Teradata aims to simplify on-premises AI for data scientists with AI Factory

Running AI workloads on premises has often meant overloading shared infrastructure or relying on complex hybrid setups. In an effort to reduce this friction, Teradata has bundled some of its existing tools into a new AI Factory to help data scientists manage the full AI lifecycle on premises without disrupting core systems or moving to the cloud.

“A lot of IT departments are already fully using their Teradata system for mission-critical workstreams and really don’t want data scientists doing AI exploratory work on that same system,” said Dan Spurling, SVP of product management at Teradata.

Until now, he said, to keep the two apart enterprises would have to use a hybrid solution of Teradata’s existing on-prem hardware and VantageCloud. “But AI Factory simplifies this process,” he said, separating the infrastructures where needed and making it possible to do all of the AI workstream, or at least the most expensive part of it, on premises.

This can give organizations “predicable costs and more control over their data sovereignty,” Spurling said.

AI Factory, which wasn’t previously available on-premises in a single, integrated, and ready-to-run stack, combines the data analytics software provider’s database engine and AI Workbench (consisting of ClearScape Analytics, ModelOps, JupyterHub, AI and ML In-Engine Analytics, Enterprise Vector Store, and support for the Open Table format) with AI microservices such as RAG, retrieval embedding, reranking, and guardrails.

Together, these tools gives enterprises a way to scale AI with an out-of-the-box packaged solution — with governance, performance, and control baked in, said Michael Ni, principal analyst at Constellation Research.

“What stands out is how Teradata integrated ModelOps, vector search, and LLM pipelines into a turnkey platform providing traceability and compliance from experimentation to production,” Ni added.

AI Factory is also developer- friendly, said Moor Insights and Strategy principal analyst Robert Kramer.

“The built-in AI Workbench includes tools such as JupyterHub and ModelOps, which are familiar to data teams and help keep the development process moving. That’s important for customers who want to keep AI projects from getting stuck in setup or compliance bottlenecks,” Kramer said. The predictable pricing model should help customers avoid surprise costs, he said.

Regulatory pressure

Enterprises’ desire to move AI activities on-premises isn’t just driven by questions of cost or control: For some businesses, it’s also about regulatory norms and scrutiny.

“Between stricter data privacy rules, growing concerns over data sovereignty, and unpredictable cloud costs, many enterprises are rethinking what should stay in the cloud and what’s better off staying in-house,” Kramer said. “This is especially true for industries like healthcare, finance, and government, where regulations are tight and data sensitivity is high.”

Other vendors including IBM, Dell, and HPE are looking to help enterprises build AI on-premises, but Kramer said Teradata has an edge, especially with customers who are already using it for data management and want to bring AI into the mix without standing up a bunch of new systems or re-architecting what they already have.

“While others might have gotten there first, Teradata’s offering is built around real integration with enterprise data operations, which can save time and reduce complexity for its existing customers,” he said.

Integration with Nvidia

For AI Factory, Teradata has partnered with Nvidia by using its Nvidia Enterprise AI Factory Validated Design, a full-stack featuring its Blackwell-accelerated computing systems, AI Enterprise software, and networking hardware.

The Nvidia GPUs, though, in this case, have to be owned by the enterprise deploying AI Factory on-premise, Teradata said.

As Constellation Research’s Ni put it, “Nvidia brings the AI muscle, Teradata brings enterprise-grade data and analytics platform.”

Having their Nvidia chips on premises, Kramer said, enables enterprises to run heavier AI workloads such as training or inferencing models without relying on cloud-based GPUs, which can lead to high or unpredictable costs. “To avoid that, what Teradata does is connects to those GPUs using its AI Microservices layer, which helps customers get more out of what they’ve already invested in,” he said.

AI Factory is now generally available.