For decades, the SQL query language has been the cornerstone of database technology.
But what happens when you bring SQL together with modern generative AI? That’s the question that Google Cloud is answering as part of a series of database updates rolling out at the company’s Google Cloud Next conference today.
Over the past year, all Google Cloud databases have added some form of vector support enabling AI use cases. At Google Cloud Next, multiple databases are being updated including Firestore, which is getting MongoDB compatibility. Google Bigtable is getting support for materialized views and support for Oracle Database in Google Cloud is also expanding.
The biggest and most transformational news, at least from a database AI perspective however is in the AlloyDB database. Google first launched AlloyDB in 2022 as an enhanced version of the open source PostgreSQL database. At the Google Next event in the summer of 2024, vector embeddings landed in AlloyDB as well as support for Duet AI to enable database migration.
Today AlloyDB is being expanded with integration with Google’s Agentspace, which is also making its debut at the Google Cloud Next event. Perhaps more interestingly though is the new AI query engine that allows natural language expressions directly within SQL queries for the first time.
AlloyDB’s API query engine brings natural language directly Into SQL
Google’s new AI query engine for AlloyDB, allows developers to use natural language expressions within standard SQL queries—not just replacing SQL, but enhancing it with AI capabilities.
“We’re bringing in an AI query engine to AlloyDB,” Andi Gutmans, GM and VP of Engineering, Databases at Google Cloud told VentureBeat in an exclusive interview. “Within a SQL query we will have operators that both can use natural language and foundation models and your traditional SQL operators And you can bring these together.”
This innovation marks a significant evolution in database interfaces. SQL, an acronym that stands for Structured Query Language, was first introduced in 1973. For the last 50 years it has been the de facto standard for structured database queries. The original promise of SQL was to make it easy to execute database queries with a language that used English words in a natural way. Common SQL queries and actions include terms such as ‘insert’ and ‘join’ but it’s not quite natural language.
“We’re delivering on a 50 year old promise where SQL should mimic English now,” Gutmans said.
The query engine enables developers to combine precise SQL syntax with flexible natural language expressions.
Unlike other approaches that simply translate natural language to SQL, Google’s implementation integrates natural language directly into the query language itself. Google runs foundation model-powered semantic operators alongside traditional relational operators in the database engine.
“When SQL first came out in 1973 it was all about, hey, we want a natural language for query data and so SQL was kind of that natural language,” Gutmans said. “But really, the way you should think about it is now, this is more the promise of SQL, because now you can use even more natural language as part of your SQL query, but it’s still well structured.”
Agentspace integration democratizes database access
Google Cloud is also connecting AlloyDB with its Agentspace platform, creating a natural language interface that extends database access beyond technical specialists to virtually any employee in an organization.
While developers and database administrators benefit from AlloyDB’s AI query engine, regular business users will utilize Agentspace.
“It’s for the average employee in an organization, trying to get their job done,” Gutmans said. “One of the ways to get their job done is actually to have a natural language interface, being able to ask questions about all the enterprise data they have access to.”
What makes this integration particularly powerful is how it maintains security while expanding access. Unlike other natural language database interfaces, Google’s implementation leverages its powerful Agentspace platform, which knows how to reason, not just about a single data source, but multiple data sources. It could be a web search, AlloyDB or other enterprise unstructured data.
Vector search optimization delivers measurable business outcomes
Google has also dramatically improved AlloyDB’s vector search capabilities, optimizing both performance and usability. AlloyDB’s Scalable Nearest Neighbor (ScaNN) index now delivers up to 10x faster filtered vector search queries compared to hierarchical navigable small world (HNSW) indexes in standard PostgreSQL.
“We’ve seen AlloyDB’s vector search adoption increase nearly seven times since the launch of the state-of-the-art Scalable Nearest Neighbor (ScaNN) for AlloyDB index in 2024,” Gutmans said.
This rapid adoption reflects real business impact, as evidenced by retail giant Target’s experience. Gutmans noted that Target has used AlloyDB to improve its online search experience.
“They’re using vector search, and they’re using these capabilities to really improve the accuracy,” he said. “And if you think about the 20% improvement in accuracy that translates to revenue…20% better targeting means more conversions, more revenue.”
Real-time processing capabilities advance with Bigtable’s materialized views
One of the more technically significant announcements is Bigtable’s new continuous materialized views feature, designed for high-throughput, real-time applications.
“This is a really cool capability that is very specific to Bigtable,” Gutmans explained. “Bigtable is used a lot in clickstream counters, like real time counters for real time applications, there’s very low latency, and it scales.”
Unlike traditional materialized views that require periodic refreshes, Bigtable’s implementation updates automatically.
This capability eliminates the need for complex data flow pipelines to calculate aggregates, simplifying architectures for real-time analytics.
What this means for enterprise AI adoption
For enterprises developing AI applications, Google’s database enhancements offer several immediate advantages. The AI query engine enables more intuitive data access while maintaining SQL’s structure and security. The optimized vector search delivers measurable performance improvements for semantic search applications. Finally, the Agentspace integration extends database access throughout organizations without requiring SQL expertise.
For enterprises looking to lead in AI adoption, these innovations mean database infrastructure can now actively participate in AI workflows rather than just storing data. The convergence of SQL’s structure with natural language’s flexibility creates opportunities for smarter applications that leverage both human and machine intelligence without requiring complete system redesigns.
Perhaps most importantly, Google’s approach demonstrates that enterprises don’t need to abandon existing database investments to embrace AI capabilities. As Gutmans succinctly put it when asked if SQL was becoming obsolete: “SQL is dead. Long live SQL.”
The post What’s next for Google Cloud databases? AI inside SQL and more appeared first on Venture Beat.
