We raised $30M to build the future of AI data! 🚀 Learn about our Series A, the new Multimodal Lakehouse for unified text, image, audio & video pipelines, and powerful new enterprise features.
Lance DB's $30M Series A, Multimodal Lakehouse Launch & Product Updates
LanceDB is Now a Series A Company
Over the past year, we have witnessed the Lance columnar formatbecome the new standard for multimodal data. As of June 2025, Lance remains the fastest growing format across the data ecosystem. During this period, our open source packages have been downloaded more than 20 million times.
This milestone represents a significant validation of our vision to democratize multimodal AI development. The $30M Series A funding will accelerate our mission to build the most efficient and scalable data platform for AI applications.
As of June 24th, 2025, along with the celebration of LanceDB’s Series A, we are introducing the Multimodal Lakehouse Suite of Products into LanceDB Enterprise.
The LanceDB Enterprise offering now consists of four features: Search, Exploratory Data Analysis, Feature Engineering and Training.
The Multimodal Lakehouse represents a breakthrough in unified data management for AI applications, seamlessly handling text, images, audio, and video data in a single platform. This comprehensive solution eliminates the complexity of managing multiple data silos and provides enterprises with the tools they need to build, train, and deploy multimodal AI models at scale.
Richer full-text search capabilities: Unlock advanced FTS with boolean logic, flexible phrase matching, and autocomplete-ready prefix queries.
Blazing-fast full-text search: Optimized FTS engine now delivers P99 latencies under 50ms on 40M-row tables, lightning speed at scale.
Streamlined Kubernetes deployments: Native Helm chart support makes BYOC setups faster and easier to manage. A deployment can be up and running in a couple of hours.
Smarter vector search with tight filters: Fine-tune recall with new minimum_nprobes and maximum_nprobes controls for better results on queries with highly selective filters.
From Text to Video: A Unified Multimodal Data Lake for Next-Gen AI
Ryan Vilim from Character AI shares how their Data & AI Platform team builds self-service tools and infrastructure to power LLM training and research. He explains how they leverage data lakes, Spark, Trino, Kubernetes, and Lance to prepare, annotate, and serve massive multimodal datasets—while keeping workflows fast and researcher-friendly.
The talk also covers why Lance’s open multimodal lakehouse format fits their needs, enabling unified storage, search, and analytics at scale. Packed with practical insights on managing AI data pipelines, this session is perfect for anyone building or scaling AI systems.
Building a Data Foundation for Multimodal Foundation Models
Ethan Rosenthal from Runway delivers an in-depth exploration of the unsung heroes behind generative models: data pipelines.
Ethan shares pragmatic lessons on system design—covering topics like columnable storage formats (moving beyond tarballs and Parquet), schema evolution with LanceDB, efficient video decoding, asynchronous data loading to eliminate GPU bottlenecks, and on-the-fly augmentation for flexible experimentation.
A Thank You to Our Contributors
A heartfelt thank you to our community contributors of Lance and LanceDB OSS projects this past month:
