Databricks hits $188B valuation as its AI pivot pays off — and its coding benchmarks turn heads

Databricks has done it again. On Thursday, the company announced a new funding round that values it at $188 billion — the latest in a dizzying series of raises that have transformed it from a big-data SaaS darling into one of the most talked-about AI infrastructure companies in the world.

The round, led by Coatue, is not yet closed, but it's said to be worth roughly $3 billion. That follows a $5 billion Series L in February at a $134 billion valuation, a $1 billion raise at $100 billion in September 2025, and a $10 billion round at $62 billion in December 2024. The pace has been so relentless that tech Twitter has already started making memes about running out of alphabet letters for Series rounds.

But the valuation leap isn't just hype. Databricks has successfully repositioned itself as an AI company by leveraging the one thing it has always had: enterprise data. Its platform already sat on troves of corporate data, making it a natural home for secure, governed AI deployments. The company has since rolled out a suite of AI products — Lakebase for AI agents, Unity as an AI gateway, and Omnigent for multi-agent orchestration — that have cemented its credibility.

Perhaps most interesting for developers and AI engineering teams is Databricks' growing role as a champion of open-weight models. The company has become a high-profile advocate for cost-conscious AI, particularly Chinese open-weight models like Z.ai's GLM 5.2. Last week, CEO Ali Ghodsi published internal benchmark results comparing AI models on real coding tasks performed by Databricks' own 3,000 software engineers.

The findings were striking: open models, and GLM 5.2 in particular, matched or exceeded proprietary models from Anthropic and OpenAI on the hardest coding tasks — at lower cost. But the real surprise was that the choice of agentic coding harness — the tool that wraps around a model to manage context and instructions — had just as much impact on cost as the model itself. Databricks found that the open-source harness Pi excelled at managing context, delivering high quality at a fraction of the cost.

"The lesson here isn't that one harness is always cheaper or that native harnesses are worse," the blog post noted. "Instead, model choice is only one piece of the puzzle."

This kind of pragmatic, cost-driven research is exactly what enterprise engineering teams need to hear. Databricks isn't just selling the AI dream; it's doing the messy work of figuring out what actually works in production, at scale, and sharing those lessons. That's a big reason why investors keep opening their checkbooks.