LanceDB has raised $57.5 million in a Series B led by Bessemer Venture Partners and Innovation Endeavors, announced November 17, 2026. The round brings the company's total funding to $98.5 million since its Series A in June 2025, and adds Theory Ventures, Dimension, Swift Ventures, CRV and Y Combinator to its cap table.
LanceDB builds what it calls a multimodal AI lakehouse: a system for organizing, curating and serving the video, image and sensor data that AI models train on, including the kind of footage physical AI and robotics systems need to learn from.
What LanceDB builds
The company's core product is an open-source data format called Lance, which it says has been downloaded more than 120 million times, alongside a commercial LanceDB Enterprise platform built on top of it. The format is designed to handle large-scale multimodal datasets efficiently, letting AI teams retrieve and train on mixed video, image and sensor data without the performance penalties of older data formats built mainly for text and tables.
LanceDB counts Netflix, Microsoft AI, Runway, Luma AI, Midjourney and Nvidia Cosmos among its customers. The Nvidia Cosmos relationship is notable because Cosmos is Nvidia's platform for generating synthetic training data for physical AI and robotics, putting LanceDB's infrastructure directly in the pipeline that feeds robot and autonomous-vehicle models.
Why the money matters
LanceDB says the new capital will go toward expanding the platform's capabilities across the broader model-development workflow: training-data management, feature engineering, experiment execution and data loading. Rather than just storing data, the company wants to own more of the pipeline between raw footage and a trained model.
That expansion comes as demand for multimodal training data has surged alongside interest in robotics and embodied AI, where models need far more video and sensor data than text-only large language models. Companies building those models are running into the same bottleneck: existing data infrastructure wasn't built for petabytes of video and lidar.
Investors and competitors
Bessemer Venture Partners and Innovation Endeavors led this round; LanceDB's Series A in June 2025 brought in backers including CRV and Y Combinator, both of whom returned for the Series B. The company operates in a field that includes vector databases like Pinecone and Weaviate, though LanceDB positions itself more broadly as a lakehouse for raw multimodal data rather than purely a vector search tool.
What to watch next
The test for LanceDB is whether robotics and physical AI labs become a durable growth driver rather than a side benefit of serving media and generative-AI customers. With Nvidia Cosmos already in its customer list, expect LanceDB to lean harder into pitching itself as default infrastructure for teams training robots and autonomous systems on real-world footage.
