Databricks announced Thursday a new funding round valuing the company at $188 billion, led by Coatue, according to TechCrunch AI. The company didn’t disclose the amount and admitted the money isn’t in hand yet. Other outlets peg the raise at roughly $3 billion. The round closes later this summer.
Announcing a round before the wire clears is unusual. A VC told TechCrunch AI the deal is solid, with so many firms wanting in that Databricks had no reason to sit on its new number.
The Valuation Ladder
Four rounds. Nineteen months. Here’s the climb:
- December 2024 – $10B raised at a $62B valuation. Record-breaking at the time.
- September 2025 – $1B at $100B.
- February 2026 – $5B Series L at $134B.
- July 2026 – roughly $3B at $188B.
That’s a tripled valuation in a year and a half. The company has raised so many rounds that the alphabet itself became a punchline. “Turning on alerts for when we get a Series AA,” one person posted.
Why This Isn’t Just Hype
Databricks was founded in 2013, in what TechCrunch AI nicely calls the BC era. Before ChatGPT. It won the big data wave with software that let enterprises dump massive data volumes in the cloud and still run fast analytics.
That’s the whole story of this valuation. Databricks already sat on enterprise data before anyone needed AI on top of it. When companies started demanding AI with the same security and governance they expect from boring enterprise software, Databricks didn’t have to build a new business. It had to ship products onto an existing moat.
So it shipped:
- Lakebase – a database built specifically for AI agents
- Unity – an AI gateway
- Omnigent – a “meta-harness” that manages multiple agents at once
What stands out here is the sequencing. Most AI-halo companies bolt AI onto a product nobody asked to be intelligent. Databricks sold the storage layer first and the intelligence second.
The Cost Story Practitioners Should Read
This is the part that matters more than the valuation.
Databricks became one of the loudest enterprise champions of Chinese open-weight models, meaning models whose underlying code is published for anyone to use and modify. Cost control is driving that shift, and it’s one of the defining trends of 2026. The company is a particular booster of Z.ai’s GLM 5.2 for coding work.
Last week CEO Ali Ghodsi published internal benchmarks from managing AI costs across 3,000 software engineers. The company tested models on the actual tasks its programmers do, not synthetic benchmarks. The finding: “open models, and GLM 5.2 in particular, are now able to handle even the highest level of task difficulty” in coding, at lower total cost than proprietary models from Anthropic and OpenAI.
The surprise was elsewhere. Databricks found the harness mattered just as much as the model. A harness is the agentic coding tool wrapping the model, the thing managing context and instructions. Codex and Claude Code are harnesses. Databricks found the open-source harness Pi among the best at managing context per prompt, making it one of the cheapest options without a quality hit.
“The lesson here isn’t that one harness is always cheaper or that native harnesses are worse,” the post said. “Instead, model choice is only one piece of the puzzle.”
If you’re tracking your own AI spend, that’s the actionable finding. You can swap to a cheaper model and still bleed money on a harness that stuffs context badly.
What To Expect Next
Three things worth watching:
- The AI halo is doing real work on valuations. TechCrunch AI notes the effect is strong enough that sandwich chain Jersey Mike’s mentioned AI 22 times in its S-1. Databricks earned its repositioning. Plenty of companies are just saying the word.
- Open-weight models are now enterprise-viable for hard coding tasks. When a company with 3,000 engineers publishes numbers saying so, procurement teams notice.
- Harness selection becomes a budget line item. Expect more benchmarking of tooling, not just models, through the rest of 2026.
The round closes later this summer. Whether Databricks stays private long enough to reach Series AA is the open question.
Full details are at the original source.