spanner, Google Cloud’s globally consistent, scalable, fully managed database, is built for the AI era. It offers a unified approach to integrate your data and power sophisticated, agentic AI.

In this post, we’ll look at how Spanner is helping organizations achieve production-ready AI, and if you want more real-world examples, you can read this companion article. It covers four key areas — fraud detection, personalized recommendations, hybrid search, and autonomous network operations.

The Power of interoperable multi-model databases

Today’s common, fragmented, multi-database strategy introduces critical challenges:

Data inconsistency: The absence of a single source of truth forces developers to simulate consistency through brittle application code, leading to data duplication, stale data, and significant governance and security vulnerabilities. Your application's overall reliability is effectively capped by the database with the lowest SLA.
Operational and skill silos: Managing disparate systems creates massive operational overhead and fractures your team into skill-silos, imposing a "complexity tax" that directly inflates costs and drags down development velocity.
Unified intelligence blockers: Inherent data silos and the ETL latency create an architectural barrier, preventing the real-time, context-aware intelligence required to power cutting-edge AI applications.

Together, these challenges create a significant structural disadvantage in the AI era.

Spanner’s AI advantage

We’ve been constantly evolving Spanner over the years, to keep up with the most demanding cloud workloads and Generative AI offers some of the most unique and novel database challenges yet.

Building on its core foundation, Spanner introduced fully interoperable multi-model capabilities last year, allowing you to access and query your data using diverse models — including relational, key-value, graph, and vector —all within a single, highly capable database foundation. Each data model is powerful in their own right:

Relational: Spanner has pioneered the relational scale-out database, offering ANSI SQL with both Google SQL and PostgreSQL dialects that delivers 5 9s availability and strong consistency across the globe.
Key-value: Spanner also offers high-performance, key-value database capabilities. Its Cassandra native endpoint allows customers to easily lift and shift their Cassandra workload to Spanner without change to application codes
Graph: Built on the ISO standard GQL, allowing you to model data natively as a graph or as a simple overlay on top of relational data.
Vector: A fully integrated semantic search solution offering both KNN and ANN, with the latter built using Google's state-of-the-art ScaNN technology, capable of supporting indexes with over 10 billion vectors.
Full-Text Search: building on Google’s decades of search expertise that allows you to search for across structured and unstructured data with advanced information retrieval tools out of the box for 40+ languages
Integration with Data Warehouses: Seamlessly connecting transactional data with insights from analytical processes.

This unified architecture eliminates the need for complex data synchronization across multiple specialized databases, radically simplifying your environment and accelerating development. Customers can now leverage Spanner's interoperable multi-model capability to combine distinct data types seamlessly.

Companies making AI a globe-spanning success

MakeMyTrip is one of many companies putting AI into production in demanding real-world environments thanks to the multi-model capabilities of Spanner.

An India-based online travel company, MakeMyTrip has successfully consolidated four specialized databases into a single Spanner instance, significantly reducing operational complexity, while accelerating AI innovation and achieving high performant and high quality outputs.

Ravindra Tiwary, director of technology development at MakeMyTrip, saw multiple benefits to her organization’s Spanner migration: “We consolidated MongoDB, Neo4j, Elasticsearch, and Qdrant into one unified system, achieving a 75% reduction in operational complexity. This transition eliminated the friction of managing multiple database nuances and redundant synchronization pipelines.

The most transformative advantage is executing unified queries across lexical, keyword, and embedding searches in a single platform. This consolidation accelerated our feature innovation cycle by 30% to 50% for consolidated-view features and saved 5.5 to 9.5 operational hours per week, per destination.

With Spanner as our single source of truth, we eliminated data drift and ensured facts are updated atomically. This directly improved our answer-quality score by 9%. Spanner now provides the scalable, high-performance foundation essential for our 'Destination Expert' ecosystem (a homegrown GenAI-powered travel planner bot providing personalized travel recommendations) and future AI workflows.”

And they’re not the only ones who’ve reached new destinations on their AI journey. To learn how leading organizations like Target, Palo Alto Networks, and MasOrange are succeeding with Spanner, check out our next post.

To get started on your own AI journey with Spanner, visit our Spanner page to learn more and get started today.

Real-world success with Spanner’s fully interoperable multi-model database

Tue, 31 Mar 2026 17:00:00 +0000

In the first post on the power of multi-model databases to lay the foundations for gen AI, we highlighted how Google Cloud Spanner helps organizations overcome some of the challenges presented by traditional approaches to database architecture and management. In this post, we dive deeper on the specific examples, across four common use cases.

We are seeing customers increasingly choose Spanner's multi-model capabilities to address three key strategic goals:

A foundation of scale and reliability: Many specialized databases for graph, vector, or search, are built on traditional, single-machine architectures. As a result, they face fundamental challenges with scalability, availability, and consistency. We see customers migrate off these specialized systems because they have - or are about to - hit a wall. All Spanner’s data models are built on its tried-and-true platform offering 99.999% availability, automatic scaling, and limitless horizontal scale, and they can easily extend to new capabilities. For example, adding a vector embedding column to the existing graph schema.
Consolidating database sprawl and eliminating ETL: Managing, securing, and patching multiple disparate databases, each with its own data model, query language and backup policy can be an operational nightmare for users. Extract, transform, load (ETL) pipelines required to sync data, are especially frustrating as they often create inconsistency and delays. Spanner eliminates this complexity by offering multiple data models in a single unified database, eliminating extra data copies, inconsistency and management overhead. Moreover, Spanner’s interoperable multi-model capabilities allow a developer to write one SQL query that joins relational tables, traverses a graph relationship, and filters on a vector or text search function.
Future-proofing for evolving application needs: While many customers start with a simple application, they know it will need to get smarter and more complex over time. In Spanner, adding a graph-based recommendation or AI-powered vector search can be an afterthought. A developer can simply turn on graph or search capabilities on their operational data, with a simple data definition language (DDL) command. With Spanner, there is no painful migration, no complex re-architecturing and no growth ceiling. Instead, customers can build on a reliable relational database while seamlessly adding new, advanced data models as their application evolves.

Here’s how customers across industries are already leveraging Spanner's evolving multi-model capabilities to solve their toughest data challenges and achieve early success:

1. Fraud detection

Fraudsters often exploit complex, non-obvious patterns across multiple transactions and accounts. Traditional relational databases struggle to detect these intricate relationships in real-time. Spanner combines relational queries with graph analytics to enable real-time pattern recognition. This allows businesses to efficiently identify suspicious clusters or unusual connections that might indicate fraudulent behavior, significantly reducing financial losses and enhancing security.

DANA: Anti-money laundering for fast growing customer base

DANA, an Indonesia-based e-wallet app, offering payments and digital financial services including lending, insurance and investments, has adopted Spanner Graph to support its critical anti-money-laundering, or AML, efforts.

Challenge: With a massive and still rapidly growing user base, DANA struggled to scale and meet query performance SLAs using existing relational databases to detect money laundering patterns in transactions. Moving to do the analytics in graph databases was obvious, but many graph database providers in the market simply could not handle the scale.
Solution: Spanner was selected after an elaborate RFP process due to its high availability, virtually unlimited scale, and external consistency. The ability to use full-text search (FTS) and vector search directly within the Graph model were key differentiators.

Palo Alto Networks: Access graph for SaaS identity

Palo Alto Networks, one of the leading cybersecurity firms, leverages Spanner to provide insights into organizational identity posture, surfacing misconfigurations and over-privileged accounts, dormant accounts, unrotated credentials, over-privileged accounts, and accounts missing in the Identity Provider (IDP).

Challenge: The team needed to build a world-class agent security product for the AI era that could innovate quickly while ensuring highly scalability without creating data silos.
Solution: They built an "Access Graph" on Spanner to connect user identities, access permissions, and the associated user activities within the SaaS applications. Spanner allows them to achieve massive scale with a single schema for both graph and non-graph use cases seamlessly.

