Architecting intelligent storage for 50% storage and 25% compute savings at PB scale

At petabyte scale, cloud bills are not just a finance issue; they are an architectural verdict. While most organizations treat cost optimization as a reactive cleanup task, this team made it a platform capability that runs invisibly.

This talk shows how they embedded finops by design into their data platform, automating cost efficiency so developers never have to think about it:

• Automated profiling at scale: Heuristics analyze data distributions across billions of rows, with no manual inspection required.
• Intelligent physical layouts: Storage structures align to actual access patterns, using techniques such as adaptive encoding and modern compression.
• Self-tuning file optimization: ML-driven compaction minimizes cloud storage overhead and eliminates wasteful reads.

The result is a 50% reduction in storage footprint and a 25% boost in compute performance, achieved transparently without changing a single line of user SQL.