Spanner in 2025: Innovations powering intelligent, multi-model AI applications

For over a decade now, Google has relied on Spanner, an always-on, virtually unlimited database to power global-scale applications like Gmail, YouTube, and Google Photos. Today, Spanner handles over 6 billion queries per second at peak and more than 17 exabytes of data, with five 9s of availability, plus global consistency. 

2025 was a big year for Spanner on Google Cloud, too, where it powers demanding workloads for customers like Walmart, Goldman Sachs, Palo Alto Networks, and MercadoLibre. At the same time, in an era defined by artificial intelligence, the role of databases is undergoing a profound transformation. The database is evolving beyond its traditional function as a passive data repository, emerging instead as an intelligent context hub. Spanner now powers key use cases like product search and recommendations, fraud detection, identity resolution, and autonomous network operations (ANO). 

As we gear up for an exciting 2026, let’s take a minute to review some of last year’s highlights. In a nutshell, in 2025, we:

• Made Spanner a first-class citizen for AI workloads

• Bridged the gap between operational and analytical data

• Made migrations simpler

• Raised the bar on price performance, and 

• Enhanced enterprise safety and data protection

Let’s review!

Spanner AI with multi-model support

To build truly intelligent generative AI applications, you shouldn't have to glue together multiple point solutions. This is why in 2024, we integrated vectors, graphs, and relational data to provide context for your AI models. In 2025, we expanded these multi-model capabilities, adding new functionality across graph, vector, text search, as well as Vertex AI integrations. 

• AI capabilities: We introduced multiple capabilities to make Spanner an AI-first database. This includes built-in AI integrations such as ML.PREDICT for natural language queries, MCP Toolbox for Databases, Spanner extension for Gemini CLI, conversational data agents, and Agent Development Kit for Spanner, which together, empower developers to build sophisticated, agentic applications with Spanner. We also extended our multi-model platform to power advanced AI use cases, such as using graphs with vector search for GraphRAG use-cases. Finally, Vertex RAG Engine now lets you use Spanner as a RAG-managed database for data indexing and retrieval operations.

• Graph enhancements: Spanner Graph added support for schemaless data that enables iterative development and frequent updates without requiring schema changes. You can also now build graphs on named schema objects and on SQL views, organizing and encapsulating important aspects of your app.

• Full text search: The enhance_query option provides automatic synonym matching and spell correction for a number of functions, reducing the manual tuning needed to improve recall. 

• Approximate Nearest Neighbor (ANN): ANN with Google's ScaNN is a fast, scalable vector search technique optimized for efficiency in high-dimensional vector searches, crucial for AI recommendation systems, image search, and semantic search. In 2025, we announced the general availability of ANN search to enable developers to perform fast similarity searches on vector embeddings, powering generative AI applications where consistency at scale matters. 

AI-powered operations

In 2025, we delivered multiple AI powered improvements that make it easier for developers to leverage Spanner for their applications: 

• Spanner index advisor: Index Advisor analyzes your query patterns and proactively suggests new indexes (or identifies unused ones) to optimize performance and cost.

• Schema recommendations (Preview): An intelligent tool that scans your schema design for anti-patterns — like hotspot-prone primary keys — and suggests improvements before they become production issues.

• Spanner CLI: The new command line interface, bundled directly with gcloud, allows you to run SQL, manage sessions, and automate scripts within existing developer workflows.

Unified analytics: Minimize silos and complexity

Last year, we bridged the gap between operational and analytic data with multiple new capabilities:

• Spanner columnar engine (Preview): This engine automatically lets you analyze vast amounts of operational data in real time, while maintaining Spanner's global consistency, high availability, and strong transactional guarantees — without impacting transactional workloads. Verisoul.ai, a provider of an all-in-one platform to automate fake account detection used Spanner and Spanner columnar engine to support low-latency transactional writes and rich analytics to provide fast responses to their users.

• The "Better with BigQuery" ecosystem: We strengthened the integrations between Spanner and BigQuery to bring operational and analytical engines together to provide real-time insights:

• Apache Iceberg support: With this capability, data analysts can join live Spanner data with Apache Iceberg tables on BigQuery, as well as export Iceberg data into Spanner. This allows you to query and combine your real-time operational data (from Spanner) with your curated data lakehouse (in Iceberg).

• Spanner external datasets materialized views: We announced the general availability of Spanner external datasets in BigQuery that enable BigQuery users to query live operational data in Spanner with zero ETL requirements. We further extended this capability to integrate with BigQuery materialized views for ultra-fast reporting using pre-computed query results.

