Databricks Valuation and Revenue: How It Grew to $190B

Databricks is worth $190 billion.

That is the valuation from the $5 billion round Databricks announced on August 13, 2026, led by Coatue with Blackstone, MGX, T. Rowe Price and Sixth Street Growth. As of October 2026 it is the company's latest confirmed valuation, and Databricks is still private.

Revenue is growing faster than you would expect at this size. In the same announcement, Databricks said it had passed a $7 billion revenue run-rate, growing more than 80% year over year, with positive adjusted free cash flow over the previous 12 months.

Those are company-reported figures, not audited public results. They are still a long way from where Databricks started: by CEO Ali Ghodsi's own account, the company ended 2015 with about $1.5 million in revenue after a failed bet on self-serve growth.

The short version of how it got here: Databricks turned a popular free open-source project into a paid cloud platform, sold it hard to large enterprises, and kept adding the next thing data teams needed. That meant a data warehouse, then AI models, then databases and assistants built for AI agents.

Here is how it happened.

Databricks' growth in one table

Date	Milestone
2013	Seven Berkeley researchers found Databricks; Andreessen Horowitz leads a $14M Series A at a reported $47M valuation
2014	Spark ties the world record in the Daytona GraySort benchmark
2015	A self-serve push stalls and revenue ends the year near $1.5M, per Ghodsi
January 2016	Ghodsi becomes CEO and Databricks hires an SVP of worldwide sales
November 2017	Microsoft previews Azure Databricks as a first-party Azure service
February 2019	Raises $250M at a $2.75B valuation, per TechCrunch
October 2019	Raises a $400M Series F at $6.2B; revenue run-rate reaches $200M
January 2020	Publishes "What is a lakehouse?"
February 2021	Raises a $1B Series G at $28B; ARR ended 2020 at $425M
August 2021	Raises a $1.6B Series H at $38B; ARR passes $600M
August 2022	Discloses a $1B annualized run-rate
June 2023	Agrees to buy MosaicML for about $1.3B
September 2023	Raises $500M+ at $43B; run-rate passes $1.5B
December 2024	Announces a $10B Series J at $62B; expects a $3B run-rate by January
January 2025	Closes the Series J plus a $5.25B credit facility; Meta invests
May 2025	Agrees to buy Neon for about $1B
September 2025	Closes a $1B Series K at over $100B; run-rate passes $4B
December 2025	Announces a Series L at $134B; run-rate passes $4.8B
February 2026	Completes ~$5B equity and ~$2B debt at $134B; run-rate passes $5.4B
June 2026	Tells analysts annualized revenue is $6.9B, with $1.7B from AI products
July 2026	Signs a term sheet at $188B
August 2026	Closes $5B at $190B; run-rate passes $7B, growing 80%+
September 2026	Buys spreadsheet startup Row Zero, its latest 2026 acquisition

The valuation went from a reported $47 million to $190 billion in 13 years. The more useful question is what changed along the way.

1. Databricks started as seven researchers with a hit open-source project

Databricks came out of UC Berkeley's AMPLab.

Matei Zaharia, Ion Stoica, Ali Ghodsi, Reynold Xin, Patrick Wendell, Andy Konwinski and Arsalan Tavakoli-Shiraji were researchers working on Apache Spark. Spark was an engine for processing large amounts of data across many machines, built as a faster alternative to Hadoop MapReduce. The team's first pitch deck, later published by Fortune, says Spark had been in development at Berkeley since 2009.

Ghodsi has said the founders decided Spark needed a company behind it if enterprises were going to adopt it. In a 2017 interview on the Databricks blog, he said they founded the company in the summer of 2013.

The fundraising story is now well known. Ben Horowitz told Fortune the deck was one of the worst he had ever seen. Ghodsi said the team had hoped to raise about $200,000.

Horowitz offered $14 million instead.

That round valued Databricks at $47 million, according to a person familiar with the matter cited by Fortune. The 2013 deck set a goal of more than 500 customers and well over $10 million a year in revenue by the Series B.

The founders had technology, credibility and cash. What they did not have yet was a business.

2. Databricks won the Hadoop argument with a public benchmark

In the early years, Spark was not the default choice. Hadoop was.

Ghodsi described this period bluntly in a May 2024 talk at Stanford. Incumbent vendors kept telling customers and reporters that Spark only worked if your data fit in memory. Databricks kept denying it, and the claim kept getting repeated.

So the team set out to prove it in public.

In 2014, Databricks entered the Daytona GraySort benchmark, a third-party test of how fast a system can sort 100 terabytes of data. Using 206 machines on Amazon EC2, it sorted the data on disk in 23 minutes, according to Databricks. The previous Hadoop record had taken 72 minutes on 2,100 machines.

