# Twelve Labs raises $100 million as Amazon turns video search into an AWS bet > Source: > Published: 2026-07-01 11:26:46+00:00 Jae Lee's [Twelve Labs](https://www.twelvelabs.io/?ref=runtimewire) is raising $100 million in a Series B round that pulls Amazon deeper into the video intelligence stack, [Bloomberg Technology](https://www.bloomberg.com/news/articles/2026-07-01/video-search-startup-raises-100-million-from-amazon-vc-funds?ref=runtimewire) reported Wednesday. The round is being co-led by NEA Management Co. and Naver Ventures, with participation from Amazon.com Inc., Radical Ventures, Index Ventures, Korea Investment Partners and others, according to Bloomberg. Bloomberg described Twelve Labs as Nvidia-backed, but did not identify Nvidia as a participant in this Series B. The valuation and total capital raised after the round were not disclosed. That matters because this is not just another AI infrastructure check. Amazon is arriving as both investor and distribution partner. Bloomberg reported that AWS signed a multiyear contract to host Twelve Labs workloads on AWS Trainium chips, and that new Twelve Labs models will debut on AWS for developers building AI applications. The commercial terms Bloomberg described do not establish exclusivity, minimum spend, revenue share or how much of the $100 million may come from strategic capital versus traditional venture dollars. For Lee, the financing extends a thesis he has been pushing: video understanding as infrastructure rather than a consumer feature. The idea is to build AI that can search what happens inside footage, not just what someone labeled in a title, transcript or metadata field. ### The round buys compute, but also a lane into AWS Twelve Labs sits in a harder part of AI than the phrase "video search" suggests. A text search engine can index words. A video intelligence system has to map objects, motion, speech, scene changes, sound, timing and intent into a representation developers can query. Twelve Labs sells that work through models and APIs the company describes as foundation models for multimodal video understanding. Twelve Labs' own site says the platform turns raw video into searchable, AI-ready data, with use cases across media, sports, advertising, government and security. Twelve Labs says users can search libraries using natural language to locate actions, scenes, dialogue and emotions without manual tags. Those are company claims, but they clarify what Lee is selling: not a consumer search interface, but infrastructure for organizations sitting on video archives they cannot fully use. Amazon has already been close to the business. In December 2024, [Amazon said](https://press.aboutamazon.com/2024/12/generative-ai-startup-twelve-labs-works-with-aws-to-make-videos-as-searchable-as-text?ref=runtimewire) Twelve Labs was building and scaling proprietary foundation models on AWS as part of a Strategic Collaboration Agreement. Amazon said Twelve Labs used Amazon SageMaker HyperPod to train multimodal models, and said the models were available through AWS Marketplace. The new financing extends that relationship from cloud customer story to cap-table alignment. For Amazon, Twelve Labs gives AWS a specialized multimodal model partner at a time when every major cloud provider is trying to keep AI workloads on its own infrastructure. For Twelve Labs, AWS can offer more than compute: Marketplace distribution, enterprise procurement, developer access and a route into large organizations that already centralize software buying around cloud contracts. The strategic question is how much flexibility Twelve Labs keeps. Video models are compute-hungry, and startups with frontier-model ambitions often have to trade independence for infrastructure certainty. Bloomberg's report says AWS will host workloads on Trainium, Amazon's in-house AI accelerator line. That is useful if the economics work and the software stack is mature enough for Twelve Labs' models. It also ties a young model company to Amazon's chip strategy at the moment when cloud providers are trying to reduce dependence on Nvidia GPUs. ### Lee's original wedge was search inside the frame In 2022, when Twelve Labs was still early, [TechCrunch reported](https://techcrunch.com/2022/12/05/twelve-labs-lands-12m-for-ai-that-understands-the-context-of-videos/?ref=runtimewire) that Lee, described as a data scientist by training, thought it made little sense that video had become central to daily life but remained difficult to search because machines lacked context understanding. Twelve Labs raised a $12 million seed extension led by Radical Ventures at the time, according to TechCrunch. The pitch has since moved from video search to broader video intelligence. In June 2024, Twelve Labs [said it raised $50 million](https://twelvelabs.dev/blog/series-a-announcement.html?ref=runtimewire) in Series A funding co-led by New Enterprise Associates and Nvidia's NVentures, with participation from existing investors including Index Ventures, Radical Ventures, WndrCo and Korea Investment Partners. In that post, Lee wrote that Twelve Labs had spent three years building a video understanding platform driven by perceptual-reasoning research, and said AI systems need to learn from video to understand the world the way humans do. That line explains why investors keep underwriting the category despite unclear near-term economics. Video is abundant, expensive to process and operationally messy. Media companies have archives. Sports leagues have game footage. Security, public sector and enterprise customers have recordings they need to search, summarize, audit or turn into workflows. The value is obvious. The difficulty is whether models can deliver accurate retrieval and analysis at a cost customers will pay, without forcing teams into brittle tagging systems or manual review. Twelve Labs has been careful to position itself as developer infrastructure rather than a single-purpose application. Its public materials point developers to API docs, SDKs, a playground and AWS Marketplace. That is the same distribution logic behind many foundation-model startups: build the model, abstract it behind APIs, and let customers create vertical workflows on top. The challenge is that specialized infrastructure can get squeezed from both directions. General-purpose model providers keep adding video capabilities, while cloud platforms offer their own services for indexing, recognizing and analyzing media. ### What remains undisclosed The missing pieces are the ones that usually decide whether a round like this is strategic momentum or expensive runway. Bloomberg did not disclose Twelve Labs' valuation. Revenue, ARR, headcount, paying customer count, model usage and gross margin were not disclosed. Bloomberg also did not state whether the Series B has formally closed or whether the financing was still in process as of July 1, 2026. Those omissions do not weaken the core news. They define the stakes. Twelve Labs has persuaded NEA, Naver Ventures and Amazon that video understanding is becoming a large enough infrastructure category to justify another major round. Lee has also found a cloud partner willing to put AWS distribution and Trainium compute behind that bet. The next proof point is not whether video search is useful. It is whether Twelve Labs can make video-native AI cheap, accurate and easy enough that developers build on it before the biggest model labs and cloud platforms absorb the market Lee helped define.