Verisoul.ai: Real-time fake user detection

Verisoul offers a unified AI-powered platform to detect and prevent fake users, ensuring accounts are real, unique, and trustworthy.

Challenge: Verisoul previously built and maintained 10 different independent services across Postgres, Cassandra, and Neo4j to handle a variety of types of data, such as network intelligence, device intelligence, behavioral and sensor data, email and multi accounting. This complexity made it difficult to provide zero-latency detection to counter the speed, scale and sophistication of modern-day fraud attacks,
Solution: By consolidating onto Spanner, Verisoul now can monitor hundreds of customers with millions of accounts in real time, capturing every login, page view, click, and mouse move. Spanner provided an all-in-one database for Graph, vector search, and seamless integration with BigQuery, allowing them to eliminate maintenance overhead while delivering unlimited throughput all with a simple architecture.

2. Recommendation engines

Personalized recommendations are at the heart of online consumer businesses. Building an effective recommendation engine requires analyzing vast amounts of user behavior data, product and service attributes, and historical interactions. Spanner’s interoperable queries allow you to combine user profiles (relational), interaction history (search), and product similarity (graph) to generate highly relevant recommendations in real time, driving better user engagement and improving conversion rates.

Target: Combining Vector and Graph Search for gift recommendations

Target sought to elevate the holiday shopping experience with a generative AI-powered Gift Finder for highly personalized gift recommendations.

Challenge: The application was run on a specialized search database, providing limited gift recommendations. To enhance and personalize the experience, Target needed a sophisticated upgrade.
Solution: Target selected Spanner Graph for its versatile hybrid query model. The solution blends graph traversals with vector search with their proprietary embeddings delivering intuitive, real-time product suggestions — all delivered just in time for the 2025 Black-Friday-Cyber-Monday shopping rush.

True Digital Group: Consolidating AI search

True Digital Group, Thailand's leading telecom-tech company, offers customers a wide array of high-quality digital services, encompassing both streaming and print media, along with customer loyalty tracking. Their AI-driven intelligent search feature ensures accurate content retrieval based on keywords and user intent.

Challenge: A fragmented stack with multiple databases resulted in outdated data, inconsistent tokenization, multiple query languages, and poor search quality, causing users to avoid the search feature.
Solution: True Digital consolidated all search functionality onto Spanner. By combining keyword and intent-based search results using SQL, they significantly improved search relevancy and accuracy, leading to increased customer engagement and satisfaction.

3. Hybrid search

Information retrieval is the critical bridge that grounds AI models in factual, up-to-date data and enables agentic workflow. Often, users must locate a specific needle in a haystack — searching through a massive corpus of legal documents, financial reports, or research papers. Interoperable multi-model Spanner empowers customers with hybrid search capabilities, ensuring AI models retrieve the most relevant context at any scale with pinpoint accuracy.

Rogo: Financial workflow automation

Rogo connects proprietary internal data with external financial sources like filings, PitchBook, LSEG, FactSet, and S&P Capital IQ to help finance professionals automate their workflows, from building pitch decks to drafting investment memos.

Challenge: Rogo needs to ingest and connect data from dozens of sources at once, across both structured and unstructured formats. Finding the right backend to support that wasn't straightforward.
Solution: Rogo chose Google Cloud Spanner for its high performance, scalability, and easy management. It lets them store and query both relational and document-based data in one place, which has made it easier to audit and maintain as the platform has grown.

Inspira: Streaming legal intelligence

Inspira, a leading legal tech company, provides AI-driven solutions tailored for legal research and general workforce optimization. Their platform serves law firms, corporations, and government entities, managing a massive repository of 75 million legal documents and 440 million vectors.

Challenge: Before migrating to Spanner, Inspira struggled with a complex, fragmented architecture, relying on a polyglot system consisting of Elasticsearch, BigQuery and Cloud SQL. This led to complicated data synchronization, and complex “two-stage” query filtering to combine keyword and vector searches. The team also needs a path to scale beyond 1 billion vectors without sacrificing latency and high read/write throughput.
Solution: To resolve these inefficiencies, Inspira consolidated their entire stack intoSpanner, drastically simplifying a 4.5 TB data pipeline into a unified, high-performance single-source of truth. Leveraging Spanner’s native support for both FTS and vector search, Inspira enabled single-stage filtering for hybrid queries and achieved high-precision snippets for LLM-based legal analysis with RAG workflow.

4. Autonomous network operations

Autonomous network operations (ANO) represents the transition from reactive maintenance to predictive, self-healing networks. By creating a comprehensive digital twin of the network topology and overlaying it with real-time operational data, telecommunications providers can automate root cause analysis, predict anomalies, and resolve network incidents without human intervention.

MasOrange: The digital twin

A temporal digital twin is at the heart of MasOrange’s ANO efforts, replicating its country-wide wireless network topology, alongside operations support systems (OSS), and business support systems (BSS) data.

Challenge: MasOrange needed a graph database that was highly available, infinitely scalable with zero RPO/RTO to serve as the foundation of its ANO stack. They required vector, and FTS capabilities without the operational overhead of managing multiple disparate solutions.
Solution: MasOrange chose Spanner for its ability to meet strict scalability and availability requirements while offering fully interoperable Graph, vector and FTS capability. Today MasOrange’s digital twin is live on Spanner, powering end-to-end anomaly detection and root cause analysis.

Looking Ahead

With the scale insurance, high reliability, global consistency, and versatility in handling different data models interoperably, Spanner is a future-proof database for your agentic workload.

We envision a future where the database becomes a simple implementation detail, allowing you to focus purely on accelerating developer productivity, improving operational efficiency and delivering your business goals. Visit our Spanner page to learn more and get started today.

How ID.me Scaled to 160M Users While Reducing Operational Risk

Mon, 30 Mar 2026 16:00:00 +0000

Editor’s note: ID.me is transforming digital identity security, proving that establishing your identity can be easy. What's more, their platform has scaled to support 160 million members and can support up to 40,000 users per minute.To support services like tax filing that require massive scale and power real-time AI, the team migrated 50 terabytes of data from their legacy platform to Google Cloud, adopting a modern architecture on AlloyDB, Cloud SQL, and Vertex AI. This architecture resulted in faster development, more accurate fraud detection, and a 40% reduction in their data teams’ overall work completion time.

Phone, driver’s license, credit card – you probably don’t leave home without some form of ID. It proves who you are, and it works almost anywhere. But online, you’re made to prove your identity again and again, and create new logins for every new service or tool you use. At ID.me, we think people should be able to verify their identity once, securely, and bring that same credential everywhere they go online.

Our goal is to create the digital wallet for identity: a trusted sign-in that works across the public and private sectors. Today, we serve over 160 million members. As identity grows ever more essential to how we live and work, we’re scaling to make it as easy to prove who you are online as it is to flash your driver’s license in person.

But demand doesn’t wait

In the last few years, ID.me has gone from serving 50 million members to more than 160 million. We no longer track usage by the day, but instead monitor it moment-to-moment. At this stage, our platform is designed to support up to 40,000 members per minute. That kind of demand changes the equation. For ID.me members, access is everything. It’s not just about uptime; it’s about ensuring that when a member needs to prove who they are—whether for government benefits, healthcare, or exclusive offers—we verify them securely and instantly. As usage grew, we realized our infrastructure couldn’t scale the way we needed it to. We were reaching the limits of our architecture. So we made the decision to rebuild the ID.me data foundation on Google Cloud for our next phase of growth.

A new database for us and our 160 million closest friends

The first order of business was to select a database that could truly provide the massive scalability and reliability we needed. AlloyDB for PostgreSQL immediately stood out. It directly addressed the scaling bottlenecks and operational complexities we faced with our prior setup, allowing us to confidently handle our peak demands. This shift didn't just solve our technical hurdles; it dramatically improved our developer experience. Our teams now spend far less time on provisioning, maintenance, and patching, accelerating our development cycle from weeks to just days.

