We’ve sat in enough of these conversations to know the pattern by heart. A customer has already picked a model, already has a use case, already has budget approved. Then someone asks where the training data actually lives, and the room goes quiet for a second. It’s on ONTAP, sure. But also in a Snowflake instance nobody fully trusts anymore, a Databricks lakehouse spun up for last year’s project, and a Hadoop cluster the team keeps meaning to shut down and never does. Nine in ten organizations say they’re deploying AI within the next year. Almost none of them have actually sorted this part out first.
NetApp and Starburst have been closing that gap together. Here’s how, and why it’s a different pitch than the usual partner-blog boilerplate.
More storage, same blind spot
Every customer we talk to is buying more storage, not less. Data volumes keep climbing every year regardless of anything either of our companies does. This partnership isn’t trying to shrink anyone’s storage footprint; it’s aimed at something else: nobody can see or query across all of it at once, and that gap shows up in nearly every deal we work on.
Every hop between a storage system and a compute layer costs something, whether latency, dollars, and one more copy of sensitive data that somebody now has to secure and track. An AI model can only ever be as smart as the narrowest slice of data it’s actually allowed to see, and right now that’s whatever silo happens to be closest, which is rarely the one that matters most. Storage built for steady, predictable reporting jobs just wasn’t built for the bursty, unpredictable way training and inference hit it.
Where AI Data Engine actually comes in
NetApp built AI Data Engine to fix exactly this, and Starburst works alongside it as the federation and query layer. Software talking to software. Starburst handles the analytic federation and querying, while AI Data Engine feeds it valuable context from unstructured data. One thing worth clearing up, since people sometimes assume otherwise: there’s no joint hardware appliance here, no single box to rack and stack. NetApp brings the storage and data platform, we bring the federation engine, and the two work together as a named integration rather than a partner logo somebody pasted onto a slide.
What that gives you is one governed SQL layer reaching ONTAP and StorageGRID, plus Snowflake, Databricks, Iceberg and Delta Lake tables, and yes, the older stuff too like Hive, Hadoop, Teradata, all without copying anything out of place first. Starburst’s AI Data Assistant, AIDA, sits on top and does the actual work:
- Planning each query
- Pulling from the right sources
- Citing exactly where an AI agent’s answer came from
Governance rides along with the answer instead of getting checked after the fact.
The numbers we’d lead with
Customers get one integrated platform instead of the five or more tools most AI projects start with, which alone cuts a lot of setup time from a typical build. Data prep costs drop by roughly 70 percent when the work happens in place instead of in a pipeline. Now, federation doesn’t mean a query never touches a network. If the source and the compute aren’t in the same spot, data still has to move to get processed; that part doesn’t go away. What disappears is paying to copy and store that same data a second time just to make it queryable. In hybrid and multi-cloud setups, Starburst’s Stargate capability keeps compute as close to the data as it can and lets clusters share data directly, so you’re not defaulting to one central pull every time. Pull back to Starburst’s full customer base, and the numbers get bigger still. More than 100,000 ETL pipelines retired, analytics costs cut by about half versus a cloud data warehouse, running at a scale that includes 4 of the top 5 global banks.
As NetApp and Starburst put it in a joint post from both companies’ product leadership on the NetApp Community: “NetApp and Starburst are a perfect pairing.”
Governance had to come first
None of the speed matters much if it costs you control over the data. Starburst Enterprise’s single point of governance keeps data access secure while it’s queried in place inside NetApp’s secure perimeter, with row-level and column-level access control, masking, and full audit trails across every source it touches, on top of NetApp’s encryption and compliance posture. For financial services, healthcare, and telecom accounts especially, that’s the difference between a proof of concept that earns a demo and one that actually survives an audit six months later.
What this looks like in practice
A real data problem tells this story better than any slide could. So bring us your data, the messiest version of it–data scattered across silos, an AI project that stalled six months ago because nobody could agree where the data actually lived, a budget stretched thin across too many overlapping platforms. We’d rather hear about it directly than in a follow-up email.
