Twelve Labs Raises $100 Million to Build Video Understanding for AI Agents

Twelve Labs, a startup building AI models for video understanding, announced the close of a $100 million Series B funding round. The round was co-led by NEA and Naver Ventures, with Amazon also among the investors. Amazon simultaneously signed a multi-year agreement making Amazon Web Services the preferred cloud provider for Twelve Labs.

Video as a distinct data category

From the outset, Twelve Labs has built on the premise that the world doesn't unfold in text but in motion, and that treating video as an add-on to language models limits what AI can actually extract from that data. Rather than converting footage into text descriptions early on, the company is building an architecture that treats video understanding as its own category, alongside image and text generation.

The company's system rests on three pillars: perception, meaning understanding raw video without early conversion to text; memory, meaning creating persistent, searchable representations indexed over time; and reasoning, meaning answering questions that span multiple segments and time periods within a recording. The Marengo model converts visual signals, audio, speech, and on-screen text into searchable representations, while Pegasus uses that foundation to generate descriptions, answers, and summaries.

Amazon steps in as investor and partner

Amazon's involvement goes beyond capital. The company has secured a role as Twelve Labs' preferred cloud provider, with new models set to launch on AWS first, and video inference optimized for Amazon's own Trainium chips. This fits a broader pattern in which major cloud providers invest in specialized AI startups in exchange for infrastructure exclusivity, rather than simply selling them compute on standard terms.

Naver Ventures, a Korean fund affiliated with the Naver search engine, was Twelve Labs' first investor and is now co-leading the round, something a partner at the fund called the strongest possible signal of confidence in the company. The presence of Red Bull Ventures and Korea Investment Partners suggests interest in AI video extends beyond traditional tech investors, toward media and sports companies that themselves hold massive video archives.

The road to video superintelligence starts here - Jae Lee, CEO of Twelve Labs

The AI video market is heating up

Twelve Labs' round is part of a record first half of 2026 for AI startups, which according to market data raised roughly $510 billion in combined funding. Within the video segment specifically, Twelve Labs faces competition both from major players building video generation models and from specialized firms focused, like Twelve Labs, on understanding and searching existing footage rather than creating new video.

For Polish companies in the media, security, and automotive sectors that accumulate hours of surveillance footage, television production material, or vehicle test recordings, the development of such tools offers a real alternative to manually searching through video archives. Twelve Labs says its models index footage in context for up to two hours of recording at a time, allowing AI agents to answer questions spread across time, such as pinpointing exactly when a specific person or event appeared in surveillance footage.

The company plans to use the new funds to develop further generations of the Marengo and Pegasus models, as well as for geographic expansion, including new teams in New York and London tasked with growing the business beyond its existing Korean-American market.

Sources: GlobeNewswire (globenewswire.com), TwelveLabs (twelvelabs.io), Bloomberg (bloomberg.com), Tech Funding News (techfundingnews.com)