Over the last two years, we migrated more than 50 terabytes of data across 15 database instances to Google Cloud – with minimal downtime. We also introduced a two-tier architecture where Cloud SQL supports our smaller and more standard services, while AlloyDB runs heavier workforces that form the backbone of the ID.me platform. That way, we can move fast without sacrificing stability. It also frees up our teams to focus on the work that actually drives innovation, like making sure AI works for, not against, our identity efforts.

aside_block: <ListValue: [StructValue([('title', 'Build smarter with Google Cloud databases'), ('body', <wagtail.rich_text.RichText object at 0x7f89702d3fd0>), ('btn_text', ''), ('href', ''), ('image', None)])]>

Training AI to fight . . . AI?

Of course, everyone is racing to use AI. At ID.me, it’s just as important to defend against its misuse. The threat landscape is evolving, especially as generative models get better at impersonating individuals and even creating synthetic identities. And since we’re in the business of verifying that people are who they say they are, that threat lands squarely on our doorstep.

One of the great things about AlloyDB is its ability to create multiple read pools. For us, those read pools have become data clean rooms that we can quickly share out with our data engineers and data scientists. Fraud analysts can go in, find what’s wrong, and either remediate or prevent it in real time.

Overall, AlloyDB AI has allowed us to scale our systems 10-20X of what we were able to handle – and with a decrease in price to boot. The impact of this is huge. ID.me has been recognized by the U.S. federal government for its role in preventing large-scale fraud within national systems. A crucial factor in this success was AlloyDB's built-in high availability and easy-to-scale read pools, which enabled the Internal Revenue Service (IRS)—the U.S. national taxing authority— to seamlessly process over 120,000 transactions per second during the last peak tax season without a blip. This effectively doubled their previous self-hosted PostgreSQL performance.

We’ve been playing with a lot of new capabilities, but the ones we’re most excited about are Gemini Enterprise and AlloyDB AI natural language as they represent a fundamental shift in how we build and interact with AI.

One login. Every system. Zero friction.

Our data teams are big fans of Google Cloud; it's made their work substantially easier. Since migrating, they can make changes much faster, leading to a 40% reduction in their overall work completion time. And across ID.me engineering teams, the developer experience has improved dramatically. Our teams can ship full product features in days instead of weeks, spending more time solving meaningful problems for our members.

We’ve been able to scale both our infrastructure and trust. With a platform that’s faster, smarter, and built to handle portable identity at massive scale, we’re one step closer to our goal: a secure, digital way to prove who you are, wherever you need it, that works everywhere you need it. You may never get asked security questions about your childhood pet again.

Learn more:

Learn more about the power of AlloyDB and AlloyDB AI.
Start your AlloyDB free trial today.
Learn how customers like Bayer and Character.ai are leveraging AlloyDB to transform their business.

Manhattan Associates powers over a billion daily API calls with Google Cloud databases

Thu, 26 Mar 2026 16:00:00 +0000

Editor’s note: Manhattan Associates, a global leader in supply chain and omnichannel commerce solutions, modernized its Manhattan Active SaaS platform by moving from legacy Oracle and DB2 systems to Google Cloud databases. With Cloud SQL and BigQuery, the company now processes over a billion API calls per day with average response times under 150 milliseconds, supporting hundreds of thousands of monthly active users across tens of thousands of stores and distribution centers.

From monolithic roots to cloud resilience

Our Manhattan Active SaaS platform supports global supply chains, requiring constant uptime and performance. Our legacy Oracle and DB2 infrastructure created operational drag through manual scaling, complex licensing, and high maintenance overhead. We needed a new database foundation that provided contractual SLAs for availability, automated resilience, and a predictable cost model.

We chose Google Cloud databases because they give us the right balance of flexibility, scalability, and operational simplicity needed to run Manhattan Active at global scale. With managed databases like Google Cloud SQL, we traded manual upkeep for built-in high availability, scalability, and cross-region disaster recovery.

Each capability of Manhattan Active now runs as an independent, containerized service orchestrated by Google Kubernetes Engine (GKE). Data streams through Pub/Sub into BigQuery for real-time analytics, while Cloud Logging and Cloud Monitoring deliver observability at scale. This microservices-first design, powered by Google Cloud’s managed services, gave us the agility to evolve faster and the confidence that mission-critical operations would remain resilient across regions.

aside_block: <ListValue: [StructValue([('title', 'Build smarter with Google Cloud databases.'), ('body', <wagtail.rich_text.RichText object at 0x7f898525c160>), ('btn_text', ''), ('href', ''), ('image', None)])]>

Building resilience and speed into every transaction

With a new foundation on Google Cloud, we could rethink how our platform delivers value at scale. The Manhattan Active architecture works alongside managed databases to turn supply chain complexity into responsive, resilient systems.

The benefits show up across the platform. Cloud SQL powers the core of Manhattan Active, quickly and reliably running millions of supply chain transactions per day. Real-time analytics flow into BigQuery, giving retailers sharper forecasting and faster anomaly detection. Automated failover and cross-region replicas safeguard business continuity, so critical services stay available even when disruptions hit.

A modernized foundation: From database to intelligence

The move from legacy Oracle and DB2 systems to Google Cloud databases solved more than just a performance issue; it gave us a resilient foundation for what came next. That reliability and scale let Manhattan Associates bring generative AI directly into the supply chain.

Our Agentic AI suite includes prebuilt agents — like the Intelligent Store Manager and Labor Optimizer — that coordinate real-time decisions across store and distribution center operations. The Manhattan Agent Foundry also lets customers build custom AI agents using a low-code environment. That same foundation powers internal efficiency too, with use cases like real-time log analysis, developer code assistance, and scenario simulations.

Unprecedented speed, scale, and operational efficiency in practice

For retailers, the impact of this platform modernization is immediate: tangible speed and reliability.

With the help of Cloud SQL and BigQuery, Manhattan Active now supports an astounding number of API calls per day, with an average response time under 150 milliseconds. This speed supports hundreds of thousands of monthly active users across tens of thousands of stores and distribution centers, enabling real-time decision-making where it matters most.

Operationally, the platform has become more elastic and efficient. The system automatically handles hundreds of thousands of scaling events per day, ensuring performance remains consistent during peak surges without expensive overprovisioning. Database observability tools like query insights give engineers clear visibility, so we spend less time on database patching and reactive troubleshooting and more on feature development and performance tuning.

For Manhattan Associates, resilience is now a built-in capability. And for retailers depending on our software, that translates into supply chains that are smarter, faster, and ready for whatever comes next.

Learn more

Discover how Cloud SQL can transform your business! Start a free trial today!
Download this IDC report to learn how migrating to Cloud SQL can lower costs, boost agility, and speed up deployments.
Learn how Ford and Yahoo gained high performance and cut costs by modernizing with Cloud SQL.

Streamline read scalability with Cloud SQL autoscaling read pools

Wed, 18 Mar 2026 16:00:00 +0000

A common pattern for applications that read frequently from a database is to offload read-heavy workloads to a read replica. This allows applications to scale without impacting critical write operations on the primary database instance. However, these read-heavy workloads can easily exceed the capacity of a single read replica. While developers can manually implement multiple replicas behind a load balancer, this approach is complex and difficult to maintain and scale.

Cloud SQL provides a simplified, fully managed solution to scale your reads using read pools for MySQL and PostgreSQL. This feature allows you to provision multiple read replicas that are accessible via a single read endpoint, so you can easily add and remove read replicas without having to make any application changes. To further improve efficiency, we recently introduced autoscaling for Cloud SQL read pools, which dynamically adjusts your read capability based on real-time application needs. Read pools with autoscaling are now generally available under Cloud SQL Enterprise Plus edition. You can find more details in the MySQL and PostgreSQL documentation, as well as the release notes.