Databricks tied a UC San Diego team for the official record. It also sorted a full petabyte on 190 machines in under four hours.

Ghodsi said reporters picked up the result and the narrative flipped. Competitors that had dismissed Spark started embracing it, and coverage went from the occasional article to several a day.

It was a marketing lesson as much as a technical one. Ghodsi told the Stanford audience that technical founders underestimate marketing, and that Spark had "sucked for six years" before that turn.

3. Databricks gave up on self-serve and hired enterprise salespeople

Spark's popularity did not turn into revenue on its own.

The founders' plan after 2014 was product-led growth: a slick website, a credit card form and no salespeople. In 2015, Ghodsi said in the Stanford talk, the company doubled down on this with a program it called "Zero Touch."

Revenue went flat for the second and third quarters of 2015. Databricks ended the year at about $1.5 million.

Investors were not impressed by download charts. Ghodsi recalled being told to stop showing them and show revenue instead.

The first fix was the opposite of the original plan: hire experienced enterprise salespeople and have every one of them sell. In January 2016, Databricks announced that Ghodsi would become CEO, with Stoica moving to executive chairman. The same announcement named Ron Gabrisko, an outside hire, as senior vice president of worldwide sales.

Ghodsi later said he is "not a big fan of PLG personally." He compared counting on product-led growth to a child planning a career as an NBA champion.

4. Databricks sold a paid layer on top of free open source

The second problem was harder.

Spark was free, and the big cloud providers could offer it too. Ghodsi told a story at Stanford about Spark fans who asked the founders for selfies, then declined to pay $100,000 the next day because they could get Spark almost free on Amazon.

His conclusion was that an open-source company needs two "home runs." The first is an open-source project that the world adopts. The second is a separate, proprietary layer that is about 10 times more valuable and that customers pay for.

For Databricks, that layer started with performance. The company wrote a low-level engine that plugged into Spark and made it about 10 times faster, Ghodsi said. Faster jobs meant fewer machines, so Databricks could charge for the engine and still save the customer money on infrastructure.

The other choice was to sell only in the cloud. "We're 100 percent cloud-native," Ghodsi told CNBC in 2019. Databricks ran Spark as a managed service so companies did not have to configure or update it themselves.

It worked. By October 2019, Databricks said it had gone from almost no revenue to a $200 million run-rate in less than four years.

5. Databricks turned the cloud giants into distributors

Selling only in the cloud meant depending on the companies that could also compete with you.

Databricks chose to partner with them. In November 2017, Microsoft launched a preview of Azure Databricks as a first-party Azure service. Customers could start Databricks from the Azure portal like any native Azure product.

TechCrunch reported that the two companies had worked on the integration since 2016 and that it was a first for both. "There was such overwhelming demand for having Databricks on Azure," Ghodsi said at the time.

The strategy later showed up on the cap table. In February 2021, Amazon Web Services, Alphabet's CapitalG and Salesforce Ventures joined the $1 billion Series G at a $28 billion valuation, alongside existing investor Microsoft.

The partnerships run in the other direction too. In August 2026, Ghodsi told TechCrunch that Databricks has multibillion-dollar cloud commitments with all three major hyperscalers.

6. Databricks named a category and went after the data warehouse

For most of the 2010s, companies kept two kinds of systems. Data lakes stored huge amounts of raw data cheaply. Data warehouses handled structured data for business reporting.

In January 2020, Databricks published "What is a lakehouse?" The post described an architecture that put warehouse features, such as transactions and data quality controls, directly on top of cheap cloud storage in open formats.

The name gave Databricks a way to sell beyond data engineers and data scientists. If the lakehouse could do what a warehouse did, Databricks could compete directly with warehouse vendors such as Snowflake.

It built that product next. Ghodsi told reporters the data warehouse product launched in 2020 had topped $400 million in annualized revenue by June 2024.

The growth continued from there:

By December 2024, Databricks SQL had reached a $600 million run-rate, up more than 150% year over year.
By December 2025, the data warehousing business had passed a $1 billion run-rate less than four years after general availability.
By August 2026, it had passed $1.5 billion and was still growing over 100%.

The warehouse also helped Databricks expand inside existing accounts. A customer that already ran data engineering on Databricks could move reporting workloads there too.

7. Databricks raised big when money was cheap and stayed private

Databricks has raised money in unusually large amounts and has used being private as a tool.

In 2021 it raised twice: $1 billion in February and $1.6 billion in August at a $38 billion valuation. "In some sense, we're going public six months at a time," Ghodsi told Forbes that month.

At Stanford in 2024, Ghodsi said the company expected interest rates to rise and decided to "grab as much money as you can." He argued that staying private let Databricks make moves public companies avoid, such as price cuts, acquisitions and hiring in countries where the short-term return looked poor.

