Google Cloud expands Spanner for AI & multi-model data

Google Cloud has expanded Spanner with a broader multi-model database strategy and made those capabilities available through Spanner Omni, placing the database more directly at the centre of Google's data and artificial intelligence portfolio.

Spanner now combines relational, graph, vector, key-value and full-text search in one database. Spanner Omni extends that model to containerised deployments on Kubernetes across on-premises systems and other public clouds.

The update reflects a wider shift in how large technology suppliers are positioning databases for generative AI and autonomous software agents. Rather than treating operational data stores solely as systems of record, vendors are trying to make them the point where transactional data, search, relationships and semantic context can be queried together.

Spanner's graph model is built around Spanner Graph, which lets users represent data natively as a graph or layer graph structures over relational data. Google is targeting knowledge graph use cases in which software agents need to connect entities, roles and events across large datasets.

Vector search is also part of the package, with support for both K-Nearest Neighbours and Approximate Nearest Neighbour methods. Google says the system can support indexes with more than 10 billion vectors, allowing semantic search to sit alongside transactional and relational workloads.

Full-text search is integrated into the same platform, supporting retrieval across structured and unstructured data, including query expansion for synonym matching and spell correction.

Another element is a built-in columnar engine for analytical queries on live operational data. Google says that can run some analytical workloads up to 200 times faster without moving data into separate systems, addressing a longstanding issue for businesses that maintain different stores for transaction processing and analytics.

Customer examples

Google cited Palo Alto Networks as a user of Spanner Graph for access-control workloads, saying it avoids the need for a separate graph database. It also pointed to legal intelligence company Inspira, which consolidated a 4.5 TB data pipeline into a single data store using full-text and vector search for retrieval-augmented legal analysis.

Google also highlighted fraud prevention specialist Verisoul as a user of the columnar engine for analytics on high-volume transactional data. In that case, the system avoids the need for separate copies of data and reduces replication delays, according to Google.

Interoperability is a notable part of the latest push. Developers can combine graph traversal, vector similarity, relational logic and keyword search in a single SQL query rather than stitching together separate database engines and moving data between them, Google says.

Omni rollout

Spanner Omni is intended to widen the reach of that model beyond Google's own infrastructure. According to Google, it is a downloadable, containerised version of Spanner that runs on Kubernetes and does not require dedicated hardware.

That gives customers a way to run the database on-premises, at the edge or in rival public clouds including AWS and Microsoft Azure while keeping the same underlying database model. The approach mirrors similar moves across the cloud market as suppliers try to meet demand for hybrid and multi-cloud deployments rather than insisting workloads remain in a single provider's environment.

For Google, the strategy also addresses concerns about database fragmentation. Many businesses have accumulated separate relational, search, graph and vector systems over time, particularly as AI projects have grown. The cost and operational burden of maintaining those systems has become a stronger talking point as companies look to put AI applications into production.

Market position

Google pointed to external benchmarks and economic studies to support the case for Spanner's broader role. It said Gartner ranked Google's Spanner first in the Lightweight Transactions use case in the 2025 Critical Capabilities for Cloud Database Management Systems for Operational Use Cases report, the second year in a row it has held that position.

Google also referred to a Forrester Consulting Total Economic Impact study commissioned by Google Cloud, which found that a composite organisation achieved a 132% return on investment, a payback period of nine months and total benefits of USD $7.74 million over three years after deploying Spanner.

Those figures come as cloud providers compete more aggressively to link core database products with AI infrastructure. Google is positioning Spanner as a foundation for its broader Agentic Data Cloud strategy, alongside services such as BigQuery and its Gemini tools, with the aim of more closely combining live operational data and large-scale analytical data.

The underlying technical message is that a single database should support multiple data models and serve both operational and AI-driven workloads. In Google's view, that makes the database less a passive storage layer and more a central point for context, search and decision-making across applications.

Google also said Spanner's architecture relies on technologies including TrueTime, Paxos, dynamic resharding and its ScaNN vector indexing method, which it says distinguish it from products that add new data models through separate engines. The result, according to Google, is a system in which distributed consistency, analytical querying and semantic retrieval run within one database architecture.

Spanner has also received the SIGMOD Systems Award, according to Google, adding to the product's profile as competition intensifies among cloud database vendors over which platform will underpin the next wave of AI software.