Getty Images/iStockphoto

Teradata unveils vector store to fuel AI development

The longtime data management and analytics vendor's new feature will enable developers to discover the relevant data needed to train business-specific generative AI applications.

Teradata on Monday unveiled Enterprise Vector Store, a feature that will enable customers to process high volumes of vectors nearly instantaneously to fuel AI development.

Vectors are numerical representations of data assigned by algorithms. They are a key means of giving structure to unstructured data, such as text, so it can be searched and discovered. In addition, vectors enable similarity searches, so large amounts of related data can be discovered.

Vector search and storage have become critical as interest in AI development has increased, given that AI tools require large volumes of high-quality data to deliver accurate outputs.

By adding Enterprise Vector Store, now in private preview with general availability expected in July, Teradata addresses a rising need in a way that makes adding vector capabilities simple and cost effective, according to Steven Dickens, CEO and principal analyst of HyperFrame Research.

There are vendors that are positioning vector databases as a standalone. What Teradata is doing largely negates that need.
Steven DickensFounder and principal analyst, HyperFrame Research

When vector search and storage capabilities are not part of a data management platform such as Teradata, users need to add third-party tools, which can be complex and costly.

"We're seeing a reduction in the need for a separate vector database," Dickens said. "There are vendors that are positioning vector databases as a standalone. What Teradata is doing largely negates that need. What that means for enterprises is less data movement, better security and reduced operations."

Among others, Oracle and AWS have similarly added vector search and storage capabilities to their databases.

"You can get excited about the feature functions of a standalone vector store, but the value comes from operational security and cost," Dickens said.

Adding vector store

With generative AI's potential to make workers better informed and more efficient, many enterprises are investing heavily in AI development.

In response, with proprietary data needed to train AI to understand an enterprise's operations, many data management and analytics vendors have built environments that enable customers to use their data with language models to develop AI applications.

Critical to those development environments have been vector search and storage that enable developers to discover relevant data for training AI.

Among competitors, Databricks made vector search and storage generally available in May 2024. SingleStore did so in late 2023. Now, Teradata is doing the same, spurred by the rising interest in generative AI development, according to Louis Landry, the vendor's chief technology officer.

Teradata already enables AI development, such as machine learning. Adding capabilities that support generative AI development was a logical next step.

"The market started to explode around this stuff," Landry said. "Vectors are the next data type that needs to be thought about in an enterprise data landscape, and we wanted to be part of that."

In addition, user feedback played a part, he continued, noting that customers related that they had tried specialized databases that hadn't held up to the workload demands of large enterprises and needed help.

"That really lit a fire," Landry said.

Using Enterprise Vector Store, Teradata customers can combine structured and unstructured data to feed AI development pipelines while taking advantage of existing Teradata features such as data governance.

In addition, Enterprise Vector Store includes the following:

  • Integration with the Nvidia NIM AI development platform to boost the performance, scalability and accuracy of AI tools.
  • Management of the full vectorized data lifecycle, from generation and indexing through intelligent search.
  • Workload processing within the existing Teradata environment, which enables flexible deployment options including cloud, on-premises and hybrid.
  • Support for development frameworks such as LangChain.
  • Features aimed at ensuring outputs can be trusted, such as tracking and explaining changes to data over time.

While none of the features differentiate Teradata's Enterprise Vector Store from the capabilities provided by specialists such as Pinecone and Chroma DB, or those offered by competing data platform vendors, the new capabilities are still significant, according to Dickens.

A specialized database might have more features than Enterprise Vector Store. Those extra features, however, are not must-haves. Enterprise Vector Store includes the needed capabilities and is therefore a far more attractive option for Teradata customers than adding a vector database from a third party.

"Vectors are obviously important but turning that feature function on in what you've already got rather than having to replicate that function with a standalone platform removes complexity," Dickens said. "Removing complexity saves money, saves time and reduces your security exposure."

As a result, any reasons Teradata customers may have had to migrate to another data platform are gone, Dickens continued. Teradata modernized its cloud-based data management and analytics capabilities in 2022. Now, it's adding capabilities that enable AI development.

Next, it might add new customers if Teradata can improve its messaging and work more seamlessly with hyperscale clouds such as AWS, Google and Microsoft.

"Teradata has a huge installed base of some premium names, and if they can add some names to that through some of the new capabilities -- maybe in the midmarket enterprise -- that's going to fuel growth," Dickens said. "The feature functions are there with things like vector store, and they have a cloud delivery model."

A chart shows the differences between traditional search and vector search.

Next steps

With Enterprise Vector Store in preview, enabling AI development will continue to be part of Teradata's product development plans, according to Landry. In particular, helping customers build applications they can trust is a focus.

In addition, enabling customers to automate business processes using AI agents is part of Teradata's roadmap, Landry continued.

"Teradata's play in all of this isn't to necessarily be on the bleeding edge but to bring real business value as technology matures," he said.

Dickens, meanwhile, suggested that Teradata would be wise to focus on making its platform more integrated with hyperscale clouds through partnerships that make Teradata a preferred platform.

SAP and Databricks recently unveiled such a partnership. Likewise, Oracle is aligning with other hyperscale clouds.

"Yes, you can consume VantageCloud [on other clouds], but is it truly a first-party service?" Dickens said. "If you're an AWS shop and can see that Teradata works with AWS, that's great. It's that type of simple flow that Teradata needs to do a better job of."

Eric Avidon is a senior news writer for Informa TechTarget and a journalist with more than 25 years of experience. He covers analytics and data management.

Dig Deeper on Data integration