Saturday, November 21, 2026Sat, Nov 21 Updated

Mergers, acquisitions and startup deals, and what they mean.

LanceDB Raises $57.5 Million to Build a Data Layer for AI Training

LanceDB raised $57.5 million in a Series B led by Bessemer and Innovation Endeavors. What the AI data startup sells, who uses it and where the money goes.

Detailed view of server racks with glowing lights in a data center environment.
Photo: panumas nikhomkhai / Pexels

LanceDB Inc., a startup that builds database software for training artificial intelligence models, has raised $57.5 million in a Series B round, SiliconANGLE reported. Bessemer Venture Partners and Innovation Endeavors led the round.

The new money brings the company’s total funding to $98.5 million. Theory Ventures, Dimension, Swift Ventures, CRV and Y Combinator also took part. The company did not disclose a valuation.

What LanceDB sells

LanceDB offers what it calls a multimodal lakehouse. A lakehouse joins two older ideas: a data lake, which holds huge amounts of raw files cheaply, and a database, which lets users organize and query that material. “Multimodal” means the system handles text, images, video, audio and other file types in one place.

The product sits on Lance, an open-source file format built for AI work. Lance stores structured records next to large files such as video, and it lets software pull or change single records quickly. Older analytics formats are tuned for scanning whole tables, which makes them slow for searching and preparing training sets.

The platform connects to the cloud storage a customer already uses. Co-founder and Chief Executive Chang She told SiliconANGLE that customer data is not copied or moved. “The data stays in place,” he said.

The problem it targets

Companies building self-driving cars, robots and so-called world models collect far more than text and pictures. Their data includes 3D sensor scans, map coordinates, time-series readings and labels that say what the sensors saw, such as roads, pedestrians or trees.

According to SiliconANGLE, a single table in many large training jobs averages 10 to 20 petabytes, and some approach 80 to 100 petabytes. Research teams often spread this data across several systems and spend weeks or months moving, cleaning and assembling it before a training run. She said that delay comes from a missing data layer for AI, which LanceDB wants to supply.

Customers named by the company

  • Runway AI Inc. moved to LanceDB’s enterprise platform as its training needs grew. Its chief technology officer, Kamil Sindi, said the company began using the Lance format in 2024.
  • Netflix Inc. built a media data lake on the technology for content creation.
  • Microsoft Corp. uses the platform as a data layer to train models on tens of billions of rows of image and code data.

Sindi said Runway picked Lance because, at the time, no other option handled unstructured data, added columns, distributed reads and ad-hoc SQL queries together.

What this means for the AI data market

The round sends investor money to the storage and preparation layer that sits under AI models. LanceDB’s pitch to buyers is speed: training data that is ready sooner means more experiments and quicker releases. She called this the “AI flywheel” and said a faster cycle lets research teams reach higher quality models.

For the investors, the raise is a bet on growth in video generation, world models and robotics, the fields the company names as its next targets.

Where the money goes

She said LanceDB will spend the funds on expanding the platform and shortening the model development cycle. Planned work covers large-scale data curation, feature engineering and data loading fast enough to keep thousands of graphics processing units busy during training.

The company also plans deeper support for video generation, world models and robotics. Researchers in those fields handle growing volumes of first-person sensor data and synthetic data, and LanceDB wants that material managed on its platform.

Source: SiliconANGLE