Why read pools?

When you create a read pool replicating from a primary instance, Cloud SQL automatically provisions multiple read replicas, referred to as read pool nodes, and creates a single load balancer (called the read endpoint) that dispatches queries to the nodes round-robin. A pool can contain anywhere from 1 to 20 nodes.

A read pool is managed as a single entity to reduce your operational burden. The pool represents a homogeneous set of nodes; whenever you make any configuration change such as updating database flags, the VM type, or other parameters, the change is automatically applied to all nodes in the pool. You can also scale out and in by adding or removing nodes from the pool at any time in response to changes in the workload. Since all queries flow through the read endpoint's load balancer, you don’t need to reconfigure your applications, even as the nodes in the pool are being updated, added, or removed.

Easier scaling with read pool autoscaling

Read pools truly shine in scenarios with variable workloads. With the general availability of autoscaling for Cloud SQL read pools, managing these fluctuations is even easier.

Key benefits of read pool autoscaling:

Manage traffic spikes automatically: Maintain application responsiveness during peak demand as the pool dynamically scales up to 20 nodes based on database connections or CPU usage.
Simplify operations: Because autoscaling is integrated with a single read endpoint, your applications stay connected to the same address even as underlying nodes are adjusted.
Optimize costs: Pay only for the resources you actually use by automatically scaling in during low-traffic periods, preventing the expense of over-provisioning.

How Cloud SQL improves availability with read pools

Engineered for mission-critical reliability, Cloud SQL read pools provide the foundation for your high-availability read workloads. By maintaining at least two nodes, your read pool is backed by a 99.99% availability SLA, which includes coverage for maintenance downtime. To do this, Cloud SQL intelligently manages your environment in the following ways:

When you add nodes to a read pool, existing connections continue without interruption on the preexisting nodes. New connections may be directed to the newly added nodes and the load shifts to be evenly balanced across the nodes as pre-existing connections finish their work.
For read pools containing two or more nodes, when you modify the VM type or database flags, or perform most other configuration updates, the read pool is updated with near-zero downtime.
As for any other Cloud SQL instance, Cloud SQL automatically detects and repairs issues with the underlying hardware for your read pool instances. Whenever a node is found to be unhealthy, it is removed from the load balancer rotation and a new node is created to replace it.

Enabling and using read pools with autoscaling

Read pools excel in environments with highly variable workloads. With Cloud SQL autoscaling, you can define minimum and maximum node counts while setting targets for key metrics like CPU utilization or database connections.

Retail is a prime example. Traffic often fluctuates based on daily cycles, seasonal shifts, or flash sales. By dynamically scaling out to meet these peaks and scaling once demand subsides, you provide a strong customer experience without the overhead of an over-provisioned environment.

If you've already created an Enterprise Plus edition instance, you can create a read pool that replicates from it. The following command creates a read pool with two nodes and enables autoscaling, configured to scale between 2 and 10 nodes based on keeping CPU utilization around 60%:

code_block: <ListValue: [StructValue([('code', 'gcloud sql instances create myautoscaledreadpool \\\r\n --tier=db-perf-optimized-N-4 --edition=ENTERPRISE_PLUS \\\r\n --instance-type=READ_POOL_INSTANCE \\\r\n --master-instance-name=myprimary \\\r\n --region=us-west1 \\\r\n --node-count=2 \\\r\n --auto-scale-enabled \\\r\n --auto-scale-min-node-count=2 \\\r\n --auto-scale-max-node-count=10 \\\r\n --auto-scale-target-metrics=AVERAGE_CPU_UTILIZATION=0.60'), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7f89833e7af0>)])]>

You can obtain the read endpoint's IP address by inspecting the instance in Google Cloud console or using the gcloud CLI. This IP address doesn't change for the lifetime of the pool.

If you have an existing read pool, you can easily enable autoscaling on it:

code_block: <ListValue: [StructValue([('code', 'gcloud sql instances patch myreadpool \\\r\n --auto-scale-enabled \\\r\n --auto-scale-min-node-count=2 \\\r\n --auto-scale-max-node-count=10 \\\r\n --auto-scale-target-metrics=AVERAGE_DB_CONNECTIONS=100'), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7f89833e7a90>)])]>

This command configures myreadpool to automatically scale between two and 10 nodes, targeting a maximum of 100 database connections per node on average.

While autoscaling is recommended for most variable workloads, you can still manually scale your read pool if needed by updating the node count directly:

code_block: <ListValue: [StructValue([('code', '# Manually scale out to 4 nodes\r\ngcloud sql instances patch myreadpool --node-count=4\r\n\r\n# Manually scale back in to 2 nodes\r\ngcloud sql instances patch myreadpool --node-count=2'), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7f89833e78e0>)])]>

Get started today

By using read pools with new autoscaling capabilities, you can avoid the tedious tasks associated with managing a large set of read replicas, and give your applications the read capacity they need, when they need it, without overspending. To get started today, please refer to the following resources:

Cloud SQL documentation for read pools
- About read pools (MySQL, PostgreSQL)
- Creating and managing read pools (MySQL, PostgreSQL): includes example commands for gcloud CLI, Terraform, and REST API
- Read pool autoscaling (MySQL, PostgreSQL): includes example commands for gcloud CLI, Terraform, and REST API
Sign up for the new Cloud SQL free trial, a dedicated 30-day program designed to give both new and existing Google Cloud users hands-on access to premium, enterprise-grade features of Cloud SQL (PostgreSQL and MySQL).

Next-gen caching with Memorystore for Valkey 9.0, now GA

Wed, 18 Mar 2026 16:00:00 +0000

Backend developers and architects building high-throughput, low-latency applications increasingly rely on Valkey, an open-source, high-performance key-value datastore that supports a variety of workloads such as caching and message queues. At Google Cloud, we offer a fully managed version as part of Google Cloud Memorystore, and today, we’re excited to announce the general availability (GA) of Valkey 9.0, delivering both massive performance gains and powerful new developer capabilities.

During the preview, we saw remarkable uptake and excitement from customers who require the highest levels of performance. Organizations pushing the boundaries of scale and latency are putting Valkey 9.0 to the test for their most demanding workloads:

"A high-performance caching layer is critical to our infrastructure at Snap. We are excited to see the General Availability of Valkey 9.0 on Google Cloud Memorystore. The new architectural enhancements, including SIMD optimizations, offer strong performance benefits for throughput and latency. Having access to a managed service backed by an open standard gives us valuable flexibility in how we deploy and manage our caching workloads." - Ovais Khan, Principal Software Engineer, Snap

This need for uncompromising speed and flexibility extends beyond social networking infrastructure into the financial sector, where real-time transaction processing and reliability are non-negotiable.

"In the financial services sector, milliseconds matter and data reliability is paramount. By utilizing Memorystore for Valkey on Google Cloud in the critical GPay stack powering the UPI Acquirer Switch for Top Indian Banks, Juspay is proud to leverage Memorystore to handle high-throughput transactional data with exceptionally low latency.

We are excited about the GA release of Valkey 9.0. The performance gains from features like pipeline memory prefetching, combined with the assurance of a fully managed, truly open-source solution, provide us with the scale and reliability necessary to securely serve all our customers." - Arun Ramprasadh, Head of UPI, Juspay

Similarly, in the media and entertainment space, delivering uninterrupted experiences to massive audiences requires a caching layer capable of instantly absorbing traffic spikes.

