
LanceDB has raised $57.5 million in a Series B round to grow its multimodal lakehouse, a data platform built for teams that train AI models on video, images and sensor data. The LanceDB funding brings the company’s total raised to $98.5 million and puts fresh money behind a problem most AI users never see: getting huge, messy training datasets ready fast enough.
Bessemer Venture Partners and Innovation Endeavors led the round. Theory Ventures, Dimension, Swift Ventures, CRV and Y Combinator also joined.
What the LanceDB Funding Will Pay For
The company says the new money will help it support AI research teams across the full cycle of improving a model. That means storing, curating, enriching and training on multimodal data in one system instead of many.
LanceDB plans to invest in feature engineering, where a team writes a data transformation once and the platform applies it across the whole dataset. It also wants to improve data loading and caching so large GPU clusters stay busy during training runs.
Planned focus areas
- End-to-end support for research teams, from raw data to training
- Feature engineering that backfills old rows and processes new data automatically
- Faster data loading and caching for big training jobs
- Workloads such as video generation, world models and robotics
Why Multimodal AI Data Is So Hard to Manage
Training a modern model often means juggling video, audio, images, 3D sensor scans and labels all at once. Chief executive Chang She told SiliconANGLE that research teams often spread their data across different systems and lose weeks or months assembling one training set.
Some of these tables are enormous, often 10 to 20 petabytes and sometimes far larger. A self-driving car, for example, produces 3D maps of its surroundings that must be linked to labels for roads, people and objects.
LanceDB’s pitch is that the data stays where it already lives, in the customer’s own cloud storage. The platform connects to it directly instead of copying or moving it around.
How LanceDB’s Platform Works
The product is built on Lance, an open-source data format designed for AI. Unlike older analytics formats built to scan whole tables, Lance can quickly fetch and update individual records, which helps when searching and preparing massive training sets.
The company shared these figures about what the platform can handle today:
| Metric | Figure |
|---|---|
| Series B amount | $57.5 million |
| Total funding to date | $98.5 million |
| Requests per second | 130,000+ at under 210ms P95 latency |
| Vectors in a single table | Over 40 billion |
| Open-source downloads | Over 120 million |
| Revenue growth since Series A | More than 10x year over year |
LanceDB Enterprise has also added a memory-accelerated write path for real-time ingest, distributed indexing and managed background compaction. The company says this lets teams work with up to 100 billion multimodal rows while several researchers use the same table at once.
Who already uses it
LanceDB lists Netflix, Microsoft AI, Runway, Luma AI and Midjourney as customers. Runway’s CTO said the team started with the Lance format in 2024 and later became an enterprise customer, while an Nvidia technical lead said the company curated about a million hours of video with Lance for its Cosmos world models.
The round follows other recent bets on AI data infrastructure, such as Edelbyte’s $70M raise. It also reflects how valuable good training data has become, a trend we covered when AI labs started buying startup data from shuttered companies.
What This Means for You
You probably won’t use LanceDB directly unless you build AI models. But the tools that prepare training data shape how quickly the AI video, image and robotics products you use can improve.
- If you run an AI or data team, the open-source Lance format is free to test before considering the enterprise product.
- If you work with video or sensor data, a single system for search and curation could cut weeks from each training cycle.
- If you follow AI investing, data infrastructure keeps drawing large rounds as model builders look for an edge.
For more tools in this space, see our guide to data analytics tools and our latest AI news coverage.
Sources
Primary source: LanceDB’s Series B announcement. Additional reporting from SiliconANGLE.
Conclusion
LanceDB’s $57.5 million Series B is a bet that the next big bottleneck in AI is data, not just compute. The company already counts major AI names as customers and says revenue has grown more than tenfold in a year. With the new funding, it wants to handle more of the work that happens before and during training. If it delivers, model builders could spend less time wrangling petabytes and more time improving the AI tools people use every day.


