Databricks, a data infrastructure and MLOps platform, has entered a definitive agreement to acquire MosaicML, an ML model development platform for generative AI applications, in a transaction valued at approximately USD 1.3 billion, inclusive of retention packages. Upon completion of the transaction, the entire MosaicML team is expected to join Databricks.

The acquisition will integrate MosaicML's platform within the Databricks' Lakehouse platform, offering organizations a unified platform with tools to create and manage a wide range of AI use cases and large language models (LLMs) using their own data without the high costs associated with building such applications. MosaicML claims that its automatic model training optimization feature enables 2x–7x faster training than typical approaches, and when combined with near linear resource scaling, multi-billion-parameter models can be trained in hours rather than days. 

Formed in 2021, MosaicML provides a platform for developing generative AI models, as well as pre-training and fine-tuning LLMs. Furthermore, it has created its own LLMs, MosaicML Pretrained Transformer (MPT) LLMs, which are open source, licensed for commercial usage, and equal to the quality of LLaMA-7B created by Meta , developed with 65 billion parameters. The penultimate version of MosaicML’s LLM, MPT-7B, had 3.3 million downloads, and its clients include AI2 (Allen Institute for AI), Generally Intelligent, Hippocratic AI, Replit, and Scatter Labs.

Analyst QuickTake: The transaction value of MosaicML—which is 5.9x higher than its previous valuation of USD 222 million—signifies the increasing activity and demand for generative AI tools and applications. Notable product launches in this space by ML infrastructure players over the past month include 1) AMD launching a new AI accelerator chip specifically designed to handle generative AI workloads, 2) OctoML's self-optimizing compute service tailored for generative AI applications, 3) Datasaur's tool for training personalized ChatGPT models, and 4) the emergence of RefuelAI , which provides clean and labeled training data for AI models utilizing LLMs.