A founder said "we just need $1 billion." Investors showed up with $15 billion instead.
The easy read here is "another AI bubble story." But dig into the real reason, and it's actually about companies getting spooked by their AI bills.
Everyone assumes it's just another AI bubble
Here's what happened. Databricks originally planned to raise about $1 billion. Then, during the company's June conference, The Information published a story saying Databricks was doing a big fundraise — and everything flipped from there.
CEO Ali Ghodsi told TechCrunch:
"As soon as that article went out, there was a long line of investors that started calling. My phone blew up."
— Ali Ghodsi, Databricks CEO
Before the round had even formally launched, $15 billion of interest materialized from just a select group of investors they'd talked to. The company ended up taking $5 billion at a $190 billion valuation. Coatue led, with Blackstone, MGX, T. Rowe Price affiliates, Sixth Street Growth, and roughly two dozen other firms joining in.
On the surface, this reads like the classic "money's everywhere so it chases anything AI" bubble narrative. But the numbers say otherwise. Databricks is running at a $7 billion revenue run-rate, growing more than 80% year over year, and it's already cash-flow positive. Its core data warehouse product alone does $1.5 billion in revenue at 100% growth, with over 1,000 customers spending more than $1 million a year.
So this isn't a "hope and hype" valuation — there's real paying revenue behind it. Which means the question changes: what did investors actually buy?
What they really bought was "savings"
Ghodsi summed up the backdrop in one line: "AI is expensive." Coming from a company with multibillion-dollar infrastructure deals across three hyperscalers and a 100-person research team, that's not just a throwaway comment.
And this cost story isn't just about Databricks' own spending. In a separate announcement a month earlier — the round that valued the company at $188 billion — the company pointed to a shift in how enterprises are actually using AI.
"Enterprises are moving from tokenmaxxing to valuemaxxing. They don't want to burn expensive tokens on the smartest model."
— Databricks, official announcement
In plain terms: the old habit was throwing your priciest, smartest model at every task ("tokenmaxxing"). The new move is routing tasks to whatever model actually fits the job, squeezing more value out of the same budget ("valuemaxxing"). The catch is, once an org is running multiple models and vendors, nobody can manage that by hand anymore.
That's exactly the gap Unity AI Gateway is built for. For organizations running multiple AI models at once, it handles per-model access controls, per-model cost monitoring, and security policy enforcement across mixed-model environments in a single layer. Instead of CIOs and IT ops teams managing dozens of AI integrations one by one, they get centralized oversight.
Alongside that sit Genie (an AI colleague that gives reliable, business-context-grounded answers) and Lakebase (a serverless Postgres database that agents read and write to autonomously, already past a $100 million revenue run-rate). Ghodsi put it this way: "Enterprises don't just want an AI that talks — they want an agent that remembers context, delivers accurate answers, and stays on budget."
Bottom line: the real reason $15 billion lined up wasn't "AI is exciting." It was that demand for infrastructure to control AI costs had become real and urgent.
Why the valuation tripled in 20 months
The numbers show just how steep this climb has been. The valuation was $62 billion at the Series J round in December 2024 — and it's now at $190 billion.
| When | Round | Valuation |
|---|---|---|
| Dec 2024 | Series J ($10B) | $62B |
| Dec 2025 | Strategic round (~$4B+) | $134B |
| Aug 2026 | Strategic round ($5B) | $190B |
The valuation tripled in 20 months, and it's now trading at roughly a 27x run-rate multiple. Snowflake is Databricks' closest publicly traded rival in enterprise data platforms, and analysts are reading this round as direct competitive pressure on Snowflake.
Coatue's Thomas Laffont, who led the round, put it this way: "Databricks has spent a decade being early to where AI was headed. Now it's the infrastructure the industry builds and scales on." Microsoft also extended its partnership into the 2030s, expanding its use of Azure Cobalt.
Why does the company keep needing more capital? Multibillion-dollar infrastructure deals with the major clouds, a 100-person AI research team, and an active M&A pace. Most recently, it acquired ElectricSQL, a lightweight Postgres team, to strengthen Lakebase. Ghodsi himself put it plainly: "We do a lot of M&A."
When this much capital chases the problem of controlling AI costs, it's a decent signal that your own org's AI spend is probably headed down the same track.
| Leave AI spend alone | Audit it now | |
|---|---|---|
| Next quarter's budget | Swings like a variable input cost | Controlled upfront based on TCO |
| Per-model cost visibility | You find out when the bill arrives | Monitored in real time via a gateway |
| Margin | Quietly eroded by AI spend | Protected by usage policy |
| Infrastructure timing | Reactive, after the crunch hits | Planned ahead, avoiding capital-timing risk |
How to audit your own AI bill, starting now
You don't need to be a Databricks customer for this. Here's Deloitte's AI tokenomics governance framework, translated into a checklist you can actually run at your org.
- Filter by strategy first
"If AI matters to your organization, the key question is whether it's deployed where it counts." Separate AI you're experimenting with for fun from AI you're strategically investing in. - Calculate TCO by department
Break down how token costs hit total cost of ownership at the department or product level. You can't cut what you can't see leaking. - Audit funding, usage, and unit economics separately
Check where this AI budget came from, who's using it and why, and what the underlying per-model pricing structure looks like — as three separate reviews, not one blended one. - Work through four risks in order
Forecast volatility (is spend swinging wildly?) → margin leakage (is it slowly eating margin?) → capital timing (are you only reacting after the fact?) → performance narrative (can you actually quantify ROI in numbers?). - Evaluate whether you need a multi-model gateway
If you're already running two or more AI models or vendors, check whether access control, cost monitoring, and policy enforcement live in one layer. It doesn't have to be Databricks — but if that layer doesn't exist yet, now's the time to build it.
Go deeper
TechCrunch original story The full reporting on how a $1B raise turned into $15B of demand, with Ghodsi's own words. techcrunch.com
Databricks' July strategic round announcement Where the "tokenmaxxing to valuemaxxing" line first appeared. databricks.com
Deloitte's CFO guide to AI tokenomics The source framework behind this piece's action checklist, with deeper risk analysis. deloitte.com
MarketScale — Unity AI Gateway deep dive A practitioner-level breakdown of what a multi-AI governance tool actually does. marketscale.com
PYMNTS — the investor's view Coatue's Thomas Laffont on the deal, plus the Microsoft partnership extension. pymnts.com