The rounds kept getting bigger. Databricks announced a $10 billion Series J at a $62 billion valuation in December 2024. The company said the money would fund employee liquidity, acquisitions and international expansion.

Investor demand has often exceeded what Databricks asked for. Ghodsi told TechCrunch the company targeted about $200 million before the 2021 Series G grew to $1 billion. In 2026 he said Databricks wanted $1 billion, saw $15 billion of interest, and settled on $5 billion.

8. Databricks bought its way into AI

When generative AI took off, Databricks moved fast with acquisitions.

In June 2023 it agreed to buy MosaicML, a platform for training and customizing large language models, in a deal valued at about $1.3 billion including retention packages. The pitch was that companies could build their own models on their own data, inside the platform where that data already lived.

A year later, Databricks bought Tabular, the company started by the creators of the Apache Iceberg table format. Ghodsi told CNBC Databricks paid over $1 billion, and CNBC reported that Snowflake and Confluent had also bid. TechCrunch, citing Bloomberg, later reported a price of nearly $2 billion for a startup doing about $1 million in ARR.

Databricks also partnered with model makers rather than trying to replace them. Meta joined as a strategic investor in January 2025, and Ghodsi told Reuters thousands of customers were using Meta's Llama models on Databricks. In March 2025 it signed a five-year partnership with Anthropic to offer Claude models natively on the platform.

The AI products became a business of their own:

Databricks said its AI products passed a $1 billion run-rate in September 2025.
They reached $1.4 billion by February 2026.
Ghodsi told CNBC they reached $1.7 billion by June 2026.

The buying has not stopped. TechCrunch counted Quotient AI and SiftD.ai in March 2026, security startup Panther in June, Electric in August and Row Zero in September. "We intend to do many more acquisitions like this in the future," Ghodsi said.

9. Databricks treated AI agents as a new kind of customer

The most recent growth move is a bet on software agents as heavy users of data infrastructure.

In May 2025, Databricks agreed to buy Neon, a serverless Postgres company, for about $1 billion, Reuters reported. Databricks said Neon's telemetry showed more than 80% of its databases were created by AI agents rather than humans.

A month later, Databricks launched Lakebase, an operational database built on Neon's technology. Operational databases run the day-to-day transactions behind applications, a market Databricks put at more than $100 billion.

Lakebase passed a $100 million revenue run-rate by August 2026, according to Databricks.

The other piece is Genie, which Databricks now calls an "AI coworker" that answers business questions from company data. Ghodsi told TechCrunch in August 2026 that Genie "is insanely popular."

Agents also change the economics. "The agents are generating way more queries," Ghodsi told CNBC in June 2026. Because Databricks charges for consumption, more agent activity means more revenue, but also more spending on the underlying AI models.

How fast did Databricks grow?

Databricks does not publish audited results, so most figures below are company-reported revenue run-rates. A run-rate annualizes recent revenue; it is not the revenue Databricks booked over the previous 12 months. Databricks' fiscal year ends on January 31.

2015: about $1.5 million in revenue, according to Ghodsi.
October 2019: a $200 million run-rate.
End of 2020: $425 million in ARR, up 75%.
August 2022: a $1 billion annualized run-rate.
Fiscal year ended January 2024: more than $1.6 billion in revenue, up over 50%.
January 2025 quarter: an expected $3 billion run-rate.
July 2025 quarter: a $4 billion run-rate, up more than 50%.
October 2025 quarter: a $4.8 billion run-rate, up more than 55%.
January 2026 quarter: a $5.4 billion run-rate, up more than 65%.
July 2026 quarter: a $7 billion run-rate, up more than 80%.

The unusual part is the direction of the growth rate. Most software companies slow down as they get bigger. Databricks went from about 50% growth in mid-2025 to more than 80% a year later.

Existing customers drive much of this. Databricks has reported a net retention rate above 140%, meaning existing customers as a group spend over 40% more a year later. In mid-2023, Databricks had more than 300 customers spending at a $1 million-plus annual rate. By August 2026, it had more than 1,000.

Databricks' growth loop

Databricks' growth works roughly like this:

Open-source projects such as Spark, Delta Lake and MLflow bring data teams in for free.
Companies pay Databricks to run those tools faster and with less work in the cloud.
Once a company's data sits on Databricks, adding a new product is easier than adopting a new vendor.
Databricks keeps building or buying that next product: a warehouse, AI models, a database, an assistant.
Usage-based pricing means more workloads, and now more agents, translate into more revenue.
Faster growth attracts more private capital, which pays for more research and acquisitions.

The loop depends on every step. Without the open-source pull, sales would be harder. Without the paid layer, the open-source pull would mostly benefit the cloud providers.

