Databricks and Microsoft to the 2030s: keep an exit

On July 23, Microsoft and Databricks announced they are extending their partnership into the 2030s. More Azure integration, more shared surface area, more of your data and your AI governance living inside one vendor relationship. The product is good. The lock-in is the point, and you should price it before you sign anything.

What actually happened

The two companies have been partners for close to a decade — Azure Databricks has been a Microsoft first-party service since 2018. The expansion announced this week does three things:

• Infrastructure. Databricks runs on Azure Cobalt, Microsoft's Arm-based silicon. It uses Cobalt 100 today and plans to adopt Cobalt 200, which Microsoft says delivers up to 50% better performance with memory encryption on by default.
• Integration depth. The named surface list is long: Microsoft Entra, Azure Data Lake Storage, OneLake, Power BI, Purview, Microsoft Foundry, Power Platform, Microsoft 365, Teams, and Copilot.
• Databricks capabilities pushed into Microsoft workflows. Genie, Databricks' AI co-worker, plus Genie Ontology and the Unity AI Gateway for governing models, agents, and cost.

Databricks says more than 20,000 organizations use its platform, including 70% of the Fortune 500, with Electrolux, Unilever, SMBC, and Banco Bradesco named as Azure Databricks customers. No financial terms were disclosed — "into the 2030s" is the only duration either company put in writing.

Why it matters for your business

Most small businesses do not run Databricks. Read this as a weather report anyway, because the pattern is the one you will meet at your own scale: your data platform and your cloud are being welded together, and the weld is sold to you as convenience.

It genuinely is convenient. Single sign-on through Entra, governance through Purview, dashboards in Power BI, an AI agent that already knows your schema — that is a real reduction in integration work, and integration work is where small teams die. We are not going to pretend otherwise.

The cost shows up later, and it shows up as an inability to leave. When your identity provider, your object storage, your BI layer, your governance catalog, and your AI gateway are all one vendor's, the migration estimate for any single piece includes all the others. That is not a hypothetical; it is why "we'll switch next year" turns into a five-year renewal.

The move we'd make at any scale: keep your raw data in open formats in storage you control, and treat every platform above it as replaceable. Parquet or Iceberg in a bucket, not a proprietary table format you can only read from inside one product. Keep your transformation logic in version-controlled SQL or code, not in a vendor's visual builder. Write down, once a year, the actual steps to move your warehouse somewhere else. If nobody can write that document, you do not have a vendor — you have a landlord.

Sources: Microsoft Source — Databricks and Microsoft expand partnership, PR Newswire — joint announcement.