Amazon just told the AI industry how it plans to win the video race: by backing somebody else's model and pushing its own chips into the deal.
On July 1, TwelveLabs, a video AI startup with offices in San Francisco and Seoul, said it had raised a $100 million Series B. The round was co-led by NEA and Naver Ventures, with Amazon, Radical Ventures, Korea Investment Partners, Index Ventures, Quadrille Capital and Red Bull Ventures also participating. The cheque matters. The cloud deal matters more.
Bloomberg reported the same day that Amazon's investment came with a deeper commercial arrangement: AWS would become TwelveLabs' preferred cloud partner, and the startup's next generation of models would launch first on AWS Trainium chips rather than Nvidia's hardware. TwelveLabs' own announcement confirmed Amazon's participation and the $100 million round, though it framed the raise around video superintelligence rather than the chip politics underneath it.
That distinction is important. If Bloomberg's reporting is right, this isn't just another venture round with a famous strategic investor attached. It ties a startup's model roadmap to Amazon's AI infrastructure business.
You don't have to squint to see why that matters. Amazon has poured billions into Anthropic over the past two years, and it has been building Trainium as its answer to Nvidia's grip on AI compute. Getting a video foundation model startup to launch first on Trainium, and to say so publicly, gives Amazon something it badly needs: evidence for cloud customers that its chips can carry serious AI workloads.
This is how Big Tech buys market share now.
Picking a Side #
TwelveLabs isn't a tiny lab anymore. The company has grown to roughly 178 employees, according to TechFundingNews, up from about 58 a year earlier. It raised a $50 million Series A in 2024, co-led by NEA and Nvidia's venture arm, NVentures. That earlier Nvidia connection makes the AWS arrangement sharper. TwelveLabs has taken money from both chip camps. For its next models, it appears to have picked where they run first.
The company built its name by letting customers query raw video footage the way you'd query text. That was useful, but it was still a search pitch. Now TwelveLabs is describing something bigger: what CEO Jae Lee calls a Video Cognition System, built around perception, memory and reasoning in one architecture. In the company's July 1 announcement, Lee said TwelveLabs started five years ago from the view that video, not text, is the closest digital record of reality as it unfolds.
That is a bold claim. It also has a real technical point underneath it. TwelveLabs' Marengo model maps visual, audio, speech and on-screen text signals into one searchable representation, while Pegasus turns those representations into answers, descriptions and summaries. The company is trying to make video archives addressable at the second level - not just searchable by a filename or a transcript, let alone a caption.
Google will not politely step aside for that. Gemini already has multimodal reach, YouTube gives Google a mountain of video, and Google Cloud can sell AI tooling straight into the same enterprise accounts TwelveLabs wants. TwelveLabs is making a specialist's argument: a company built only around video can move faster on this one axis than the generalists. Amazon's bet says that argument is worth funding instead of simply copying in-house.
That's the wager.
The Race for Video Isn't Over #
The expansion plans match the ambition. TwelveLabs is opening offices in New York and London. That takes it to five cities in all: San Francisco, Seoul, Los Angeles, New York and London. Its customer map is also broadening beyond the media and entertainment clients that made video search an obvious early use case. The company's own announcement points to public services, advertising, security, sports and automotive as target areas.
That's a lot of ground for one architecture.
For readers watching the funding stack rather than the model demos, the real signal here isn't only TwelveLabs. It's Amazon. A company with the resources to build almost anything in-house chose to back a roughly 178-person startup and, according to Bloomberg, connect that bet to AWS Trainium. Text had OpenAI and Anthropic. Image had Midjourney and Stability. Video is still unsettled, and the cloud providers don't want to sit around while Nvidia decides who gets the best seat. The risk is obvious too. If TwelveLabs can make Trainium look credible for video models, Amazon gets a proof point it can show to every AI customer worried about Nvidia supply and cost. If it can't, the deal becomes another reminder that alternative AI chips need more than a balance sheet behind them. They need workloads that perform in the real world.
Also read: Google's Money Just Made Proxima Fusion Europe's Best Funded Fusion Bet • Monzo Cofounder Tom Blomfield Leaves Y Combinator to Join Anthropic's Compute Team • Valarian raises $50 million to help Europe escape America's cloud grip