One of the most persistent headaches in enterprise AI is not the model — it is getting the data to the model without a six-week engineering project. Teradata announced on September 2 that its Autonomous Knowledge Platform now integrates directly with Microsoft OneLake, letting businesses run high-performance AI analytics on their Fabric data without moving, copying, or duplicating anything.

For data teams running both Teradata and Microsoft Fabric environments, that is a meaningful unlock.

What Changed

The integration connects Teradata’s Autonomous Knowledge Platform to OneLake — Microsoft’s unified data lake layer inside Microsoft Fabric — using open Apache Iceberg standards. Instead of extracting data from OneLake, transforming it, and loading it into Teradata (the traditional ETL route), analysts and AI agents can now query OneLake tables directly from Teradata’s environment.

Cross-platform authentication is handled through Microsoft Entra ID, so existing security and governance policies follow the data without being rebuilt. Complex joins, feature engineering for AI models, and SLA-bound analytical queries can now span both platforms without duplicating anything.

The announcement is timed to a broader enterprise push: Teradata will demonstrate the integration at the European Microsoft Fabric and SQL Community Conference in Barcelona, September 28 through October 1.

Why This Matters for Enterprise AI

The practical problem this solves is real. Many enterprises store operational data in Microsoft Fabric because it integrates well with Power BI, Azure services, and Microsoft 365. They also run analytical and AI workloads in Teradata because of its performance on large-scale queries. Historically, any AI project that needed both environments meant someone had to set up a pipeline to move data between them.

Data movement is not a trivial cost. Beyond the engineering time to build and maintain ETL jobs, there is the latency of waiting for syncs, the governance complexity of tracking which copy is authoritative, and the storage cost of duplication. In a world where AI agents need to query live data to make decisions, a pipeline that runs once a night is not good enough.

By treating OneLake tables as native queryable objects through Apache Iceberg’s open format, Teradata sidesteps the movement problem entirely. The data stays in OneLake; the analytics come to it.

What This Means for Business

If your organisation uses Microsoft Fabric for reporting and is exploring AI agents for operational decisions, this integration shortens the path significantly. Your AI agents can now reach the governed, live data in OneLake without a separate data engineering project to feed them.

A few things worth noting for planning:

Governance transfers automatically. Because the integration uses Entra ID and preserves OneLake’s access controls, the same permissions that govern human access also govern agent access. You do not need to rebuild a separate security model for your AI layer.

The Apache Iceberg foundation matters. Iceberg is an open standard, which means this is not a proprietary lock-in play. Other tools and platforms that speak Iceberg can also benefit from the same architecture. Organisations investing in Iceberg now are building on a durable foundation.

Hybrid environments are the real target. This is aimed squarely at enterprises running AI across cloud and on-premises environments. If all your data is already in one place, this is less relevant. But most businesses above a certain scale have data scattered across systems they accumulated over years — and this is the kind of integration that starts to stitch those environments together.

The Bigger Picture

Teradata’s play here is consistent with what the Autonomous Knowledge Platform was designed for: giving AI agents a reliable, governed foundation to operate at scale. The challenge with deploying AI agents in enterprise contexts is not the agents themselves — it is giving them trustworthy access to the data they need without creating new security or governance risks.

Integrating with OneLake is a practical step in that direction. It does not require enterprises to abandon their existing Microsoft investments or consolidate their data estate into a single platform. It adds a path for AI to reach data where it already lives.

For data teams managing this kind of complexity, that is the work: not replacing what works, but connecting it well enough that AI can actually use it.

---

The European Microsoft Fabric and SQL Community Conference runs September 28 through October 1 in Barcelona, where Teradata will show the integration running live. If your team is evaluating enterprise AI data infrastructure, it is worth watching.

If your organisation is navigating the challenge of getting AI agents to work across a fragmented data estate, Enterprise DNA’s Omni Advisory service helps business leaders map a practical path — without overhauling everything at once. Book a discovery call to talk it through.