• Continuous queries with Reverse ETL (BigQuery to Spanner): You can now combine BigQuery’s continuous query capabilities with reverse ETL to Spanner to stream computed insights — such as fraud alerts or dynamic pricing signals — from BigQuery into Spanner in real-time, for low-latency serving of these insights. Fastweb+Vodafone was able to do just that by leveraging this feature to use Spanner as their real-time serving layer with BigQuery.

Simplified migration

Customers frequently migrate to Spanner to benefit from five nines of availability, virtually unlimited scale, low operational overhead, and global consistency. In 2025, we made it it easier for them to migrate their Cassandra, and MySQL workloads to Spanner:

• Cassandra interface: Spanner’s Cassandra interface lets you take advantage of Spanner's fully managed, scalable, and highly available infrastructure using familiar Cassandra tools and syntax. Lift and shift existing CQL applications with virtually no changes. 

• MySQL interoperability: You can now install a library of 80 MySQL functions to help reduce the application changes required to migrate your MySQL workloads to Spanner.

Performance and cost improvements

Continuously raising the bar on price-performance is a north star for the Spanner team. In 2025, we introduced multiple capabilities to help you improve query performance and reduce costs:

• Repeatable read isolation (Preview): This new isolation level reduces locking overhead to improve latency for workloads that have low read-write contention.

• JSON indexing: Using the same foundation as Spanner’s full-text search, JSON indexing accelerates common queries over JSON, without having to predefine anything about the structure or data. This gives developers flexibility without having to compromise on performance. 

• Read leases: Read leases can improve read latency for strongly consistent data in multi-region configurations by trading off some write performance for common read-mostly workloads. 

• Tiered storage: Tiered storage allows you to manage data lifecycle costs efficiently by using SSD and HDD storage within the same instance, all via configuration and without having to modify the APIs you use to access the data. This functionality moves older data to lower cost HDD storage (~80% cheaper) while automatically keeping new data on high-performance SSDs.

• Managed autoscaler: We announced the general availability of managed autoscaler and enhanced it to allow scaling read-only replicas independently from the read-write replicas, thereby improving read performance while optimizing costs based on traffic patterns.

• Manual split points: While Spanner automates sharding to distribute work across nodes in an instance, sometimes you know your traffic patterns better than we do. The new split-point API allows you to “warm up” or pre-split your data to handle anticipated spikes, like a flash sale or game launch, to handle traffic immediately.

• Query optimizer: The new default query optimizer (version 8, for those counting at home) brings a host of automated enhancements, optimizing join strategies and index usage to improve query performance and predictability.

• Multiplexed sessions and lazy decoding: We enabled multiplexed sessions by default across major SDKs (Java, Go), significantly improving throughput and resource utilization. We also optimized our client libraries with lazy decoding for partitioned queries, allowing your applications to handle massive datasets with a significantly lower memory footprint. Plus, new built-in client metrics give you granular observability into your application's database performance.

Raising the bar on enterprise safety

For our customers, especially in industries like financial services, retail, and healthcare, reliability is non-negotiable. This year, we introduced additional "safety nets" that further protect your data from the most unpredictable variable in any system: human error.

• Drop protection: Accidental deletions can be catastrophic. We introduced schema object drop protection as a safeguard for your production environments to prevent the accidental deletion of critical tables, indexes, and columns, acting as a line of defense against operational mishaps.

• Zero-touch compliance with default backup schedules: Data protection shouldn't rely on manual checklists. We introduced default backup schedules to ensure a recovery baseline the moment a database is created. This minimizes the risk of forgotten backups and so that your workloads adhere to your organization's governance and compliance standards from day one.

Awards and recognition

While we continue to enhance Spanner based on customer needs and industry trends, it’s always great to get industry recognition for our work. In 2025,

• Spanner was recognized as a Leader in the Gartner Magic Quadrant for Cloud DBMS and ranked #1 for the Lightweight Transactions Use Case (and Top 3 in all categories) in the Critical Capabilities for Operational Databases report for 2025, beating AWS Aurora to take the #2 rank in “OLTP Transactions”

• Beyond technical performance, a March 2025 Forrester Total Economic Impact™ (TEI) study validated the significant business value Spanner delivers, finding that a representative composite organization realized a 132% ROI and $7.74M in total benefits over three years, with a rapid nine-month payback period.

• We were also honored with the 2025 ACM SIGMOD Systems Award for "reimagining relational data management to enable serializability with external consistency at global scale."

Looking ahead

We are proud of what we have delivered for customers in 2025, and are excited to see the innovative solutions you are building on Spanner. Needless to say, we are just getting started and we have a lot more exciting capabilities lined up for 2026. 

Want to learn more about what makes Spanner unique and how it’s being used today? 

Try it yourself for free for 90-days or for as little as $65 USD/month for a production-ready instance that grows with your business without downtime or disruptive re-architecture.