"During the preview of Valkey 9.0 on Google Cloud Memorystore, we've experienced amazing performance and stability. Live streaming demands absolutely minimal latency and maximum throughput. The architectural enhancements in Valkey 9.0 allow us to scale our caching layer more efficiently to handle traffic spikes during major events. Relying on a fully managed, open-source solution ensures Fubo can deliver a seamless viewing experience to our audience." - Kevin Anthony, Platform Engineering Manager, Fubo

Performance at scale: Speed without compromise

Valkey 9.0 is engineered for raw speed. By building upon the enhanced IO threading architecture introduced in Valkey 8.0, Valkey can handle significantly higher throughput and reduced latency on multi-core VMs. The performance gains are driven by several architectural enhancements as highlighted in the official Valkey 9.0 release announcement on the Valkey blog:

Pipeline memory prefetching: This optimization increases throughput by up to 40% by improving memory access efficiency during pipelining.
Zero copy responses: For large requests, this feature avoids internal memory copying, yielding up to 20% higher throughput.
SIMD optimizations: By utilizing SIMD for BITCOUNT and HyperLogLog operations, Valkey 9.0 delivers up to 200% higher throughput for these common tasks.

(Note: The figures above are based on open-source benchmarks; actual performance improvements will vary depending on your specific workloads.)

The mechanics of throughput: Pipelining

In Valkey, latency is largely constrained by network round-trip-time (RTT). If an application waits for each response before initiating the next request, total throughput remains bound by this round-trip latency. Pipelining addresses this by disaggregating latency and throughput, allowing multiple requests to be sent over a single connection without awaiting immediate responses.

Most Valkey clients offer native support for pipelining, with some like valkey-go providing “auto-pipelining” capabilities that handle this optimization transparently for developers.

Valkey 8.0 unlocked significant performance leaps with background thread memory prefetching. This architectural optimization provides a hint to the hardware to move data from DRAM into CPU caches before the main thread requires it. While this allowed the main thread to execute operations with minimal delay, it was limited; in pipelined traffic, only the initial operation benefited from prefetching, leaving subsequent requests bottlenecked by CPU cache misses.

Building on this foundation, Valkey 9.0 re-architects command processing to specifically optimize for pipelined workloads. Rather than prefetching keys solely for the first operation, Valkey 9.0 prefetches all operations within a pipeline simultaneously.

While seemingly incremental, the impact is profound, given that CPU caches are typically dozens of times faster than DRAM access. By maximizing CPU cache hits on the main thread, Valkey 9.0 delivers up to 40% higher overall throughput.

New capabilities for modern developers

Beyond raw performance, Valkey 9.0 addresses top feature requests from the community with powerful new commands that simplify application logic and data management.

Granular hash field expiration

A frequent request for in-memory stores is the ability to expire individual fields within a hash rather than the entire key. Valkey 9.0 introduces this capability, allowing for much more flexible data lifecycle management.

Real-world example: Imagine managing a "User Session" hash where you store authentication tokens, temporary preferences, and long-term settings. With HEXPIRE, you can set a 30-minute expiration on a temporary_session_token field while keeping the rest of the user's profile data intact. This eliminates the need to break a single logical object into multiple keys just to handle different Time-To-Lives (TTLs).
New commands: This release adds full support for HEXPIRE, HEXPIREAT, HEXPIRETIME, HPERSIST, HTTL, and more.

Advanced geospatial and conditional logic

Polygon search for geospatial indices: You can now query location data by a specified polygon. For instance, a logistics application can define precise, non-circular delivery zones — like a specific neighborhood or industrial park — and query for all active assets currently within those exact boundaries.

To use this capability, simply add your coordinates to a geospatial index:

code_block: <ListValue: [StructValue([('code', '> GEOADD gcp:regions -121.1851 45.5946 "us-west1 (Oregon)" -118.2437 34.0522 "us-west2 (Los Angeles)" -111.8910 40.7608 "us-west3 (Salt Lake City)" -115.1398 36.1699 "us-west4 (Las Vegas)" -95.8608 41.2619 "us-central1 (Iowa)" -77.4874 39.0438 "us-east4 (N. Virginia)"\r\n\r\n(integer) 6'), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7f89801b2460>)])]>

Then use GEOSEARCH with the BYPOLYGON option:

code_block: <ListValue: [StructValue([('code', '> GEOSEARCH gcp:regions BYPOLYGON 5 -125.00 49.00 -125.00 32.00 -114.00 32.00 -109.00 42.00 -116.00 49.00\r\n\r\n1) "us-west2 (Los Angeles)"\r\n2) "us-west4 (Las Vegas)"\r\n3) "us-west3 (Salt Lake City)"\r\n4) "us-west1 (Oregon)"'), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7f89801b2d30>)])]>

Conditional delete (DELIFEQ): This command deletes a key only if its current value matches a specified value. This is a game-changer for distributed locking. Now, a worker can safely release a lock only if it still holds the unique token it originally used to acquire it, ensuring it doesn't accidentally delete a lock that has already expired and been re-assigned to another process. Previously, this kind of operation was only possible using Lua scripting.

To understand the usefulness of DELIFEQ, suppose two processes (Process A and Process B) need exclusive access to the lock.

Process A executes the following to acquire the distributed lock for 30 seconds:

code_block: <ListValue: [StructValue([('code', '>SET distributed_lock process_A NX PX 30000\r\nOK'), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7f89827abac0>)])]>

If Process B attempts to do the same, it won’t succeed due to the NX option:

code_block: <ListValue: [StructValue([('code', '> SET distributed_lock process_B NX PX 30000\r\n(nil)'), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7f89827ab5e0>)])]>

When Process A finishes, it can safely remove the lock with the new DELIFEQ command:

code_block: <ListValue: [StructValue([('code', '> DELIFEQ distributed_lock process_A\r\n(integer) 1'), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7f89827ab8b0>)])]>

Now, Process B is free to acquire the lock:

code_block: <ListValue: [StructValue([('code', '> SET distributed_lock process_B NX PX 30000\r\nOK'), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7f89827ab550>)])]>

Enhanced visibility and debugging

The CLIENT LIST command now supports robust filtering options. Platform engineers can now quickly isolate clients by flags, name, library version, or even specific databases. This makes it significantly easier to identify and troubleshoot legacy clients or specific application instances that may be causing performance bottlenecks in a complex microservices environment.

New filtering option	Description
NAME <name>, NOT-NAME <name>	Include or exclude clients by their specified name.
IDLE <seconds>	Include only clients that have been idle for over a number of seconds.
FLAGS <flags>, NOT-FLAGS <flags>	Include or exclude clients based on various flags (e.g., whether they are primary or replica connections).
LIB-NAME <name>, LIB-VER <version>, NOT-LIB-NAME <name>, NOT-LIB-VER <version>	Include or exclude clients based on their provided library version or name (e.g., only valkey-py version 6).
DB <db>, NOT-DB <db>	Include or exclude clients based on their current selected DB.
CAPA <flags>, NOT-CAPA <flags>	Include or exclude clients based on their declared capabilities (e.g., whether they support redirection or not).
IP <ip>, NOT-IP <ip>	Include or exclude clients based on their IP address.
NOT-ID <id>, NOT-TYPE <type>, NOT-ADDR <address>, NOT-LADDR <local address>, NOT-USER <username>	Many existing filters now support an inverse option.

The extended CLIENT LIST capabilities are available in Memorystore for Valkey out of the box.

Enterprise-ready features

With this release, we are also introducing the clustered databases configuration, which facilitates complex use cases like blue-green caching patterns. This allows multiple services to utilize numbered databases to efficiently partition data and maintain logical namespace separation. You can now configure up to 100 numeric databases for namespacing your keyspace even when in cluster mode.

Configuration example: When creating your instance via the Google Cloud CLI, you can specify the number of databases:

code_block: <ListValue: [StructValue([('code', 'gloud memorystore instances create my-valkey-instance \\\r\n--engine-version=VALKEY_9_0 \\\r\n--replica-count=1 \\\r\n--shard-count 10 \\\r\n--engine-configs=cluster-databases=100'), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7f89827ab730>)])]>

The clustered databases configuration lets you efficiently partition data across various services or environments within a single, scalable clustered instance. This provides the logical organizational benefits of multiple databases while leveraging the massive scale of a cluster. For many customers currently constrained by legacy non-clustered Redis instances, this capability removes the friction of complex application re-architecting, making it easier to modernize workloads that rely on multi-database structures.

