Databricks: Inside the $190B Data and AI Lakehouse Platform

Databricks has never been a public company, has never traded a share on an exchange, and still just raised money twice in six months at valuations of $134 billion and then $190 billion. The business underneath that number is a 12-year-old data platform that quietly became the infrastructure layer three-quarters of the Fortune 500 build their AI on.

Most coverage of Databricks is a valuation headline followed by a Snowflake comparison. What gets skipped is how the business actually works: what customers pay for, why the company keeps buying startups instead of building everything in-house, and whether $7 billion in revenue supports a $190 billion price tag. That's what this piece is for.

Company Snapshot

What Databricks Actually Does

Databricks sells a “lakehouse” — a single platform that stores data once, in open formats on cloud object storage the customer already owns, and runs every workload against it: batch data engineering, SQL analytics and dashboards, model training, and now AI agents. The alternative most enterprises lived with for a decade was a data lake (cheap, flexible, hard to govern) bolted next to a separate data warehouse (fast, structured, expensive and rigid). Databricks' pitch, built on the open-source Apache Spark and Delta Lake projects its founders created, is that you don't need both.

That architecture is why the company has ridden three separate waves without changing its core product: 2010s big-data engineering, the cloud data-warehouse wave that made Snowflake public, and now enterprise AI, where the same governed data becomes the training and retrieval layer for agents. More than 20,000 organizations use the platform today, including over 60% of the Fortune 500, per the company's own disclosures.

The product spans four layers: data engineering and governance (Lakeflow, Unity Catalog), analytics and BI (Databricks SQL, Genie), AI and agents (Mosaic AI, Agent Bricks), and, since 2025, an operational database (Lakebase) and a security lakehouse built around the 2026 acquisition of Panther. Every layer runs on the same governed data — the actual moat, not any single product.

How Databricks Makes Money

Databricks doesn't sell seats. It sells consumption. Customers pay for Databricks Units (DBUs), a usage metric billed per second, on top of a separate cloud infrastructure bill they pay directly to AWS, Azure, or GCP. The total cost of running Databricks scales with how much data a customer moves and how many models it trains or serves, not headcount — which is why revenue can compound as customers push more AI workloads onto the platform without a new contract each time.

Pricing varies by workload: AI model serving runs around $0.07/DBU (the cheapest tier, aimed at price-competitive hosted inference), scheduled data-engineering jobs run about $0.15/DBU, interactive notebooks cost roughly $0.40/DBU, and fully managed serverless SQL tops out near $0.70/DBU on AWS — the premium for instant-start, zero-ops warehouse queries. For the full unit-economics breakdown, see our how Databricks makes money post.

Net dollar retention above 140% is the number that matters most: existing customers expand DBU consumption fast enough, largely by adopting new layers (AI, then database, then security), that Databricks grows revenue substantially before signing a single new logo — the flywheel every consumption platform wants and few achieve at this scale.

Funding History

Databricks has raised more than $25 billion across 16 rounds since its 2013 Series A, and the pace has accelerated sharply. It raised more in the seven months between its December 2024 Series J and the August 2026 round than in its first decade combined.

Sources: TechCrunch (2019 Series F), Databricks newsroom press releases (Series G through Series K), CNBC (Feb 2026 Series L), and Bloomberg (Aug 2026 round).

The August 2026 round was led by Coatue, with Blackstone, MGX, and T. Rowe Price among the largest checks, plus new investor Sixth Street Growth and additional new backers BOND, Clearlake Capital, Point72, Premji Invest, and TPG — alongside repeat participants Andreessen Horowitz and Thrive Capital, who have now backed the company across multiple rounds spanning a decade. For the full breakdown of that round's terms and investor list, see our post on Databricks' $5B round at a $190B valuation.

Product Portfolio

Databricks ships across five product layers, and the pitch to customers is that all five sit on the same governed copy of their data — nothing has to be re-exported or duplicated to move from one layer to the next.

The Acquisition Flywheel: How Databricks Bought Its Way to a Full AI Stack

Every major new product layer Databricks has shipped since 2023 started as somebody else's startup. Rather than build each new layer of the AI stack in-house from scratch, the company has run a deliberate build-then-buy pattern: identify the layer customers are asking for, acquire the best point solution, then rebuild it into Unity Catalog and DBU billing.

It started with MosaicML in 2023 (roughly $1.3 billion), the foundation of Mosaic AI's model-training layer. Next was Tabular in 2024 (over $1 billion), the company behind the competing Apache Iceberg table format — an acquisition that mattered less for the technology than for ending a format war fragmenting the open-data ecosystem Databricks depends on. Then Neon in 2025 (about $1 billion), a serverless Postgres database that became Lakebase, an operational database layer for AI agents that read and write state, not just query history. Databricks agreed in June 2026 and closed in August 2026 on Panther, an AI-native security operations platform, folding cybersecurity detection into the lakehouse under a new “security lakehouse” category.

