TwelveLabs, a startup that builds AI models for understanding video, has raised 100 million dollars in a Series B round, announced on July 1. The round was co led by NEA and NAVER Ventures, and its investor list is the first thing worth noting, because it includes Amazon alongside Index Ventures, Radical Ventures, Korea Investment Partners, Quadrille Capital, and Red Bull Ventures. TwelveLabs is not a new name in this space. It is known for models that let software search and describe the contents of video, the kind of technology that can find every moment a particular object, action, or person appears across a huge archive of footage, which is a genuinely hard problem that text focused AI does not solve on its own.
What the money is for matters more than the amount. TwelveLabs is using the round to push beyond video understanding, which is essentially very good search and tagging, toward what it calls a full stack agentic intelligence system for video. The idea is to fold three things that are usually separate, perception, knowledge, and reasoning, into a single architecture aimed at video. Where an understanding model can tell you that a clip contains a forklift and a spill, an agentic system is meant to reason about the scene, connect it to what it already knows, and take or recommend an action. The company has taken to describing the goal as video superintelligence, which is marketing language, but the underlying shift, from labeling what is in a video to reasoning about it, is a real and difficult step.
The Amazon side of the announcement is arguably the bigger story. TwelveLabs named Amazon Web Services as its preferred cloud provider and signed a multiyear strategic commitment that goes well beyond renting servers. The two companies said they would optimize TwelveLabs video inference to run on AWS Trainium, Amazon's own AI chips, rather than relying solely on Nvidia hardware, and that new TwelveLabs models would launch first on AWS. That is a meaningful signal in two directions. For Amazon, backing a video AI specialist and running it on Trainium is part of its effort to prove its in house silicon can handle serious AI workloads and to secure a foothold in a modality it does not own. For TwelveLabs, deep alignment with one cloud is a bet that guaranteed compute and distribution are worth more than staying neutral.
Step back and the raise fits a larger pattern, which is that video is becoming its own distinct battleground in AI. Most of the visible progress of the last few years has been in text and, more recently, in generating images and clips. Understanding video at scale, the far less glamorous problem of making sense of the enormous volume of footage that security cameras, media companies, factories, and phones produce every day, has lagged behind, precisely because it is hard and expensive. A funding round like this, with both venture money and a cloud giant behind it, is a bet that the next valuable layer of AI is not another chatbot but systems that can watch and reason about the physical and recorded world. That connects video understanding to the same broad shift toward agents and toward AI that acts, rather than only answers.
The caveats are worth keeping in view. One hundred million dollars is a serious sum but a modest one next to the billions flowing to frontier model labs, and it signals a strong specialist rather than a new giant. The word superintelligence should be read as ambition, not description, since reliably reasoning about messy real world video is still an unsolved problem, and enterprise buyers will judge the product on accuracy and cost, not on slogans. There are also the familiar questions that follow any leap in machines understanding video, about surveillance, consent, and who gets to point these systems at whom, which grow more pressing as the tools get better. But as a market signal the round is clear enough. A recognized video AI company just secured a large war chest, a marquee list of backers, and a cloud partner willing to commit its own chips, all pointed at the same idea, that teaching AI to genuinely understand and reason about video is about to matter a great deal more than it has.