Seamless upgrades

Existing Memorystore for Valkey customers can take advantage of these new features immediately through a simple, no-downtime, in-place upgrade path. As of today, Valkey 9.0 is available for both cluster-enabled and cluster-disabled instances, ensuring a smooth transition to the latest engine.

Looking ahead

The general availability of Valkey 9.0 represents a significant milestone in the evolution of open source databases. But we’re just getting started. We’re already hard at work engineering the next wave of innovation and will be unveiling even more enhancements for the Valkey ecosystem at Google Cloud Next 2026.

Ready to build?

The best way to experience the power of Memorystore for Valkey 9.0 is to try it out. Get started with the documentation or deploy your first Valkey instance. Don't let having to self-manage Redis hold you back. Experience the simplicity and speed of Memorystore for Valkey today and see how it can power your applications, so you can focus on what matters: innovating and creating impactful applications for your business!

PayPal's historically large data migration is the foundation for its gen AI innovation

Thu, 26 Feb 2026 17:00:00 +0000

With the dawn of the gen AI era, businesses are facing unprecedented opportunities for transformative products, demanding a strategic shift in their technology infrastructure. A few years ago, PayPal, a digital-native company serving hundreds of millions of customers, faced a significant challenge. After 25 years of success in expanding services and capabilities, we’d created complexity in our data analytics infrastructure. Some 400 petabytes of data was spread across a dozen siloed systems due to limitations of scale and acquisitions of companies like Venmo, Braintree, and others.

Our very success in growth and innovation had created complexity that threatened our next evolution.

To continue leading the next wave of innovation in financial services, we knew we had to modernize our data foundation. Today, we’re proud to share how PayPal successfully completed what’s arguably one of the largest data migrations in history, culminating with the move of our analytics to BigQuery, Google Cloud’s enterprise data warehouse. This effort marks a significant leap in creating the robust data framework we’ll need to expand and advance our business priorities and meet the ever-evolving financial needs of our customers.

This migration was essential, but the scale was daunting. In fact, by some measures, such as our now sunset Teradata system, we believe this was one of the biggest data migrations in history. Befitting of such history, we wanted to offer some insights into how we tackled this migration and what others might consider when undertaking a significant migration of their own.

Untapped potential of data

As one of the original digital payment pioneers, PayPal processes billions of transactions, and houses decades of valuable customer insights. We have a mountain of data — really a mountain range — that had developed over decades without being fully leveraged in the service of our customers and merchants.

Each acquisition and new service added valuable capabilities but also introduced new data challenges. For example, a small business owner might use PayPal for online sales and Venmo for local transactions. However, providing a unified view of their business required complex processes that were costly and slow.

The fragmentation of data limited our ability to offer personalized experiences to consumers, thereby reducing the potential to maximize the value of their money and hindering our ability to gain deeper insights from the data.

As the gen AI era dawned, our digital fragmentation was becoming more than just a technical inconvenience. With AI becoming a transformative force in financial services with huge potential ROI, we knew fragmented data would severely limit our ability to create the intelligent experiences customers have come to expect. These could run from further strengthening our industry-leading fraud detection models to providing a best-in-class commerce platform for merchants to help them succeed in the competitive global economy.

To get there, we had to get our disparate data platforms in order, first.

Legacy systems, modern ambitions

The scope was massive. We needed to consolidate multiple data platforms, including what’s believed to be the world’s largest Teradata deployment, along with Hadoop clusters, Redshift, Snowflake, and various other systems processing petabytes of transaction data. This migration also had to be executed while maintaining the uninterrupted security and reliability our customers depend on.

As a technology company, PayPal has considerable internal resources, so we first had to decide whether to tackle this challenge ourselves. We weighed the costs and benefits and decided that if we were to unify and scale our on-premise infrastructure to meet our future needs, the cost and time-to-complete would have been prohibitive. Plus, the innovations in AI were happening at a rapid pace in the cloud. To truly leverage the power of our data, we needed to be where that innovation is happening.

We assessed various data warehousing solutions and chose BigQuery due to its numerous advantages. It is a fully managed, cloud native platform with disaggregated compute and storage that can scale independently. It has powerful capabilities at the scale and performance we needed, and a familiar SQL interface meant a gentler learning curve for our developer community.

Most importantly, BigQuery’s native integrations with AI enable seamless and efficient data analytics.

The journey to unified data

After choosing Google Cloud as our data partner, we embarked on our historic data migration. This may sound hyperbolic, but when you consider the scale of PayPal’s business, the geographies across which we operate, the regulations within each, the sensitive and quite literally valuable nature of this data, the scope of the challenge starts to be clear.

With the help of partners and experts from Google Cloud Consulting, we migrated more than 300 petabytes of data and streamlined operations, decommissioning around 25% of workloads. And we managed this all while maintaining zero downtime of our business operations and with no impact to customers. Here are some key factors that contributed to our success.

Alignment: The first hurdle in achieving transformations at scale is aligning stakeholders on a shared goal. So, we made it an enterprise-wide priority.

Discovery and analysis: Detailed inventories of data, workloads and inbound/outbound data streams is crucial for defining scope, effort and forecasting budget. Establishing lineage allowed us to trace the origins and relationships of various components, thereby providing a clear and comprehensive view of the dependency graphs.

Strategy: It is crucial to establish fundamental principles for the migration process, such as deciding between lift-and-shift versus modernization, defining security principles, setting governance guardrails, and determining how consumption will be tracked.

Execution: We automated every possible task and developed live dashboards to continuously monitor the progress of migrations. FinOps was integrated through the migration process with clear visibility of consumption and performance.

Benefits from BigQuery and beyond

We’ve achieve faster insights. Queries are 2.5x to 10x faster, including complex queries used by data scientists. This unlocks real-time insights, enabling PayPal to personalize product recommendations, offers, and customer support.

We’ve built new AI foundations. Data accessible for model training is 16x fresher. Feature engineering, a crucial step in AI development, is improved by instant access to clean, governed data. This accelerates the development personalized financial guidance, and predictive analytics for both consumers and businesses.

We’ve optimized operations. By migrating to BigQuery Data infrastructure vendors were reduced from four to one, streamlining operations and reducing complexity. Data duplication between platforms was entirely eliminated.

Our new unified data platform in BigQuery has become the source for PayPal's next wave of innovation, enabling us to create more intuitive, personalized experiences across our entire ecosystem and to leverage the power of gen AI.

AI-powered innovation unleashed

Looking ahead, we're exploring how this unified data platform will enable us to deliver AI-powered experiences that weren't possible before, including:

Predictive fraud prevention that spots potential issues before they affect our customers.
Personalized financial insights that help merchants optimize their businesses.
Seamless payment experiences that adapt to each customer's preferences and patterns.
More intelligent risk assessment that could help expand financial access to underserved communities.

Agentic commerce and future possibilities we are now able to imagine.

Lessons for the AI era

While our migration may be extraordinary in its scale, we are not alone in our needs or ambitions. There are ample considerations for companies within and well beyond financial services who may be pondering their own data foundations at this time.

First off, do not underestimate how under-utilized your data may be, and how unorganized.

Making sure your data is centralized, accurate, and consistent paves the way for AI experimentation and deployment. Organizations that spend time cleaning up their data fabric will be able to bring machine learning and generative AI applications to market more quickly, and do so at scale.

Second, ensuring data is accessible to everyone within your organization, with the proper controls, unlocks so much potential. Data orchestration and enterprise search, coupled with generative AI, has the potential to break down longstanding organizational silos and speed up decision-making across your organization. It’s one of the most promising applications of AI.

