Databricks

Databricks, Inc. grew from the open-source Spark project into a commercial data platform. Customers land files in object storage and run analytics and machine-learning jobs through Databricks, often on AWS, Azure, or Google Cloud.

The company sits in a stack where cloud providers also sell competing warehouses and model tools. That makes Databricks both a partner and a rival of Amazon, Microsoft, and Google. Preferential cloud terms, or a cloud that bundles its own warehouse below cost, are the competition theories that cut both ways.

Open formats such as Delta Lake are part of the sales pitch: data should not be trapped. In practice, notebooks, job orchestration, governance, and model serving still live in the commercial product. Leaving means rewriting pipelines even if the files themselves are portable.

Databricks has raised large private rounds and bought AI tooling. A public listing or a sale to a cloud provider would be a merger review about whether an independent data layer disappears into one hyperscaler.

Cloud providers can bundle analytics with infrastructure credits. An independent platform claims that is a tie or a margin squeeze. The cloud claims it is competition.

Format openness reduces one lock-in path and leaves others: governance catalogs, identity, and scheduled jobs.

Exclusive marketplace placement on a single cloud can steer new workloads away from multi-cloud rivals.

Model-serving tied to the same lakehouse can extend data dominance into inference, which is the layer customers thought they could shop separately.

End consumers do not see Databricks. They see the apps a bank or retailer builds on it. A locked-in data platform shows up later as slower product changes and higher IT bills.

Data teams that standardize on one lakehouse train people who are less useful to a rival stack. That labor specialization is a switching cost.