How Meta Scaled AI Training Storage via Data Normalization | Sarang Masti and Weiran Liu from Meta @scaleconference
How Meta Scaled AI Training Storage via Data Normalization | Sarang Masti and Weiran Liu from Meta  @scaleconference
Uploaded June 2026 | Updated November 2026, 4 days ago
Training data for Meta's recommendation systems was entirely stored in Data Warehouse, structured as relational tables where each row captures labels and snapshotted features at the point of recommendation.

New modeling techniques, such as learning from user sequences and multi-modality, has led to a 10-100x increase in feature size, making the training data increasingly cost-prohibitive due to high duplication. The same user's features are stored repeatedly for every recommendation request, with highly popular content features being duplicated potentially over a million times.

We present a co-designed data and infrastructure in order to address the scaling challenge. By moving features out of training samples into a high-performance indexing storage and implementing model access pattern-aware pushdown optimizations, we have achieved a 10x storage cost reduction for the largest feature: long user sequences.

Want to learn more about the @Scale Conference? Visit our website here: atscaleconference.com
How Meta Scaled AI Training Storage via Data Normalization | Sarang Masti and Weiran Liu from MetaLive Panel: Behind-the-Scenes with the Team: How Meta Builds Next-Gen Wearable ExperiencesScaling AI Infrastructure: Inside CoreWeave’s Global Network with Jim JulsonMeta processes billions of video uploads daily@Scale: Systems & Reliability is coming to Bellevue, WA!The Hottest Job in the Market: Inside Meta’s AI Hardware LabWhat happens when a network incident hits at scale?How Meta Built AI That Actually Works at ScaleWhich 2026 event are you joining?Preview Player Reliability and the Agentic Defense Loop on Edits | Bhumi Sabarwal & Ravi ShahData Governance in the World of Agents | Komal Mangtani from MetaThe Hottest Job in the Market: Inside Meta’s AI Hardware Lab | Joshua Held and Yashar Bayani
@Scale |

How Meta Scaled AI Training Storage via Data Normalization | Sarang Masti and Weiran Liu from Meta

SHARE TO X SHARE TO REDDIT SHARE TO FACEBOOK WALLPAPER