Key announcements from the Snowflake Data Cloud Summit 2024

This year's Snowflake Data Cloud Summit was squarely focused on AI innovation, with the company delivering a series of announcements related to features, capabilities, and partnerships meant to empower users to leverage their data assets to build powerful AI applications.

Polaris Catalog and generally available Iceberg Tables

Snowflake introduced Polaris Catalog, a vendor-neutral, open catalog implementation for Apache Iceberg. The latter is the most popular open standard for data lakehouses, lakes, and other modern infrastructure implementations. Polaris Catalog gives users a single centralized place where any engine such as Apache Flink, Apache Spark, Dremio, Python, and Trino can find and access their Iceberg tables without losing compatibility and interoperability. Organizations can host the Polaris Catalog on Snowflake's AI Data Cloud (in public preview soon) or self-host it using containers, including Docker or Kubernetes.

The Polaris Catalog launch is part of Snowflake's commitment to the Iceberg ecosystem's growth, reflected in its investments in Iceberg Tables (now generally available) and even in the strategic partnership with Microsoft. By making Snowflake and Fabric seamlessly interoperable, both platforms can continue to support the industry’s leading open standards for storage formats – Apache Iceberg and Apache Parquet. The addition of the Polaris Catalog means that users can now harness the power of their data to build AI applications at scale, regardless of where the data is stored.

Snowflake Cortex AI and Snowflake ML

Snowflake's fully managed service to build large language model (LLM) applications, Snowflake Cortex AI, received multiple enhancements including:

• the Snowflake Cortex Analyst and Snowflake Cortex Search chat experiences (both in public preview soon) that enable users to build chatbots that leverage their structured and unstructured data in minutes;
• the Snowflake AI & ML Studio (private preview) no-code interface for easily customizing and productizing AI applications based on Snowflake's open source LLM Snowflake Arctic or a selection of leading LLMs from top providers, including Google, Meta, Mistral AI, and Reka;
• Cortex Fine-Tuning, a serverless fine-tuning customization feature available within the Snowflake AI & ML Studio;
• the LLM-based input-output safeguard Snowflake Cortex Guard (generally available soon), based on Meta's Llama Guard;
• pre-built AI powered experiences, like Document AI to extract content from documents, and Snowflake Copilot (generally available soon), the company's breakthrough text-to-SQL assistant.

Snowflake Notebooks, Snowflake Trail, Snowflake Marketplace and other developer tools

In addition to its enterprise-ready solutions, Snowflake also unveiled a suite of developer tools and innovations that accelerate tasks such as building enterprise-ready data pipelines, ML models and AI-powered applications. Some of the most notable announcements include Snowflake Notebooks for integrated development, a Snowpark pandas API, DevOps capabilities like Database Change Management and Git integration, observability features in Snowflake Trail, and a Native App Framework enabling developers to build apps once and then deploy and distribute them to Snowflake customers through the Snowflake Marketplace, which already boasts over 160 Snowflake Native Apps.

NVIDIA AI powered Snowflake AI data applications

The new collaboration between Snowflake and NVIDIA will allow customers to build custom generative AI applications in Snowflake, powered by NVIDIA. Snowflake announced that it is adopting the NVIDIA AI Enterprise software to integrate NVIDIA's NeMo Retriever microservices into Snowflake Cortex AI. The integration will allow organizations to connect custom models to internal data and enhance the accuracy of the models' responses. Additionally, the Snowflake Arctic is fully supported with NVIDIA TensorRT-LLM software for optimized performance, and is available as an NVIDIA NIM microservice. These announcements build on Snowflake and NVIDIA's initial collaboration aimed at delivering "a single, unified AI infrastructure and compute platform in the AI Data Cloud."