Using AI to Find Truth in Data
summary
Compute.ai separates compute from data management, installing a pure compute engine on every server that runs SQL directly on files and supports Iceberg/Parquet; the architecture scales from one node to hundreds, targets real‑time use cases like Uber‑style routing, and notes 42% of customers run both warehouses and lakehouses. The product uses AIML for AI-driven paging to disk, claiming 100% CPU utilization, safe memory overcommit to avoid OOMs, and linear cost scaling while citing a CSAT estimate of $25 trillion compute demand by 2026.
AI-written summary · about
chapters (16)
- 0:00Introduction to session
- 1:22Supercloud and data demand
- 2:25Semantic layer challenges
- 3:26Compute vs storage separation
- 4:53Coherent data at scale
- 6:19Defining semantic layer
- 7:04Real-time data use case
- 8:07Open standards and Iceberg
- 8:48Disruption to incumbents
- 9:53Data everywhere problem
- 10:55Partnership with Snowflake
- 11:59AI in Compute.ai
- 13:03AI infrastructure for production
- 14:47Running proprietary models
- 15:32Application market evolution
- 16:37Supercloud future with AIML
