AlphaMaven AlphaMaven

LanceDB: The data layer for multimodal AI

Reported by Bessemer Venture Partners
Nov 17, 2026· 2 min read
Save Post LinkedIn Email
Key Takeaways
  • • Bessemer Venture Partners is co-leading LanceDB's $57.5M Series B alongside Innovation Endeavors, with continued support from CRV, Theory Ventures, Dimension, Swift Ventures, and Y Combinator, according to Bessemer's published investment note.
  • • LanceDB's open-source Lance format is positioned as a storage layer for multimodal data (images, video, sensor streams, embeddings), analogous to what Parquet does for analytics tables.
  • • LanceDB reports tens of millions of monthly downloads, up 6x within a year, and cites frontier labs, WorldLabs, Applied Intuition, ByteDance, and Netflix as users.
  • • The commercial LanceDB Enterprise platform adds distributed indexing, high-throughput ingestion, training data loaders, and feature engineering, deployable inside customers' own cloud or on-prem environments.
Summary
Bessemer Venture Partners has co-led a $57.5M Series B for LanceDB, the company behind the open-source Lance file format, table format, and catalog spec. Bessemer is leading the round alongside Innovation Endeavors, with continued support from CRV, Theory Ventures, Dimension, Swift Ventures, and Y Combinator. The investment note frames the round as funding infrastructure for multimodal AI, where training and inference data increasingly consists of images, video, audio, and sensor streams rather than rows of numbers and text.

According to the Bessemer write-up, LanceDB is led by CEO Chang She, a co-author of pandas who previously ran ML engineering at Tubi, and CTO Lei Xu, whose background includes HDFS at Cloudera and ML infrastructure for Cruise's self-driving fleet. Bessemer says the founders concluded the Parquet format needed to be rebuilt from first principles for multimodal workloads.

Lance stores raw images or video alongside metadata, features, and vectors, and supports native vector search and SQL. Bessemer notes that a team can add an embeddings column to a petabyte-scale table without rewriting it. LanceDB, the company and database built on the format, reports users downloading the library tens of millions of times monthly, up 6x within a year. Frontier labs reportedly use it for pre-training to consolidate separate systems into a single Lance-based platform, while WorldLabs, Applied Intuition, and ByteDance use it for post-training and Netflix for production retrieval and ML operations.

The commercial product, LanceDB Enterprise, adds distributed indexing, high-throughput ingestion, training data loaders, and feature engineering, and runs inside the customer's own cloud or on-premises environment. Bessemer argues that, as in earlier data infrastructure cycles, an open-source project paired with a company able to scale it is likely to win the AI data layer. The figures above are company and investor claims; the article does not provide independent audited usage or revenue data.

Go deeper than the headline

Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.

Start your free 7-day trial

Sources

1 outlet

AlphaMaven summarised this story from the reporting below. Each link opens the original article.

Bessemer Venture Partners LanceDB: The data layer for multimodal AI