LanceDB announced a $57.5 million Series B on November 17, 2026, led by Bessemer Venture Partners and Innovation Endeavors, SiliconANGLE reports. The company plans to expand its platform for organizing and preparing AI training data—the less glamorous work that has to happen before a model gets to do anything impressive. Source
The round brings LanceDB’s total funding to $98.5 million. Other participants included Theory Ventures, Dimension, Swift Ventures, CRV and Y Combinator, according to SiliconANGLE and The SaaS News. Source Source
A video and a prompt are just the start
Co-founder and CEO Chang She told SiliconANGLE that research teams can lose weeks or months assembling a training dataset because its components sit in different places. Source
“There’s a missing data layer for AI today, and that’s what we’re trying to build,” She said in SiliconANGLE’s interview. Source
LanceDB calls its answer a multimodal lakehouse. It combines storage for large amounts of different data—such as images, video and structured information—with database tools for organizing, searching and preparing it. The commercial platform is built on Lance, an open-source data format designed for AI workloads. Source
In comments quoted by SiliconANGLE, Runway CTO Kamil Sindi described training examples that begin with a video and a prompt, then accumulate additional information, including resolution and duration metadata, as experiments demand it. Even a training example has a growing to-do list. Sindi said Runway began using Lance in 2024 and later became a LanceDB Enterprise customer. Source
Keeping GPUs fed is the goal
She said the funding would support improvements to large-scale data curation, feature engineering and loading, aiming to keep thousands of graphics processing units, or GPUs, efficiently supplied during training. LanceDB also plans deeper support for video generation, world models and robotics. Source
The SaaS News also reports planned work on caching and data-preparation tools to make better use of GPUs during training. Source
Neither report gives a delivery schedule for the planned improvements or independent benchmarks for training speed, costs or GPU utilization. Source Source
Comments