TwelveLabs raises $100m Series B for agentic video intelligence
TwelveLabs has closed a $100 million Series B to accelerate development of what it describes as a full-stack agentic video intelligence platform. The round was co-led by NEA and NAVER Ventures, with participation from Amazon, Radical Ventures, Korea Investment Partners, Index Ventures, Quadrille Capital, and Red Bull Ventures. The company said proceeds will fund research and development, expand headcount in San Francisco and Seoul, and support new offices in New York and London.
The raise marks a strategic inflection for TwelveLabs, which is moving beyond standalone video understanding models towards an integrated architecture that combines perception, knowledge, and reasoning into a single persistent system. Chief executive and co-founder Jae Lee said the company's thesis, held since founding, is that "the substrate of machine intelligence is recorded reality in motion, not language," and that the new funding would allow TwelveLabs to take that thesis from foundation models to a production-scale video cognition stack.
The platform
TwelveLabs' current model suite centres on two components. Marengo 3.0, released late last year, is the company's video embedding model; it converts raw footage into a semantic representation that can be searched and queried at scale by covering audio, speech, motion, and visual content across time. Pegasus 1.5, the more recently released counterpart, converts video into structured data: scene boundaries, named entities, temporal segments, and semantic context that downstream systems can act on. Both models are available through Amazon Bedrock and TwelveLabs' own API.
The new agentic layer builds persistent memory across an indexed video library, so the system compounds in analytical capability the more footage it processes, rather than resetting with each query. TwelveLabs also launched Rodeo, its first application-layer product, earlier in July, representing the company's first move up the stack toward no-integration consumer and enterprise tooling.
The AWS relationship goes beyond the equity investment. The two companies have signed a multiyear commitment under which TwelveLabs will optimise its video inference workloads on AWS Trainium chips, and new TwelveLabs models will launch on Amazon Bedrock first. Jason Bennett, VP and Global Head of Startups and Venture Capital at AWS, said the partnership reflects "a shared vision for building the intelligence layer for video at scale."
Market context
Video AI is a rapidly expanding category, though one that has historically been dominated by narrow, task-specific models for surveillance, content moderation, and sports analytics rather than general-purpose reasoning systems. Hyperscalers offer video analysis capabilities within their broader AI services, while a number of well-funded startups are pursuing multimodal understanding at varying points in the stack. TwelveLabs is positioning itself as a full-stack alternative that controls its own perception and reasoning layers, rather than composing third-party components.
The enterprise appetite for video understanding is being driven partly by the scale of the problem: the company cites estimates that video accounts for upwards of 90% of the world's data, yet the majority remains unsearchable and unanalysed. Verticals the company has named as active growth areas include media and entertainment, advertising, security, sports, automotive, and government, where TwelveLabs says it is already working with public-sector agencies on mission-critical workflows.
The involvement of both Amazon and NAVER Ventures as co-lead investors is notable from a strategic standpoint. NAVER, the South Korean internet group, has significant video and media assets; Amazon brings both cloud infrastructure and a distribution channel through Bedrock. Neither investor disclosed the valuation at which the round was priced. With the Series B closed, investors and enterprise buyers will be watching for published benchmark comparisons for Marengo 3.0 and Pegasus 1.5, alongside named customer deployments that can validate the compounding-intelligence thesis at production scale.