The financial world will continue to evolve, driven by new technologies and changing customer expectations. PayPal’s data transformation shows how even established companies can reinvent themselves to stay ahead of this change — provided they're willing to tackle the fundamental challenges that stand in their way.

In doing so, we've not only preserved our position as a digital payments pioneer but set ourselves up to continue leading the next wave of innovation in digital commerce.

Serving data from Iceberg lakehouses fast and fresh with Spanner columnar engine

Thu, 26 Feb 2026 17:00:00 +0000

The divide between data in operational databases and analytical data lakehouses is disappearing fast. As businesses increasingly adopt zero ETL lakehouse architectures, the challenge shifts from simply storing data in an open data format such as Apache Iceberg to serving it with the low-latency performance and speed that modern applications and AI agents require. Whether it’s a cybersecurity provider like Palo Alto Networks requiring real-time threat detection insights, or a telecommunications giant like Vodafone looking to reimagine data workflows for better customer experiences, organizations need to serve precomputed insights and AI models at scale.

To help, today, we are excited to announce the preview of the Spanner columnar engine, which allows you to serve your Iceberg lakehouse data with the scale and low latency of Google’s Spanner.

Uniting OLTP and analytics: The Spanner columnar engine

Traditionally, organizations had to choose between the high-performance transactional capabilities of an OLTP database and the analytical power of a columnar warehouse. Spanner’s columnar engine ends this trade-off by uniting these two worlds in a single, horizontally scalable system.

The columnar engine uses a specialized storage mechanism designed to accelerate analytical queries by speeding up scans up to 200 times on live operational data. By storing data in a columnar format alongside traditional row-based storage, Spanner can execute complex queries automatically using vectorized execution — processing batches of data at once rather than row-by-row. Most importantly, this performance boost can be isolated from critical transaction workloads, so that customer-facing applications remain responsive while you gain real-time insights from your operational datastore.

New features

Since we first announced Spanner columnar engine, we’ve added several new capabilities to accelerate performance and enhance usability. These include:

Vectorized execution: The engine supports faster columnar scans and aggregations using vectorized execution to process data more efficiently.
Automatic query handling: Spanner automatically redirects large-scan analytical queries to the columnar representation, speeding up analytical queries without affecting concurrent transactional workloads, allowing for true hybrid processing.
On-demand columnar data conversion: In addition to automated columnar data conversion, a new major compaction API helps accelerate the conversion of existing non-columnar data into the columnar format.

Why Iceberg data needs a fast, low-latency serving platform

Apache Iceberg has become the standard for open lakehouse architectures, providing a robust way to manage massive open format datasets in cloud-based storage. However, while lakehouses are excellent for large-scale analytics, they aren’t usually designed for the sub-second, high-concurrency "point lookups" or aggregated serving that live applications require.

This is where Spanner provides a unique value proposition. By moving curated, processed data from your lakehouse into Spanner — a process known as reverse ETL — you transform "cold" analytical data into "hot" operational data. Spanner provides the global consistency and high availability that applications require, making your Iceberg data accessible via low-latency APIs for real-time decision-making and agentic AI features.

Benchmarking Spanner columnar engine

To demonstrate Spanner’s new serving capabilities, we used a leading industry benchmark for web analytics and real-time dashboards — the exact scenarios where low-latency serving is critical.

Our benchmark results with a single Spanner node showcase the power of the columnar engine:

Benchmark query	Spanner columnar engine speedup
Simple count of all records	46.3×
Basic aggregation with filtering	32.7×
High selectivity scan	46.7×
Global aggregation	58.6×

These results represent the acceleration of real world workloads on Spanner columnar engine and show that Spanner can take complex, scan-heavy queries and return results in milliseconds, making it a great choice for powering real-time dashboards and user-facing features. Spanner is now a high-performance engine capable of delivering complex analytical results at the speeds required by modern digital experiences.

Universal reverse ETL: Serving data from all lakehouses

Spanner is designed to be the serving layer for your entire data ecosystem. Whether your lakehouse lives in BigQuery, Snowflake, Databricks, or Oracle, Spanner offers an integrated pathway for high-speed serving.

Through our latest reverse ETL workflows, you can easily bridge the gap between your analytical and operational worlds:

BigQuery: The tight integration between Spanner and BigQuery provides a powerful, bidirectional bridge for managing Iceberg data across both operational and analytical environments. You can perform federated queries on BigLake Iceberg and Spanner tables using BigQuery external datasets for Spanner, allowing for real-time analysis without moving data. When you need to serve curated BigQuery insights at scale, reverse ETL workflows can push data from BigQuery and BigLake Iceberg tables directly into Spanner. Furthermore, you can capture live operational changes in Spanner and stream them into BigQuery and BigLake Iceberg tables using Datastream, ensuring your lakehouse remains synchronized with your transactional data from Spanner for agentic AI and real-time decision-making.
Databricks: Using Databricks' Universal Format (UniForm), you can generate Iceberg metadata for your Delta Lake tables automatically. This allows Spanner to ingest your processed Databricks data via BigQuery or Dataflow, so that your "curated" datasets are ready to power applications with minimal engineering overhead.
Snowflake: You can export Iceberg tables to Google Cloud Storage and use BigQuery BigLake as a zero-copy intermediary to push that data directly into Spanner via EXPORT DATA commands. Alternatively, for simpler migrations, you can export Snowflake data as CSVs and use Dataflow templates for high-throughput ingestion into Spanner.
Oracle Autonomous AI Lakehouse: Oracle Goldengate 26ai now allows you to replicate your Oracle Autonomous AI Lakehouse data into Spanner to serve insights generated from Oracle’s data ecosystem with Spanner’s scale and consistency.

Get started today

It’s time to stop waiting for your lakehouse queries to finish and start serving your data hot, fresh, and fast with Google Spanner powered by Spanner columnar engine. The Spanner columnar engine is now in preview. You can enable it on your existing Spanner tables today with a simple DDL change.

You can see the performance acceleration of Spanner columnar engine by running the analytics benchmark queries on Spanner available on Github for yourself.

To help you get started, here are codelabs to build reverse ETL pipelines to Spanner from

Databricks:

and

Snowflake:

Decommission your legacy Apache Cassandra stack and build for the future with Spanner

Thu, 19 Feb 2026 17:00:00 +0000

An increasing number of customers are migrating to Spanner from legacy NoSQL environments like Apache Cassandra. The strategic drivers are evident: a markedly lower total cost of ownership (TCO), elastic scalability, and near-zero maintenance overhead.

With the general availability of the native endpoint enabling Cassandra Query Language (CQL) APIs on Spanner, your existing Cassandra applications can now leverage Spanner’s enterprise foundation, featuring strong consistency, virtually limitless scale, and 99.999% availability — all while utilizing familiar CQL.

Better yet, migrating your application to Spanner with the CQL interface typically requires only a one-line code change, as your existing CQL statements remain valid. Combined with our integrated, high performance bulk and live migration tools, your transition from Cassandra to Spanner is simple.

Beyond NoSQL: Strategic solutions for Cassandra Users

While the CQL API facilitates the move, Spanner addresses the fundamental data integrity and operational constraints inherent in traditional Cassandra architectures:

Global ACID transactions: Minimize concerns regarding eventual consistency. Achieve comprehensive global ACID transactions to help ensure data integrity at any scale.
Powerful indexing: Strongly consistent secondary indexes support complex query patterns with built-in optimization and no integrity risks.
Rich SQL: Utilize a sophisticated SQL interface that supports joins and aggregations.
High reliability: Benefit from 99.99% availability in regional setups and 99.999% in multi-regional configurations.
Compliance and latency: Simplify data residency compliance with geo-partitioning, delivering low-latency local reads and writes to a global user base.
Built-in observability: Access a suite of performance metrics and charts directly in the Google Cloud console at no additional cost.

