Twelve Labs just raised $100 million, and the real story isn't the money. It's the chips Amazon wants underneath video AI.
Bloomberg reported on July 1, 2026 that Twelve Labs Inc., the San Francisco startup building AI models that search and understand video, is raising a $100 million Series B round co-led by New Enterprise Associates and Naver, with Amazon.com among the backers. Radical Ventures, Index Ventures and Korea Investment Partners also joined. The company was founded in 2021 by Jae Lee and four co-founders, and it had already drawn money from Nvidia, Intel and Samsung.
I wouldn't read this as just another venture round. The sharper part of the Bloomberg report is the infrastructure deal sitting beside it: AWS signed a multiyear contract to host Twelve Labs' workloads on Amazon's Trainium chips, and new Twelve Labs models are due to launch on AWS for developers building video search and analysis tools. That's cloud distribution, yes. It is also Amazon putting one of the more visible video AI startups onto its own silicon instead of leaving every hard workload to Nvidia.
Lee, who worked as a data scientist for Korea's Ministry of National Defense before starting the company, built Twelve Labs around a problem the rest of the AI boom has treated as secondary. Text is easier to index. Images are easier to label. Video is hours of motion, speech, objects, faces and context stacked on top of each other. Twelve Labs' pitch is simple enough for any customer to understand: search a video library the way you'd search a document, then find the exact clip where a product appears, an event happens, or a safety incident gets buried in footage nobody wants to scrub by hand.
That is why Amazon's role matters. Amazon doesn't need to own Twelve Labs to win something here. It needs companies like Twelve Labs running inference on Trainium, because every useful workload on AWS silicon is a workload Amazon doesn't have to route through Nvidia's most expensive hardware. Trainium2 chips already power Anthropic's Project Rainier cluster in Indiana. Twelve Labs gives AWS a cleaner story in a different corner of AI: not chatbots, not code assistants, but video understanding.
The tension is hard to miss. Nvidia backed Twelve Labs when it was smaller and less proven. Now the startup is raising a serious round while signing a major compute arrangement with the cloud company trying to reduce the industry's dependence on Nvidia GPUs. Twelve Labs isn't walking away from Nvidia hardware everywhere, and the article shouldn't pretend it is. But the new capacity, the developer rollout and the Amazon money all point in the same direction.
If you're a founder building on AI infrastructure, this is the part to watch. Multimodal model companies have become public proof points for chip strategy. A benchmark slide can tell you Trainium is competitive. A production customer can show you someone is willing to build on it. When Amazon can point to Twelve Labs models running through AWS and reaching developers through its platform, that argument becomes much easier to sell to the next enterprise buyer.
The product still has to earn the money. Twelve Labs sells enterprises on a narrow, useful promise: index the video once, then let staff ask for moments in plain language instead of tagging everything manually. Media archives, retail security teams and sports organizations all know that pain already. They don't need a lecture on multimodal AI. They need the system to find the right frame quickly, without turning every query into a research project.
Frankly, that's the real test for both companies. If the $100 million helps Twelve Labs make video search fast and accurate at enterprise scale, Amazon gets a stronger Trainium case and Twelve Labs gets a business beyond demos. If it doesn't, AWS is just hosting another well-funded AI startup with expensive compute and a story that sounded better on announcement day.
What happens next is easy to measure. Twelve Labs said its new models are rolling out on AWS for developers building video search tools. The funding round is current. The useful question is whether working applications ship on top of it before the year is out.
Also read: Oracle Just Handed Wall Street Its Best Case Against the AI Spending Boom | Anthropic Gets Its Most Powerful AI Models Back After an 18-Day Fight With Washington | Tech and Finance Are Cutting Jobs by the Thousands as AI Spending Climbs
