There is a version of the AI investment thesis that is entirely about model companies — OpenAI, Anthropic, the next generation. There is another version that says the real money is in infrastructure, and that the infrastructure companies are, historically, the ones that compound over decades while the model companies cycle through.

Databricks is the infrastructure thesis. It does not build AI models. It builds the data platform that AI models run on — the unified analytics and AI environment that enterprise companies use to store, process, and deploy AI at scale. Its lakehouse architecture has become, for many large enterprises, the data foundation on which their AI strategy sits.

The numbers are remarkable for a company that gets less attention than its valuation deserves. At $134 billion private, with $5.4 billion in ARR growing at 65%, and positive free cash flow, Databricks is the kind of company that would have been the most anticipated IPO of any year that did not also contain OpenAI and Anthropic. Instead, it is the sleeper.

The S-1 is expected in H2 2026. Ali Ghodsi, the CEO, has been saying Databricks is "going public six months at a time" for years, which is the CEO equivalent of I'll get around to it. This time, the company appears to be actually getting around to it. The investment banks are in the room.

The FreeMalta read: Cleanest IPO of the cohort if it lists in 2026. Profitable, growing, mission-critical. The Snowflake comparison will hang over pricing — Snowflake traded badly after its 2020 IPO — but Databricks' unit economics are better. Watch H2 2026.