Databricks announced a new funding round at a $188 billion valuation, led by Coatue, continuing a rapid fundraising streak that began well before the AI boom. The company has raised multiple massive rounds in quick succession—$10B at $62B valuation in December 2024, $1B at $100B in September 2025, and $5B at $134B in February 2026—spawning memes about running out of alphabet letters for series names. The tension is clear: Databricks was originally a big-data analytics platform, but it has aggressively repositioned itself as an AI infrastructure provider, leveraging its existing enterprise data storage and governance capabilities to build AI products like Lakebase, Unity, and Omnigent. The article questions whether this valuation explosion is deserved or merely riding the AI hype wave, especially given that the company announced the round before the money was even secured (a rare move, but one the VC community sees as solid due to oversubscription).

Internally, Databricks has been practicing what it preaches. CEO Ali Ghodsi shared results from benchmarking AI models on real coding tasks performed by the company’s 3,000 software engineers. The comparison found that open models—specifically Z.ai’s GLM 5.2—matched proprietary models from Anthropic and OpenAI at handling the highest levels of coding difficulty, while costing significantly less. More surprisingly, the choice of harness (the agentic coding tool like Codex or Claude Code that wraps around a model and manages context) had an equally large impact on total cost. The open-source harness Pi emerged as one of the best at managing context per prompt, delivering low cost without sacrificing quality. The lesson: model selection alone is insufficient; the infrastructure layer that orchestrates the model matters just as much for practical AI deployment.

The takeaway for builders is that the AI stack is no longer just about which foundation model you pick. Databricks’ internal data shows that optimization must extend to the harness layer—the orchestration framework that controls context windows, instructions, and multi-turn interactions. The company’s aggressive adoption of cheaper open-weight Chinese models like GLM 5.2 is a strategic bet that aligns with a broader 2026 trend of cost control through open models. For serious engineers, this reinforces the need to treat AI deployment as a full system design problem: model, harness, data governance, and cost monitoring are all levers. Databricks’ valuation surge is as much about its ability to capture this integrated stack as about the AI tailwind itself.