Databricks’ US$1.3 billion purchase of MosaicML puts AI race in fast lane
Training and using artificial intelligence models will cost companies just ‘thousands of dollars, not millions’, under software unicorn’s vision
Sharanya Pillai
[SAN FRANCISCO] Yet another billion-dollar deal has landed in the red-hot artificial intelligence (AI) space: Databricks, a software unicorn backed by Microsoft, Amazon and Google, is buying a generative AI startup for US$1.3 billion.
The startup, MosaicML, is a nascent challenger to ChatGPT maker OpenAI. Announced on Monday (Jun 26), the deal follows other multibillion-dollar generative AI investments by tech majors.
Sceptics might call it a sign of froth, but Databricks chief Ali Ghodsi disagrees. He told The Business Times (BT) that it is “easy to justify” the deal’s value based on MosaicML’s growth potential, with the valuation being a “conservative” 10 to 20 times of annual recurring revenue (ARR).
“For that valuation of US$1.3 billion, if you can get somewhere between US$50 million and US$100 million ARR in the next 12 months, then multiplying that by 10 to 20 gets you to a US$1.3 billion valuation,” said Ghodsi, who spoke to BT at the InterContinental San Francisco, on the sidelines of Databricks’ annual Data + AI Summit.
“So the question then is, can this company get to US$50 million to US$100 million ARR in 12 months?... Looking at MosaicML, they absolutely can and would. Then the question is actually when we are combined, can we get there faster and even with a bigger number? I absolutely believe so. So then it’s easy to justify,” he added.
Founded in 2013, San Francisco-based Databricks specialises in enterprise tools for companies to streamline and manage their data, and harness it for AI applications. The company was last valued at US$38 billion in 2021, after raising US$1.6 billion in Series H funding. It employs over 5,000 staff globally.
BT in your inbox
Start and end each day with the latest news stories and analyses delivered straight to your inbox.
Databricks is vying for a future where training and using generative AI models, “will cost thousands of dollars, not millions”, the company said on Monday.
This is a vision shared by MosaicML, which was started in 2021 by former Intel executives and academics. The startup provides open-sourced enterprise tools for companies to build customised AI models, with a focus on making such innovation cheaper. It has also developed its own foundation models, MPT-30B and MPT-7B, which are base layers of technology that can power generative AI applications such as chatbots.
MosaicML has raised US$64 million to date from investors including deep tech-focused venture firms DCVC and Lux Capital. It was last valued at US$222 million, according to media reports, which means that the US$1.3 billion valuation would imply a valuation jump of about five times.
The price tag includes retention packages, although Ghodsi declined to disclose the proportion. All of MosaicML’s employees are expected to join Databricks.
Naveen Rao, co-founder and chief executive of MosaicML, said in a press statement on Monday that everyone should be able to build and train their own AI models, “imbued with their own opinions and viewpoints”.
“We started MosaicML to solve the hard engineering and research problems necessary to make large-scale training more accessible to everyone. With the recent generative AI wave, this mission has taken centre stage,” he noted.
The acquisition, which is subject to regulatory clearance, comes amid a wave of investments by tech majors into promising generative AI startups.
Last month, Databricks rival Snowflake acquired Neeva, a company that aims to leverage generative AI for Internet search. Google and Salesforce invested in Anthropic, another OpenAI rival. And the ChatGPT maker itself had previously raised US$10 billion from Microsoft.
Asked if Databricks is looking into more acquisitions, Ghodsi replied: “I think there’s a great opportunity to find good teams where the revenue is growing fast and they can bring customer value, and (where) one plus one equals three or four or five.”
But he noted wryly that while Databricks is speaking to many potential companies, “you have to kiss a lot of frogs” before finding the right target.
Copyright SPH Media. All rights reserved.
