Analysis
Databricks is in private preview with an agentic CDP; Zeotap runs its CDP on the customer's own open-source stack. Both avoid copying customer data into a separate system.
In 60 seconds
- June 16, 2026: Databricks announced CustomerLake, an agentic customer data platform inside its lakehouse, governed by Unity Catalog. It is in private preview, with consumption pricing planned.
- September 15, 2026: Zeotap launched Open CDP, running on Apache Spark and Iceberg in the customer's own lakehouse.
- Both products keep customer data in the customer's environment.
- Vendors use "composable" and "sovereign" in different ways.
Impact Layer
- What changed
- A major data platform vendor and a CDP vendor each launched a CDP that runs on the customer's own lakehouse instead of a separate copy of the data.
- Why it matters
- It changes where a CDP lives: from a separate system with its own data copy to a layer on the warehouse or lakehouse the company already governs.
- Who is affected
- Marketing operations and data teams at larger companies evaluating or running CDPs, and standalone CDP vendors.
- Martech stack area
- CDP, AI, Data
- Stack impact
- Customer data can stay under existing governance, but marketing teams will depend more on the data team for setup and changes.
- Business impact
- Moderate CustomerLake is in private preview and pricing is not final.
- Related technologies
- Related vendors
- What to watch
- CustomerLake general availability and pricing, whether other lakehouse vendors respond, and what "open" and "sovereign" mean in contracts.
A customer data platform (CDP) has usually meant a separate system that collects customer data, unifies it and passes it to marketing tools. Two 2026 launches argue for a different arrangement: keep the data in the company’s own lakehouse and run the CDP on top of it.
Databricks CustomerLake
On June 16, 2026, Databricks announced CustomerLake, which it calls an agentic CDP inside its lakehouse and which runs under Unity Catalog, Databricks’ governance layer. It is in private preview. Customers named in the announcement include HP, Circle K, AB InBev and Getnet by Santander. Databricks plans consumption-based pricing. Ali Ghodsi, Databricks’ chief executive, is quoted.
Zeotap Open CDP
On September 15, 2026, Zeotap launched Open CDP, which runs on open-source Apache Spark and the Iceberg table format, in the customer’s own lakehouse. Zeotap describes it as composable and “sovereign by design”, and says it is available to qualified enterprise customers.
What “composable” means here
Both vendors keep customer data where the company already holds it, rather than copying it into a vendor’s system. The word “composable” is used loosely across the industry, so it is worth asking exactly which parts of a CDP (identity resolution, audience building, activation) run in your environment and which run in the vendor’s.
Our read
For governance-minded companies, a CDP that does not copy data is easier to defend: fewer copies means fewer places to apply consent and deletion rules. The cost is operational: marketers lean on data engineers for changes they might previously have made in a visual interface. Both launches are early, with one still in private preview, so independent customer results matter more than the pitch. Compare them with the incumbent CDP approach (see our CDP page) on setup effort, not only on features.
Sources
- Databricks: Databricks enters the marketing industry with CustomerLake (opens in a new tab)
Private preview, customers named, Unity Catalog, pricing plan.
- GlobeNewswire: Zeotap launches Open CDP (opens in a new tab)
Apache Spark and Iceberg, availability.