AI Platform Giant Databricks Hit With Patent Suit in Texas—Another East District Gambit

The Filing

Poulin Holdings LLC filed suit against Databricks Inc. on March 10, 2026, in the US District Court for the Eastern District of Texas, targeting several widely used components of the AI platform provider's Lakehouse architecture. The complaint alleges that Databricks products including Feature Store, MLflow tracking and evaluation, Unity Catalog, and Marketplace infringe US Patent Nos. 8,160,977 and 9,009,090.

The patents at issue cover predictive-model databases, search functions, and access controls—bread-and-butter technology that sits at the heart of modern machine learning operations. Poulin seeks at least a reasonable royalty for infringement, a formulaic demand that signals this is likely a straightforward licensing shakedown rather than a dispute over genuine innovation boundaries.

What Was Actually Decided

Nothing yet. But the venue choice tells the story. The case was filed in the Texas Eastern District Court, the perennial favourite of patent assertion entities seeking plaintiff-friendly juries and expedited discovery schedules. This is forum shopping in its most transparent form—Databricks has no meaningful presence in Marshall or Tyler, Texas.

The timing is equally suspect. Databricks has built its reputation among AI companies by helping businesses optimise available features, apps, and autonomous agents, making it a natural target as enterprise AI adoption accelerates. Wait for the company to achieve market prominence, then strike with patents covering foundational infrastructure.

The Practical Impact

This lawsuit represents a textbook example of the patent tax that AI platform companies now face. The targeted Databricks tools—Feature Store for managing ML features, MLflow for experiment tracking, Unity Catalog for data governance—are precisely the kind of workflow infrastructure that any serious machine learning deployment requires.

For Databricks, the immediate concern isn't damages but distraction. Patent litigation in the Eastern District of Texas moves fast, with magistrate judges who pride themselves on keeping discovery on tight schedules. The company will need to allocate significant engineering resources to document how its systems work, potentially slowing product development.

But the broader signal here is troubling. If patent holders can successfully target the scaffolding that makes AI development practical—the cataloguing, versioning, and governance tools that let companies actually deploy models at scale—they're effectively levying a tax on AI adoption itself.

Historical Parallel

This mirrors the software patent wars of the 2000s, when assertion entities targeted basic e-commerce and web functionality. Companies like Amazon, eBay, and Google faced waves of suits over one-click purchasing, online auctions, and search algorithms. The difference now is the stakes. AI infrastructure patents cover technologies that didn't exist when most of these applications were filed, creating a dangerous mismatch between narrow patent claims and broad technological impact.

The Eastern District of Texas has enabled this dynamic by refusing to apply meaningful venue restrictions. The Federal Circuit has previously rejected efforts by companies like Lenovo and other defendants to move similar lawsuits out of the Eastern District of Texas, cementing the district's role as the forum of choice for patent monetisation.

What's Next

Databricks will likely file a motion to transfer venue, arguing that California's Northern District—where the company is based and where its employees and documents are located—is more convenient. Don't expect success. The Eastern District of Texas has developed an entire cottage industry around rejecting such motions.

More likely, this settles quietly within 12-18 months for an undisclosed sum. Patent assertion entities like Poulin Holdings don't typically want the expense and uncertainty of trial. They want licensing revenue, and they've calculated that the cost of settlement will be lower than the cost of defence.

The next case management conference is scheduled within 60 days, where the court will set the discovery schedule and claim construction timeline. For Databricks, the real work begins now: mapping every aspect of how MLflow handles model metadata, how Feature Store manages data lineage, how Unity Catalog implements access controls.

The patent system wasn't designed for this, but it's what we have.