The native CQL endpoint provides a clear pathway to decouple your existing Cassandra applications and modernize them using the full power of Spanner. Let’s look at the next steps after migrating your data and applications from Cassandra to Spanner.

Tweak Spanner for your specific workload

Following your migration, here’s how to optimize your Spanner environment for your workload.

1. Optimize costs and operational efficiency

Workload Characteristics	Recommended Solution	Primary Benefit
Write-intensive traffic	Throughput-optimized writes	Up to a 6x increase in write throughput via request bundling (with minimal latency impact).
Variable or fluctuating traffic	Autoscaler	Automatically aligns capacity with demand, eliminating over-provisioning costs.
Steady-state, baseline capacity	Committed use discounts (CUDs)	Secure up to 40% savings on steady-state operational costs.
Storage-intensive workloads	Tiered storage (HDD)	Utilize cost-effective HDD storage for a significant reduction in long-term storage expenses.

2. Achieve low latencies
We continuously enhance Spanner's performance to support mission-critical, high-concurrency workloads.

Single-digit millisecond performance: Spanner consistently delivers under 5ms latency for both read and write operations.
Repeatable read isolation: This feature utilizes optimistic concurrency to reduce latency and transaction aborts in read-heavy, low-contention scenarios.
Read leases: Enable strongly consistent reads in multi-region instances without cross-region coordination, maximizing node efficiency and performance.

3. Prepare for peak traffic surges
For planned events like marketing launches or massive data ingestions, you can proactively manage capacity:

Manual split APIs: While Spanner handles data partitioning automatically, the pre-splitting capability allows you to exactly define how your database distributes data ahead of peak loads. This helps ensure immediate utilization of new capacity for stable performance.

4. Isolate operational and analytical pipelines
Prevent resource contention by isolating BI and ETL processes from core operations:

Dedicated resources: Leverage read-only replicas and directed reads to achieve workload isolation.
Advanced analytics: Spanner provides high-performance operational analytics through its columnar engine, and integrates with BigQuery via Data Boost, reverse ETL, and continuous queries.

Deconstruct your Cassandra ecosystem

Migrating from Apache Cassandra to Spanner is a strategic opportunity to decouple your architecture from a complex web of sidecar utilities. While the Cassandra-compatible API serves as the entry point, the true value lies in collapsing the operational "Cassandra tax" into a unified, managed multi-model ecosystem.

Here’s a quick guide to drastically lowering your TCO while significantly boosting performance.

1. Transition easily via connectors

Minimize initial friction by leveraging the Spanner Cassandra Adapter and connectors to port management and application layers with near-zero code modifications:

Orchestration: Maintain your existing Airflow DAGs by redirecting to the Spanner proxy.
Application frameworks: Transition from Spring Data Cassandra to Spring Data Spanner, preserving established repository patterns while unlocking a superior transaction model.
Data processing: Continue leveraging Apache Spark by connecting to Spanner’s native CQL endpoint or using the specialized Spanner Connector for Spark.

2. Retire sidecars via native integrations

Cassandra demands a "babysitting" layer that Spanner automates. Decommission legacy maintenance tools and unlock Spanner’s advanced capabilities:

Anti-entropy: Decommission Cassandra Reaper. Spanner’s Paxos-based replication manages consistency natively, removing the need for manual repair cycles.
Data protection: Replace Medusa with Spanner-native backups. Move away from fragile SSTable snapshots to reliable, Point-in-Time Recovery (PITR).
Observability: Swap complex JMX exporters for Cloud Monitoring. Focus on high-value metrics like Query Insights and Lock Statistics rather than maintenance debt.
Integrated search: Replace external Elasticsearch sidecars and complex ETL pipelines with Spanner full-text search to eliminate indices-synchronization issues.
Modern streaming: Swap legacy CDC for Spanner change streams, providing native integration with Dataflow and Kafka.
Graph analytics: Port JanusGraph to Spanner Graph to run openCypher queries directly on operational data without complex ETL processes.
Query federation: Replace Trino/Presto with BigQuery Federation via Data Boost. Join real-time transactional data with massive data lakes without impacting production I/O.

Build the future with Spanner’s multi-model advantage

Spanner is the definitive always-on database that integrates relational, key-value, graph, search, and vector search capabilities into a single, interoperable platform. By transitioning to Spanner, you eliminate the overhead of managing fragmented databases and unlock the ability to develop innovative applications on a unified data foundation. Build new applications or modernize your legacy applications to take full advantage of Spanner.

Get started today

Ready to refocus on building rather than managing? Initiate your migration today and experience the power of Spanner with your existing Cassandra Query Language.

Codelab: Gain practical experience with the native CQL endpoint.
Free trial: Explore Spanner for 90 days or start for as little as $65/month for a production-ready instance that scales without disruption.
Migration guide: Access deep technical documentation and comprehensive migration resources.

Databases

UKG unlocks real-time workforce intelligence at scale with the Agentic Data Cloud

Unifying the systems behind the suite

Bringing people intelligence to intelligent people

Driving impact across every layer

Learn more

What’s new with Google Data Cloud

April 20 - April 24

April 13 - April 17

April 6 - April 10

March 23 - March 27

March 16 - March 20

February 23 - February 27

February 16 - February 20

February 2 - February 6

January 26 - January 30

January 19 - January 23

January 12 - January 16

What’s new with Google Data Cloud - 2025

Announcing Spanner Omni: Your infrastructure, Google’s innovation

Why Spanner Omni is a big deal

Introducing Spanner Omni

The tech behind Spanner Omni

Where you can get access

Converging operational and analytical data for AI transformation

Flexible, real-time data agents

Grounding agents in a unified governance foundation

Unified engines for deep reasoning

Built for performance at agent scale

Oracle AI Database@Google Cloud: The foundation for the Agentic Enterprise

New capabilities announced at Next ‘26

The agentic future

Get started

What’s new with Databases: Powering the agentic future

Embed AI into every layer of the data stack

Break down walled gardens with lakehouse integrations

Commitment to open data and multi-cloud flexibility

Get started today

What’s new in the Agentic Data Cloud: Powering the System of Action

Enabling agents with a universal context engine

Agentic-first practitioner experiences

A cross-cloud foundation built for agentic scale

Automating the future

Build your success on a System of Action

Introducing BigQuery Graph: Unlock hidden relationships in your data

Graph challenges faced by enterprises across industries

Common BigQuery Graph use cases

BigQuery Graph in the real-world

How to use BigQuery Graph

Ready to get started?

From operational to analytical: The unified Spanner Graph and BigQuery Graph solution

Spanner Graph for operational workloads

BigQuery Graph for analytical workloads

Spanner Graph and BigQuery Graph as a unified solution

One unified solution for all your graph needs

Get started with Spanner Graph and BigQuery Graph today

QueryData helps agents turn natural language into queries for AlloyDB, Cloud SQL and Spanner

The opportunity for agentic systems: from intent to action

The developer’s dilemma: why natural language for agents with databases is hard

Understanding the accuracy gap

How QueryData solves for near-100% accuracy

Deterministic security for your queries

Ease of use for quality hill-climbing and tool integration

What you can build with QueryData today

Next steps

Near-100% Accurate Data for your Agent with Comprehensive Context Engineering

Agentic Tools: A Path to Accuracy and Explainability

The Key Ingredient: Comprehensive Database Context

1. Schema Ontology

2. Query Blueprints

Accuracy As Foundation for Agentic Actions

Spanner's multi-model advantage for the era of agentic AI

The Power of interoperable multi-model databases

Spanner’s AI advantage

Companies making AI a globe-spanning success

Real-world success with Spanner’s fully interoperable multi-model database

1. Fraud detection

2. Recommendation engines

3. Hybrid search

4. Autonomous network operations