TwelveLabs raises $100M Series B from NEA and Amazon to build video-AI models

TwelveLabs, a video-intelligence startup with offices in San Francisco and Seoul, said on July 1, 2026, that it raised $100 million in Series B funding. Venture firms NEA and NAVER Ventures co-led the round, with Amazon among the participating investors.

The round lands as investors keep paying up for AI systems built specifically for video rather than repurposed text-and-image models. Other backers include Radical Ventures, Korea Investment Partners, Index Ventures, Quadrille Capital and Red Bull Ventures. TwelveLabs said the money will go toward its Marengo and Pegasus video models and its Video Cognition System, plus new offices in New York and London.

The company describes Marengo 3.0 as a multimodal video-embedding model and Pegasus 1.5 as a model that converts video into structured data such as scene boundaries and on-screen entities. Together, TwelveLabs says, they let developers search and analyze footage the way text search works over documents. “The world does not happen in text. It happens in motion,” TwelveLabs co-founder Jae Lee wrote in the announcement.

The pitch puts TwelveLabs among a set of specialists building video-native models rather than adapting general-purpose systems, in a market where hyperscalers are also investing. Amazon’s presence on the cap table ties a cloud provider’s interests to a company whose models are compute-hungry to train and run; an AWS executive said in the announcement that TwelveLabs has been serving customers on Amazon Bedrock for more than a year.

One caveat: the model capabilities described here are TwelveLabs’ own account, not independently benchmarked in the announcement. Performance claims for video-understanding systems are hard to verify without shared evaluations, and the company did not publish comparative results alongside the raise. What this does not prove: that the Marengo and Pegasus models outperform rival video models, a claim the company did not substantiate with third-party benchmarks.

The test now is commercial, not technical. Video AI has plenty of demos; the $100 million buys TwelveLabs time to show that enterprise customers will pay for search and analysis over their footage at scale.