Databricks reached a $43 billion valuation after raising $500 million in a Series I funding round led by T. Rowe Price Associates. Existing investors including Andreessen Horowitz, Fidelity Management & Research Company, and GIC also participated in the round. The company surpassed a $1.5 billion revenue run rate with more than 50 percent year-over-year growth. During the second quarter, Databricks saw its strongest quarterly incremental revenue growth. More than 10,000 organizations use the Databricks Lakehouse Platform to unify data and AI. Capital One, a major customer of Snowflake, also participated in this latest funding round.
The company builds its generative AI stack by integrating technology from acquisitions like MosaicML and Okera. Databricks acquired MosaicML in July 2023 for $1.3 billion to improve AI training and deployment. This acquisition brought in a team of highly sought-after engineers to help customers build custom machine learning models. Databricks also acquired Lilac to help customers manage unstructured data for generative AI. The company provides a full-stack solution where customers train and deploy models inside their own environments. This approach helps organizations keep sensitive data within their control.
The Lakehouse AI toolbox helps developers build intelligent assistants. This toolbox includes vector search, low-code AutoML, and MLflow 2.5. Vector search helps developers improve the accuracy of generative AI responses by using embeddings managed in Unity Catalog. AutoML provides a low-code way for technical and non-technical users to fine-tune large language models using enterprise data. Users access these tools via Notebooks or the Assistant in the SQL editor. The Assistant helps with writing queries and answering data questions. While these tools aim to simplify workflows, the Assistant remains in preview.
DBRX architecture and enterprise deployment
Databricks employs Naveen Rao, the former CEO of Nervana, as the vice president of artificial intelligence. Rao leads the effort to bring DBRX, an open-source large language model, to enterprise users. DBRX is a Mixture-of-Experts model with 132 billion total parameters and 36 billion active parameters per token. The model was trained on approximately 12 trillion tokens and uses a 32,000 token context window. Databricks offers two public variants: DBRX Base and DBRX Instruct.
The DBRX Instruct model focuses on efficient training and inference for enterprise applications. Users can access these models through Foundation Model APIs and Model Serving using pay-per-token or provisioned throughput. The Mosaic AI Training toolkit allows users to point the service at a Unity Catalog dataset and a base model to run fine-tuning jobs. This feature is currently in public preview.
| Model Spec | DBRX Instruct Detail |
|---|---|
| Architecture | Mixture-of-Experts |
| Total Parameters | 132B |
| Active Parameters | 36B |
| Training Data | 12T tokens |
| Context Window | 32K |
The company also provides LakehouseIQ, an engine that uses machine learning models to infer data context. This engine uses natural language processing to help users search for data without technical help. Unlike the knowledge graph methods used by Salesforce or Glean, LakehouseIQ uses generative AI to understand jargon and organizational structures. Users access LakehouseIQ via APIs for custom applications or through the SQL editor. One limitation remains that DBRX Instruct does not fit certain specific workloads in a single request.
Competitive positioning against Snowflake
Databricks positions its platform to lead in generative AI development. Analysts note that Databricks is ready to help customers build custom machine learning and AI models. This puts pressure on Snowflake, which is currently in private preview with its Snowpark Container Services. Snowflake is also building generative AI capabilities through its partnership with NVIDIA and the acquisition of Reka AI. Snowflake acquired Reka AI for $1 billion to improve its ability to handle multimodal data like text, images, and video.
The rivalry involves how each company handles enterprise data security and model training. Databricks uses MosaicML to let customers escape vendor lock-in from providers like OpenAI or Google. This setup allows enterprises to use their data securely without sending it to third-party services. Snowflake also aims to enable Retrieval-Augmented Generation applications through its integration of Reka AI and Streamlet.
| Capability | Databricks Status | Snowflake Status |
|---|---|---|
| Generative AI Tools | General Availability / Public Preview | Private Preview |
| Model Acquisition | MosaicML ($1.3B) | Reka AI ($1B) |
| Data Unification | Lakehouse Platform | Data Cloud |
Databricks faces competition from Snowflake, which has a large customer base and is building its own language models. However, Databricks reported its highest-ever quarterly incremental revenue growth while Snowflake saw its value drop by nearly 45 percent over the last two years. Some analysts believe Databricks differentiates itself by offering query performance gains for generative AI use cases. The company also works with NVIDIA to accelerate data processing and generative AI models. Does the aggressive expansion into model development through acquisitions create long-term sustainability for the Lakehouse architecture?
