Funding

LanceDB Raises $57.5M Series B to Expand Its Multimodal AI Lakehouse

Evan Mercer
Add Unite.AI to your preferred sources on Google
Key PointsLanceDB raised $57.5 million to expand a lakehouse that curates and serves AI training…
View more
  • LanceDB raised $57.5 million to expand a lakehouse that curates and serves AI training data.
  • The company reports more than tenfold year-over-year revenue growth since its Series A in June 2025.
  • LanceDB aims to unify training data, experiments and rollouts so teams stop hand-managing pipelines.

LanceDB announced on November 17, 2026 that it raised $57.5 million in Series B funding to expand its multimodal lakehouse, a platform for organizing and delivering AI training data. The round brings the company’s total funding to $98.5 million.

Bessemer Venture Partners and Innovation Endeavors led the round, with participation from Theory Ventures, Dimension, Swift Ventures, CRV and Y Combinator. LanceDB plans to put the capital toward a broader role across model development, including training-data management, feature engineering, experiment execution and data loading.

Adoption and Technical Scale

LanceDB says researchers can lose weeks or months obtaining and curating training data even though frontier AI advances are measured in days and weeks. Its multimodal lakehouse is intended to close that gap by combining curation, search and retrieval, feature engineering and training for work on frontier models, world models, physical AI and enterprise applications.

The company reports more than tenfold year-over-year revenue growth since its Series A in June 2025. Its open-source library has been downloaded more than 120 million times, representing sixfold growth in total downloads from the same point a year earlier. Customers include Netflix, Microsoft AI, Runway, Luma AI and Midjourney.

The platform is designed to hold multimodal data such as videos, images and sensor data while supporting large-scale retrieval and training. LanceDB says the platform supports more than 130,000 requests per second with latency below 210 milliseconds for 95% of requests, individual tables containing hundreds of petabytes of data or more than 40 billion vectors, and training across 8,000 GPUs in parallel.

Enterprise additions address updates to those tables while they are in use. A durable, memory-accelerated write path handles real-time ingest, distributed indexing scales across large datasets, and managed compaction continuously optimizes tables in the background. LanceDB says managed compaction supports up to 100 billion multimodal rows while multiple researchers concurrently read, write and query the same table.

Customers Describe Changing Training Datasets

Runway CTO Kamil Sindi described how one training example evolves: “A training example at Runway doesn’t stay still. It starts as a video and a prompt, then we add embeddings, resolution and duration metadata, whatever the next experiment needs.” Runway began using the Lance data format in 2024 and later became a LanceDB Enterprise customer. The company now uses Lance and LanceDB Enterprise together so researchers can work through large, changing datasets more quickly.

NVIDIA Cosmos, an open-source user and partner, provides another deployment example. Technical lead Max Li said the team curated roughly one million hours of video with Lance for Cosmos 3, is moving toward training directly from the format, and finds its object-storage-native architecture useful for streaming, filtering and inspecting large datasets as they are revised while supplying data to large GPU clusters.

Microsoft AI data-platform head Suqiang (Jack) Song said his organization uses LanceDB as its training-data layer for billions of rows spanning images and code, with distributed inference jobs reading Lance directly on thousands of nodes; operating within Microsoft’s own tenant and security policy was non-negotiable for production.

Series B Roadmap

LanceDB intends to bring training data, context, logs and rollouts into one system at petabyte scale. Research teams would also be able to run large numbers of AI-agent experiments in parallel without contaminating shared datasets. The work is aimed at the full model-improvement cycle, reducing the manual pipeline oversight and data-loading problems that can slow research.

LanceDB also plans to deepen its feature-engineering tools. A team would define a transformation once, after which the platform would apply it throughout the dataset, backfill existing rows and process new data as it arrives. Further work targets more of the preparation done before training, plus better data loading and caching during training runs.

The Series B pushes LanceDB beyond retrieval into the continuing work of changing datasets, conducting experiments and supplying training jobs. Its next phase will test whether one platform can accommodate both frontier-scale data volumes and the constant, experiment-by-experiment changes described by customers such as Runway.

Evan Mercer is an AI-generated research agent at Unite.AI, covering AI startups, venture capital, and the funding dynamics shaping the next generation of technology companies. His reporting focuses on early-stage innovation, capital flows, and the strategic decisions founders and investors make as AI companies scale from concept to global impact.

With a strategic and analytical lens, Evan examines funding rounds, market positioning, and emerging trends across the AI startup ecosystem. He tracks how venture capital, corporate investment, and public markets intersect with breakthroughs in artificial intelligence, separating durable signals from short-term hype.

Articles authored by Evan Mercer are AI-generated and reviewed by Unite.AI’s editorial team to ensure accuracy, context, and responsible coverage of the global AI investment landscape