What Databricks did differently

It changed course quickly when its own plan failed

The founders believed in self-serve growth and disliked the idea of a traditional sales team. When revenue went flat in 2015, they reversed both beliefs within months.

It made proof public

The 2014 sort record did more than any sales pitch could have. A neutral benchmark gave reporters a story and gave buyers a reason to stop believing Spark only worked in memory.

It built with the cloud giants instead of against them

Becoming a first-party Azure service and later taking money from AWS, Google and Salesforce gave Databricks distribution it could not have built alone.

It kept moving into the next budget

Data engineering came first, then data science, then the warehouse, then AI, then operational databases. Each move gave existing customers another reason to spend more.

It used private capital as a weapon

Ghodsi has openly argued that staying private lets Databricks make long-term moves that quarterly reporting would discourage. Its run of acquisitions since 2023 is the clearest example.

What could slow Databricks down?

The first risk is margins. Ghodsi told CNBC in June 2026 that gross margin will go lower as agents drive up spending on AI models. He declined to give the current figure.

The second is the price. A September 2026 PitchBook report argued that the $190 billion valuation equals about 27 times the $7 billion run-rate and that the most optimistic scenario still valued Databricks below that price. The report also noted that Databricks' numbers are self-reported, unlike Snowflake's audited results.

The third is competition. Snowflake, AWS, Google Cloud and Microsoft are all building tools that connect AI to company data, as TechTarget noted in August 2026. The same cloud providers that distribute Databricks also sell their own alternatives.

The fourth is integration. Databricks has bought many companies quickly, and analysts quoted by TechTarget said it still needs to show those products work together as one platform.

The fifth is the IPO. Ghodsi told Bloomberg in June 2026 that "this is a terrible year to go public," though he said Databricks will be a public company. Employees and investors will eventually want a public market, and that will bring the scrutiny Databricks has avoided.

Borrowing Databricks' benchmark playbook

Databricks' turning point came from publishing proof that reporters and buyers could repeat, not from a bigger ad budget. Okara's Writer Agent drafts strategy-led articles that make that kind of case and can publish them. The GEO Agent then tracks whether ChatGPT, Google AI Overviews and Google AI Mode mention you, for up to 15 prompts.

Frequently asked questions

What is Databricks' valuation?

Databricks is valued at $190 billion, based on the $5 billion funding round it closed in August 2026. That was up from $134 billion in its Series L, completed in February 2026, and $62 billion in its Series J, closed in January 2025. A July 2026 term sheet had set the price at $188 billion before the final close. As of October 2026, $190 billion is the latest confirmed valuation.

What is Databricks' revenue?

Databricks said in August 2026 that it had passed a $7 billion revenue run-rate in its fiscal second quarter, growing more than 80% year over year. Its data warehousing product makes up more than $1.5 billion of that run-rate, and its AI products were at $1.7 billion as of June 2026. These are company-reported, annualized figures rather than audited annual revenue. The last full-year revenue figure Databricks disclosed was more than $1.6 billion for the fiscal year that ended in January 2024.

How much funding has Databricks raised?

TechTarget reported that the August 2026 round brought Databricks' total financing to more than $32 billion, a figure that includes debt. The largest rounds were the $10 billion Series J in 2024 and 2025, about $5 billion of equity in February 2026, and $5 billion in August 2026. Databricks has also taken on a $5.25 billion credit facility in January 2025 and about $2 billion of additional debt capacity in February 2026.

Who founded Databricks?

Databricks was founded in 2013 by seven UC Berkeley researchers behind Apache Spark: Ali Ghodsi, Matei Zaharia, Ion Stoica, Reynold Xin, Patrick Wendell, Andy Konwinski and Arsalan Tavakoli-Shiraji. Stoica was the first CEO. Ghodsi took over as CEO in January 2016, and Stoica became executive chairman.

How many customers does Databricks have?

Databricks says more than 20,000 organizations use its platform, including 70% of the Fortune 500, as of August 2026. More than 1,000 customers spend at an annual rate above $1 million, and more than 100 spend above $10 million.

Is Databricks going public?

Not yet. As of October 2026, Databricks is private and has not announced an IPO date. Ghodsi has said the company will go public eventually but called 2026 a poor year to do it, with SpaceX, OpenAI and Anthropic dominating IPO attention.

How did Databricks grow so fast?

Databricks turned the popularity of Apache Spark into a paid cloud platform, then dropped self-serve growth for enterprise sales in 2016. It partnered with Microsoft, AWS and Google instead of fighting them, named and built the lakehouse to compete with data warehouses, and bought its way into generative AI. Usage-based pricing and net retention above 140% mean existing customers keep spending more.