That pattern is the real competitive story, more than any single product. Snowflake, AWS, and Google are all racing to bolt AI features onto existing platforms. Databricks is instead assembling an increasingly complete AI-native stack — training, serving, agents, a database, and now security — through acquisitions that all land on the same governed data layer. Whether that keeps working depends on integration execution, a much harder problem than the press releases suggest.

Revenue and Key Metrics

Revenue growth has been remarkably steady for a company at this scale: $4.8 billion at the end of Q3 2025, $5.4 billion by January 2026 (+65%+ YoY), and past $7 billion by mid-2026 (+80%+ YoY) — growth accelerating even as the base roughly doubled in 12 months, according to Databricks' own August 2026 announcement. Data warehousing alone crossed a $1.5 billion run-rate in June 2026, up from $1.0 billion in Q3 2025, showing the AI story hasn't come at the expense of the older analytics business.

What separates Databricks from most AI-era unicorns: it turned free-cash-flow positive for full-year 2025 and stayed there — a real business subsidizing an aggressive growth rate, not the reverse. That combination of 80%+ growth and positive free cash flow at a $7 billion base is genuinely rare, and a large part of why growth investors kept re-upping in back-to-back mega-rounds instead of waiting on a public listing for exposure.

Competitive Landscape

Databricks' closest and most-covered competitor is Snowflake, the public data-cloud company it's most often benchmarked against. The two started from opposite ends of the same problem — Snowflake built a fully managed proprietary warehouse for teams that didn't want to manage infrastructure, Databricks built an open lakehouse for teams that wanted direct control of raw data — and both are now racing to own the AI layer on top.

Databricks is private and self-reported; Snowflake figures are from public investor disclosures. Not a perfectly like-for-like comparison, but directionally reliable.

Databricks is growing roughly 2.7x faster than Snowflake off a larger revenue base, carries higher net dollar retention, and is free-cash-flow positive — why private investors have been willing to value it at nearly 1.7x Snowflake's public market cap despite still being private. The open-format architecture is the real product difference: Databricks stores data in Delta Lake and Parquet on cloud storage the customer controls, so other tools can point directly at the same files, while Snowflake's proprietary format keeps data inside its own managed system.

The bigger long-term threat may not be Snowflake at all, but the cloud hyperscalers — AWS Bedrock/SageMaker, Google Vertex AI, and Microsoft Fabric — each bundling similar governance-plus-AI capabilities into infrastructure customers already pay for. Databricks' answer is staying multi-cloud and open-format by design, so leaving a hyperscaler's native tools never means re-platforming the underlying data.

Leadership Team

Bull Case / Bear Case

IPO Outlook

Ghodsi told Bloomberg Television in June 2026 that 2026 was “a terrible year to go public,” citing a crowded IPO calendar including SpaceX, OpenAI, and Anthropic, and pointed to 2027 instead. By August 2026, alongside the $190 billion round, he said it is now “very unlikely” Databricks lists before either OpenAI or Anthropic does — a softer commitment than “2027,” and a sign the timeline could keep slipping.

No S-1 has been filed, confidentially or publicly, as of September 2026. Ghodsi has said the main reason to eventually go public is a liquidity mechanism for employees, not a need for capital — the company has raised $10 billion in the last twelve months alone from private investors on increasingly favorable terms, removing most of the usual pressure to list.

The practical read: as long as growth-stage capital keeps showing up at rising valuations, Databricks has little incentive to trade public-market scrutiny for a liquidity event it can partly replicate through secondaries inside private rounds. That puts an IPO more plausibly in late 2027 or 2028 than in the near term.

The Bottom Line

Databricks is a rare case in the current private-market cycle: valuation growth roughly matched by revenue growth rather than running ahead of it. $190 billion on $7 billion of annualized revenue is a rich multiple, but it's attached to 80%+ growth, positive free cash flow, and a customer base that keeps expanding its own spend. The acquisition strategy is building a genuinely differentiated AI stack faster than most competitors can build the same pieces in-house, even if integrating four separate companies is still underway. The open question isn't whether the business is real — the free-cash-flow numbers say it is. It's whether growth this fast holds as the revenue base climbs toward $14 billion, and whether an IPO ever beats another private round at a higher price.

All financial figures are based on publicly reported data, company press releases, and credible media reports as of September 2026. Databricks is a private company and does not publicly disclose audited financials; revenue and headcount figures cited here are from company disclosures and reported estimates, not independently verified.