Case Study: Netflix builds a multimodal media data lake with LanceDB

Netflix powers its Media Data Lake with LanceDB, sustaining 20,000+ QPS and 5M+ IOPS

Netflix faced the challenge of managing petabytes of multimodal media assets, including video, audio, images, and text, which were spread across formats and workflows. Their traditional data systems were not built to handle this unstructured, large-scale data or to support the machine learning pipelines needed to make these assets discoverable and queryable. To address this, they partnered with LanceDB to build a Media Data Lake using its Multimodal Lakehouse platform.

LanceDB provided a unified solution with its Lance format, enabling zero-copy data evolution, hybrid querying combining SQL, vector search, and full-text search, and efficient random access to unstructured blobs. The platform integrated seamlessly with Netflix's machine learning workflows, supporting GPU-powered batch processing and scalable feature engineering. This resulted in a production-ready system that allows for faster experimentation, richer insights into creative workflows, and the efficient management of tens of petabytes of data under a single abstraction.

Netflix