LanceDB said Nov. 17 that it raised $57.5 million in a Series B round co-led by Bessemer Venture Partners and Innovation Endeavors, with Theory Ventures also participating. The startup builds a multimodal data platform around Lance, its open-source data format.
Total funding now stands at about $95.5 million, based on the company's disclosed rounds: an $8 million seed, a $30 million Series A and the new $57.5 million Series B. The company announced its previous round, a Series A, on June 24, 2025, roughly 17 months earlier.
A bet on the data underneath AI
LanceDB is aiming at the layer that sits beneath AI training for video, images, robotics and agents. "There's a missing data layer for AI today, and that's what we're trying to build," co-founder and CEO Chang She told Kyt Dotson of SiliconANGLE.
Davis Treybig of Innovation Endeavors wrote that he confirmed his firm's role as co-lead alongside Bessemer, and that he had followed She and the development of Lance for years. Bessemer published its own investment note, titled "The data layer for multimodal AI," co-authored by Mike Droesch, Eric Kaplan, Libbie Frost and Jason Scheller.
Who is backing the company
Tomasz Tunguz, a general partner at Theory Ventures, wrote that his firm partnered with the founders at the seed stage, led the Series A and joined the Series B. SiliconANGLE reported an $8 million seed round on May 15, 2024.
RunwayML and Databricks Ventures took part in the June 2025 Series A. The available reporting does not confirm that either joined the new round. The same is true of Wayfinder Ventures, which lists a 2022 investment in the company.
LanceDB did not disclose a valuation in the materials reviewed, and Captables noted that revenue figures were also absent. For a company raising at this scale, those gaps leave an open question about how much commercial traction sits behind the enthusiasm.
Customers and the wider ecosystem

Several companies presented Lance-based systems at LanceDB's Reverie event on Nov. 5, among them Runway, Exa, Adobe, Luma and NVIDIA's Cosmos team. According to the Series A announcement, Runway also uses the commercial Enterprise product.
LanceDB's funding announcement carried customer statements. Kamil Sindi, Runway's chief technology officer, said, "A training example at Runway doesn't stay still." Max Li, technical lead for NVIDIA Cosmos, said, "Physical AI is entering a new scaling regime." Both appear in the announcement as excerpts; whether longer statements exist could not be confirmed.
Jan van der Vegt, who leads Exa's index team, offered a more concrete picture in a Reverie presentation. "These column backfills allow us to keep the old data in place and only write what is new," he said.
A case study hosted by the company says ByteDance's Volcano Engine cut its training pipelines from seven days to one. The case study is undated and has not been independently verified.
The robotics push

Hugging Face's LeRobot v0.7.0, released Nov. 5, reads natively from Lance tables, according to its release post. It keeps original MP4 files as blobs and supports in-place frame reads and global shuffle, a method of randomizing training data across an entire dataset.
Innovation Endeavors traces Lance's origins to two backgrounds: Lei Xu's work on autonomous-vehicle data infrastructure at Cruise, and She's video and machine-learning infrastructure work at Tubi.
What comes next
LanceDB's most recent newsletter, dated Nov. 12, covered the LeRobot integration, support for ASOF joins (which match records by nearest timestamp) contributed upstream to the DataFusion project, and an LLM-pretraining example. None of the company's announcement or the investor posts reviewed specify how it plans to spend the new capital.
