Hot AI News

LanceDB raises $57.5M after revenue grew tenfold in a year

LanceDB, maker of a multimodal lakehouse built on the open Lance format, closed a $57.5M Series B led by Bessemer and Innovation Endeavors, bringing total funding to $98.5M.

LanceDB raises $57.5M after revenue grew tenfold in a year
Image: LanceDB

LanceDB, the company behind the open Lance data format, has raised a $57.5M Series B led by Bessemer Venture Partners and Innovation Endeavors, according to the funding news as it was reported on November 17 and the company’s own announcement post. The money goes into its multimodal lakehouse for AI training data.

What the round pays for

LanceDB’s headline for the post reads “LanceDB Raises $57.5M as AI Research Outgrows the Modern Data Stack”. The company says the round will bring its multimodal lakehouse to researchers, model builders and platform teams working at the frontier of AI. The product stores images, video and sensor data next to metadata, features and vectors, and the company lists four uses: search and retrieval, curation, feature engineering and training.

LanceDB, which makes a multimodal lakehouse for AI training, built on the open format Lance, raised a $57.5M Series B led by Bessemer and Innovation Endeavors (Kyt Dotson/SiliconANGLE)
Image: Techmeme

According to a Data Phoenix report on the announcement, Theory Ventures, Dimension, Swift Ventures, CRV and Y Combinator also took part, and total funding now stands at $98.5M. The same report lists three stated priorities for the money: broader pre-training data preparation, feature-engineering tools that apply transformations to existing and newly arriving rows, and better data loading and caching for training runs. No delivery dates were given. Valuation and other terms were not disclosed.

LanceDB, which makes a multimodal lakehouse for AI training, built on the open format Lance, raised a $57.5M Series B led by Bessemer and Innovation Endeavors (Kyt Dotson/SiliconANGLE)
Image: Techmeme
Advertisements

Numbers the company reports

All of the following figures are LanceDB’s own claims, relayed by Data Phoenix, and have not been independently audited:

  • Revenue grew more than 10x year over year since the Series A.
  • The open-source library has passed 120M downloads.
  • The system handles more than 130,000 requests per second at under 210 ms P95 latency.
  • A single table can hold hundreds of petabytes and over 40B vectors.
  • It keeps 8,000 GPUs training in parallel.

Named customers are Netflix, Microsoft AI, Runway, Luma AI and Midjourney. The release does not say how revenue is split between the open-source project and paid products, and it gives no absolute revenue figure, so the tenfold growth has no base to be judged against.

Why a new column is the selling point

The technical argument for Lance is about how training data changes. Theory Ventures, an investor in both rounds, wrote in its own post that in conventional table formats, adding a column of new embeddings or captions means rewriting the whole table, while Lance writes the new column independently. For a team re-captioning video or re-embedding a corpus, that is the difference between touching one column and rewriting petabytes.

Theory’s post also names deployments, which are again the investor’s account rather than an audited result. Applied Intuition stores hundreds of petabytes from vehicles, machines and robots and adds over a petabyte daily. NVIDIA’s SILA platform keeps media, metadata and embeddings in Lance, and the team trains its Cosmos world model on roughly a million hours of video. Exa keeps 35 petabytes in Lance tables behind an index of about 100 billion pages, and a recent re-embedding backfill ran on 35,000 Kubernetes pods.

A fast climb from the Series A

The round follows LanceDB’s $30M Series A, announced on June 24, 2025 and led by Theory Ventures, when the company launched the multimodal lakehouse suite in its enterprise product. So the company has raised $57.5M about 17 months later, nearly double the Series A, with Theory staying in and two new leads, Bessemer and Innovation Endeavors, at the top.

The pitch has also moved. The Series A was framed around search over multimodal data. The Series B banner talks about training, with the company citing training at up to 70% MFU (model FLOPs utilization) on its use-case page, a figure it has not backed with a public benchmark in the post. Theory also points to market momentum: it says physical AI startups raised about $47B in the first half of 2026, more than their combined total for 2022 through 2024. LanceDB’s bet is that the data layer for that wave runs on Lance rather than on table formats designed for analytics.

Elsewhere in the funding cycle, agent tooling has drawn similar rounds, such as Kindo’s $30M for agent management, but LanceDB sits lower in the stack, with the data that models are trained on.

Advertisements

What is still unproven

Lance is open source, so teams can try it without the company, and the paid lakehouse is the commercial layer. The announcement does not give pricing for the enterprise product. Readers should treat the performance and customer claims as vendor statements until customers or independent benchmarks confirm them.

Sign Up for the Daily AI Pulse

One email a day. All the stories that matter.

Alex CarterAI News & Big Tech Correspondent

Alex Carter writes for Aiholics, keeping readers updated on the fast-paced world of AI and Big Tech.

Leave a Reply

Your email address will not be published. Required fields are marked *

You may also like