✉ The Friday AI Brief: the week's 5 best AI stories, tools & comparisons — in your inbox every Friday morning.

Illustration of streams of images, video frames and sensor data flowing in organized layers into rows of GPU servers in a data center

LanceDB Raises $57.5 Million to Build the Data Layer AI Training Is Missing

November 17, 2026 — LanceDB has raised a $57.5 million Series B to expand its multimodal data lakehouse, the platform Netflix, Microsoft AI, Runway, Luma AI and Midjourney use to prepare the images, video and sensor data their models train on. Why it matters: as frontier labs race to scale training, the bottleneck is increasingly the weeks spent getting data ready — and investors are paying to remove it.

What happened

The round, announced November 17, was led by Bessemer Venture Partners and Innovation Endeavors, with participation from Theory Ventures, Dimension, Swift Ventures, CRV and Y Combinator. It brings LanceDB’s total funding to $98.5 million. The raise comes about five months after the company’s Series A last June, and the growth since then is the story: LanceDB says revenue has grown more than tenfold year over year since the Series A, and its open-source library has been downloaded more than 120 million times — six times the total of a year ago.

The problem it sells against

LanceDB is built on Lance, an open-source data format designed for AI workloads. The pitch is a single system where researchers can store, search, curate, enrich and train on multimodal data — text, images, video, audio and sensor streams — instead of scattering pieces of a dataset across separate systems and losing weeks reconciling them before a single training run. Co-founder and CEO Chang She told SiliconANGLE that research teams routinely lose weeks or months getting one whole training dataset ready, and that a dedicated data layer for AI is what has been missing.

Two design choices matter for enterprise buyers. First, the data stays where it already lives: LanceDB connects to a customer’s existing cloud or on-premises storage and manages access and preparation without copying the data into a separate system. Second, the scale is aimed at real frontier workloads. The company says the platform supports more than 130,000 requests per second, hundreds of petabytes in a single table and training runs spread across 8,000 GPUs, while individual training tables commonly run 10 to 20 petabytes and can approach 80 to 100 petabytes.

Who is already using it

The customer list is the credibility. Microsoft AI uses LanceDB as the data layer for training on tens of billions of rows of image and code data, running inside its own tenant under its own security policy. Netflix built a media data lake on the technology. Runway started using the Lance format in 2024 and has since become an Enterprise customer, and NVIDIA’s Cosmos team curated roughly a million hours of video using Lance, according to the company. The new funding goes toward the full model-improvement cycle — curation, feature engineering, experiment execution and data loading fast enough to keep thousands of GPUs fed — with a push into video generation, world models and robotics workloads.

Why it matters

For two years the AI infrastructure money chased compute: GPUs, clusters, power. LanceDB’s round is a bet that the constraint has moved one layer down, to the unglamorous work of finding, cleaning and streaming training examples at petabyte scale. A data platform that sits inside the customer’s own storage, carries an open-source format underneath it and already runs inside Microsoft, Netflix and Runway is a credible claimant for that layer — and a $57.5 million Series B five months after its Series A says its investors think the window to claim it is short.

Frequently asked questions

How much did LanceDB raise?

$57.5 million in a Series B led by Bessemer Venture Partners and Innovation Endeavors, announced November 17, 2026. Total funding to date is $98.5 million. No valuation was disclosed.

What does LanceDB actually do?

It provides a multimodal data lakehouse built on the open-source Lance format, letting AI teams store, search, curate and train on text, images, video, audio and sensor data in one system — in their own storage, without copying the data elsewhere.

Who uses LanceDB?

Named customers include Netflix, Microsoft AI, Runway, Luma AI and Midjourney. NVIDIA’s Cosmos team has also used the Lance format to curate large video datasets.

Sources: LanceDB company announcement; SiliconANGLE.

Leave a Comment

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

Get the 5 best AI tools every week

Top AI news, tools, and prompts — one short email. Free, unsubscribe anytime.

Run a newsletter of your own? Monetize and grow it with SparkLoop →

Scroll to Top