AI brings the focus front and center to both new data types like embeddings, image, video, and video, and new workloads like search and training. The core of the traditional data lake is a parquet + iceberg powered storage layer for fast OLAP queries and cannot meet the new demands of AI. Instead, Lance format is the new open-source standard for the storage layer for AI data.

For the world’s most cutting-edge multimodal companies like Midjourney, Luma Labs, and WorldLabs, Lance format delivers way faster performance than parquet, superior schema evolution to Iceberg, blazing fast search on 10s of billions of vectors, and superior training performance for LLMs and VLMs. The future of multimodal AI applications and models is being built on Lance today.