December 2025 – If you’ve been paying attention to AI lately, you know these models are hungry. Really hungry. And not just for text anymore as they’re devouring video at a scale that’s hard to wrap your head around.
According to Versos, Meta is expected to need 5 million hours of video in 2026. That doubles to 10 million hours in 2027. And that’s not a typo.
But what the industry is not talking about is the infrastructure to actually get that video to AI companies is a mess. It’s fragmented, clumsy, and raises serious questions about where that content comes from and who gets paid for it.
That’s why we’re excited to announce our investment in Versos AI, a company building the first end-to-end platform for video training data.
The problem: AI needs video, but getting it is painful
Think about what AI companies actually need to train their models on video:
They need massive libraries of content. We are talking about billions of hour’s worth. They need to search through all that content to find specific scenes (imagine trying to find “a person riding a unicycle” across millions of hours of footage). They then need the video processed in very specific ways, segmented into meaningful clips, tagged with metadata and ready to feed into training pipelines.
And critically, they need to know the content is ethically sourced and rights-cleared. Because if it’s not, they’re opening themselves up to lawsuits and PR disasters.
Right now, there’s no good solution for this. The process is manual, expensive, and risky. Open-source datasets either have terrible search capabilities or get shut down with cease-and-desist orders because they scraped content without permission.
It’s a $60 billion problem by 2028, and nobody has really solved it yet.
This is where Versos comes in: making video actually searchable
Versos started with a deceptively simple insight: what if video could be as searchable as text?
Not just searching captions or titles, but actually understanding what’s happening in the video, the scenes, the dialogue and its context. What if you could type “sunset over a mountain lake with ambient bird sounds” and actually find it?
That’s what founders Chris Keevill and Cezar Grzelak set out to build.
Versos is built around a multimodal AI architecture. When you upload video to Versos, it:
- Segments the footage into semantically meaningful 7-20 second clips based on camera movement and scene changes
- Generates vector embeddings that capture what’s happening visually, what’s being said, and the ambient context
- Creates a searchable index across all those dimensions
- Tracks the complete chain of custody which means where the video came from, who owns the rights, how it’s been processed
The bigger vision
Right now, Versos is focused on processing video for AI training. But the long-term vision is much bigger: becoming the cloud-connected content hub for creators and AI innovators.
For content creators (from major studios to individual YouTubers) Versos offers searchable storage, content delivery, and direct monetization through a marketplace where video is in massive demand.
For AI companies, it’s on-demand access to curated, rights-cleared, pre-processed video content at scale.
They’re already building relationships across the supply chain: major studios, data brokers and direct buyers.
Our investment
We’re proud to join NBIF in co-leading Versos’ oversubscribed early-stage round. Also investing alongside us are the brilliant folks at Velocity Fund and RISC Capital.
For us, this investment hits on several key themes:
- Market timing. The demand for video training data is exploding, and there’s no good infrastructure to meet it. Versos is building that infrastructure at exactly the right moment.
- Technical differentiation. Their multimodal architecture and chain-of-custody system aren’t just nice-to-haves, they’re fundamental advantages in a market that’s rapidly professionalizing.
- Team execution. Chris and Cezar have done this before. They know how to build products, land enterprise customers, and scale businesses. They’ve already proven they can execute with real revenue and major partnerships.
- Ethical foundation. In an industry plagued by data scraping controversies, Versos is building the ethical infrastructure AI needs. That’s not just good karma, it’s good business.
We’re excited to support Chris, Cezar, and the team as they build the infrastructure layer for AI’s video future!
Read more
Betakit article, December 1, 2025: Versos offers copyright-cleared data to help AI generate videos, not lawsuits.