Google Cloud expands gen AI power for database and data analytics tools

Google Cloud is expanding the capabilities of its database and data analytics offerings with a series of updates announced today at the Google Cloud Next event in Tokyo. 

The announcements span across multiple services including the Spanner and Bigtable databases as well as the BigQuery data analytics and Looker business intelligence platforms. The overall goal is to integrate more flexibility into how data can be used and accessed, in an effort to help further accelerate generative AI deployments and adoption.

Key announcements and update from Google include:

• Spanner gets new graph and vector data support
• Bigtable adding SQL support
• Gemini AI is being integrated into BigQuery and Looker

“Organizations recognize that in order to get to incredible AI, they need to have incredible data,” Gerrit Kazmaier, GM & VP of Data Analytics at Google Cloud said during a briefing with press and analysts.

Google’s data analytics platforms get a new ‘look’ with gen AI

For data analytics, the big news is that Google’s Gemini AI capabilities are now available in BigQuery and Looker.  

The integration of Gemini provides a long list of over 20 new features including code generation, explanation and intelligent recommenders that will help data analysts be more productive.  Inside of BigQuery, Gemini will now also help to power advanced data preparation and analysis to accelerate time to value from data. 

“Data is messy,” Kazmaier said. “One of the great benefits that we saw in building our specialized gen AI models is for actually reasoning about data and helping our customers to align and govern data much quicker.”

AI will also help to inform the new Data Canvas feature which Katzmaier described as, “…the perfect synergy between user experience AI and a data analyst.” The key advantage of Data Canvas lies in its interactive and AI-assisted approach. It creates a self-reinforcing dynamic where users incrementally build their analysis path, and the system learns from this process.

For Looker the AI updates have a focus on helping to make it easier to get at business intelligence insights.

“We have focused our innovation on Looker on building customized agents who are really deep AI experts, which know how to select data, perform analysis and summarize it,” Katzmaier said.

Spanner database become even more multi-modal with vector and graph

Though the Google Spanner database might not be familiar to everyone, it is in fact a technology that is used by almost everyone that uses Google.

“Spanner is powering most of Google’s if not all of Google’s user products, whether that is Search, Gmail, YouTube and we had to build Spanner to really meet the level of scalability and availability that Google needed,” Andi Gutmans said. “One of the exciting things about my job is I get the opportunity to externalize that innovation to our enterprise customers.”

One of the new innovations that Google is bringing to its enterprise customers is Graph database capabilities for Spanner. Graph provides a different way of making connections across data that can enable nuanced semantic relationships.

Not only is Spanner getting graph support, it’s also finally getting vector support as well.  Google had previously announced a preview of vector support in Spanner back in February. Both vector and graph are useful at helping to enable gen AI applications. Vector in particular is commonly associated with Retrieval Augmented Generation (RAG).

While there are many purpose-built native graph and vector databases in the market, Google’s approach is to provide a multi-modal database.

“It’s not that customers have to move their data to get graph capabilities. they can take their enterprise data and start to build the graph capabilities on top of that,” Gutmans said.

The basic idea is that organizations are already relying on Spanner and trust it. The addition of graph and vector enable those organizations to extract even more utility from that data.

“We’ve expanded Spanner now, from being primarily a relational database to really being a true multi-modal database,” Gutmans said.