Databricks, a leading Data and AI company, announced today that it has signed a definitive agreement to acquire MosaicML, a startup that builds software tools for training and deploying large neural networks. Valued at approximately $1.3 billion, the deal is Databricks’ largest acquisition to date and represents a strategic bet on AI democratization, as the two companies aim to make generative AI accessible for every organization.
MosaicML is best known for creating MPT, a family of commercial Apache 2.0 licensed, open-source foundation models. Through their platform, companies can easily customize and fine-tune these models using their own data, or deploy any model (open source or proprietary) to their cloud provider of choice. Their most recent release, MPT-30B, features an 8k context window and is comparable to LLaMa-30B and Falcon-40B. A variety of organizations, including AI2 (Allen Institute for AI), Generally Intelligent, and Hippocratic AI, have leveraged MosaicML for diverse generative AI use cases.
With this acquisition, Databricks intends to enhance its Lakehouse Platform, a unified Data and AI platform, with MosaicML's generative AI capabilities. This merger is expected to offer customers a straightforward, expedient method to retain control, security, and ownership of their data, with the promise of reduced costs. MosaicML claims that its automatic optimization of model training provides 2x-7x faster training than standard approaches, allowing organizations to train multi-billion-parameter models in hours rather than days. Additionally, they believe leveraging both platforms will reduce the cost of training and utilizing LLMs from millions to thousands of dollars.
“Every organization should benefit from the AI revolution with more control over how their data is used,” said Ali Ghodsi, CEO of Databricks.
As part of the agreement, the entire MosaicML team, including its industry-leading research team, is expected to join Databricks post-transaction. MosaicML's platform will be gradually integrated into Databricks' Lakehouse platform, and it's models will remain open source.
"At MosaicML, we believe in a world where everyone is empowered to build and train their own models — and joining forces with Databricks will help us make that belief a reality," expressed Naveen Rao, Co-Founder and CEO of MosaicML.
The acquisition is subject to regulatory approval and other closing conditions. MosaicML and Databricks will continue to operate separately until the transaction is complete.