MosaicML

Pioneer in cost-efficient foundation model training, now integrated as a core AI capability within Databricks' enterprise data platform

MosaicML was a San Francisco-based AI infrastructure company founded in 2021 that developed a platform for efficiently training and deploying large language models (LLMs) and other foundation models. The company's flagship product, the MosaicML Platform, enabled enterprises to train custom AI models on their own data at dramatically reduced cost and time compared to alternatives, leveraging proprietary efficiency techniques and optimized orchestration across cloud GPU infrastructure. MosaicML also released open-source models such as MPT-7B and MPT-30B, which gained significant traction in the developer community for their permissive licensing and strong performance. MosaicML positioned itself at the intersection of MLOps and foundation model development, targeting enterprises that wanted the power of custom LLMs without the prohibitive costs typically associated with training at scale. The company's Composer library, an open-source deep learning training framework, embodied its core thesis that algorithmic efficiency improvements could dramatically reduce compute requirements. This approach resonated strongly with enterprises wary of vendor lock-in from closed model providers like OpenAI. In July 2023, Databricks acquired MosaicML for approximately $1.3 billion, one of the largest AI acquisitions of that year. Following the acquisition, MosaicML's technology and team were integrated into Databricks and rebranded as Databricks Mosaic AI. The platform continues to operate as a core pillar of Databricks' AI strategy, offering enterprises end-to-end capabilities for training, fine-tuning, and deploying custom foundation models within the Databricks Data Intelligence Platform. The MosaicML brand itself has been largely retired in favor of the Databricks Mosaic AI identity.

• 2021Founded