Related: Anthropic runs Claude on NVIDIA GPUs for the first time in Microsoft Azure.
TwelveLabs announced on July 1 that it raised 100 million dollars in Series B funding, co-led by NEA and NAVER Ventures, with Amazon joining an investor group that also includes Radical Ventures, Korea Investment Partners, Index Ventures, Quadrille Capital, and Red Bull Ventures. Amazon is not buying the company. It is one investor among several, though its involvement carries more strategic weight than the dollar figure alone suggests, because the round also formalizes a deeper cloud partnership between TwelveLabs and AWS.
Founded in 2021 by CEO Jae Lee, TwelveLabs builds AI models designed specifically to understand video rather than adapting text based systems to handle it after the fact. Its Marengo 3.0 model, released late last year, functions as what the company calls the world’s most powerful video embedding model, parsing sound, speech, and motion across time into something searchable. A companion model, Pegasus 1.5, turns that raw material into structured data, tracking scenes, entities, and context so other systems can reason over it. “Language is downstream of understanding. Video is the data understanding has to answer to,” Lee said in the funding announcement, describing the company’s original, contrarian bet that video rather than text would end up being the harder and more valuable frontier for AI to crack.
That bet is why the Amazon relationship matters beyond the funding total. AWS is now TwelveLabs’ preferred cloud provider under a multiyear agreement, and the two companies say they are optimizing TwelveLabs’ video inference workloads specifically for AWS Trainium, Amazon’s custom AI chip line. New TwelveLabs models will also launch first on AWS, and both Marengo and Pegasus are already distributed through Amazon Bedrock. “Their models have been delivering real value to customers on Amazon Bedrock for more than a year, and as they scale their video cognition system on AWS infrastructure, including our purpose built Trainium chips, we’re excited to deepen our partnership,” said Jason Bennett, AWS’s vice president and global head of startups and venture capital.
Video has been one of the more stubborn problems in applied AI. Text and, more recently, static images have had years of model development behind them, but most of the world’s video, by some estimates upward of 90 percent of all data generated, remains effectively unsearchable and unused by enterprises, sitting in archives nobody can query. Feeding entire video libraries into a model’s context window is still too expensive for most companies to justify, and treating video as a static, indexed database strips out the reasoning that makes it useful. TwelveLabs is trying to build the layer in between, a system that gets more capable the more footage it processes rather than starting from zero on every query.
For Amazon, TwelveLabs is a small but pointed data point in a much larger effort. Nvidia chips still dominate AI training and inference workloads industry wide, and every major cloud provider building custom silicon needs real customers proving that alternative isn’t just cheaper, it actually works at production scale. TwelveLabs’ commitment to run its video inference on Trainium gives AWS exactly that kind of proof point in a domain, video, that is computationally demanding and increasingly commercially important across media, advertising, security, and government use cases. It is a modest check by Amazon’s standards, but it buys AWS a flagship customer in a category that is only going to get more competitive as more of the world’s untapped data turns out to be moving pictures rather than words.
Source: TwelveLabs press release via GlobeNewswire (July 1, 2026), and Bloomberg. Credit: TwelveLabs / AWS.
Your First 10 AI Skills
10 practical AI skills, copy-paste prompts and a 7-day plan to start using AI with confidence.
Download the guide →