Interview, Fireside Chat
Benoit Dageville, Snowflake ｜ RAISE Summit 2025
- Transitioning platform focus to treat unstructured data as a first-class asset, aiming to integrate it seamlessly with structured and semi-structured information.
- Elevating SQL to the primary language for unstructured data queries through the introduction of "AI SQL."
- Optimizing GPU utilization to eliminate idle cycles via vectorized execution and pipeline stage overlap, targeting improved efficiency and cost reduction.
- Prioritizing a three-layer roadmap comprising Level Zero (data), Level One (semantic views), and a third layer dedicated to governance, security, and asset management.
- Deploying semantic views to define precise metrics, aggregation rules, and table relationships, enabling AI to process complex business definitions autonomously.
- Reporting that shipped Semantic Views have already achieved "beyond 95% quality," with an explicit future goal of reaching 100% accuracy.
- Expanding connectors for unstructured sources like SharePoint to handle semantic relationship complexities while preserving access control lists (ACLs) and security.
- Materializing specific semantic computations to optimize processing speed and relevance for frequently asked queries.
- Defining graph structures as inherent to all data types, utilizing either materialized pointers or semantic models to represent relationships.
- Establishing the open-source Polaris protocol to unify data cataloging alongside the proprietary Horizon platform, which includes compute assets in the catalog.
- Characterizing competitor claims of being "faster and less expensive" as incorrect and recommending independent TPCDS benchmarks for verification.
- Architecting the platform to support all data types and workloads, including analytics, transactional, and AI, under a single roof to minimize complexity and cost.
- Democratizing data access by enabling business users to interact directly with the platform via natural language, eliminating reliance on technical personnel for access management.
- Viewing the integration of unstructured data and direct natural language interaction as the onset of a significant industry revolution.