There's a camera in nearly every Swedish retail store. Most of them record footage nobody watches. Antrino Labs, until recently known as Preventai, wants to change that. The Stockholm and Lund-based startup just raised SEK 40 million from Pomonagruppen to scale a video AI platform that watches camera feeds in real time and flags anomalies as they happen.

The round values the company at SEK 240 million. That's a healthy multiple for a startup with roughly 200 connected cameras today. But co-founders Abdulla Salman and Noor Latif aren't selling the current state. They're selling the trajectory: 5,000 cameras within twelve months.

That's a 25x increase in deployed cameras. In a single year. The ambition alone tells you something about where video AI is heading.

Preventai launched in 2023 with a narrow focus: detecting shoplifting in real time using existing security cameras. The founders met at Lund University, bonded over a shared obsession with computer vision, and built a system that could spot suspicious behavior patterns without requiring new hardware.

The rebranding to Antrino Labs in January 2026 signals a much broader ambition. The company's core technology is now a video language model, or VLM, that can be prompted for different use cases. Retail theft was the entry point. Construction site safety, warehouse logistics, and public space monitoring are next.

"What sets us apart is that our software works on nearly any camera," Salman told Dagens Industri. That's the key differentiator. Most video analytics competitors require specific hardware or camera models. Antrino Labs runs on whatever's already mounted on the ceiling.

The SEK 40 million from Pomonagruppen is the headline. But the quiet support from Microsoft and Nvidia matters just as much. Microsoft is providing approximately SEK 2.3 million in cloud infrastructure credits. Nvidia is contributing around SEK 4.7 million in compute resources and is co-building a synthetic data pipeline to train Antrino's models on simulated scenarios.

The Nvidia partnership is particularly notable. Synthetic data, essentially AI-generated training footage, solves one of the biggest bottlenecks in video AI: getting enough labeled examples of rare events. You can't easily collect thousands of real shoplifting videos for training. But you can generate synthetic ones that teach the model what to look for.

This approach has been gaining traction across computer vision. What makes Antrino's implementation interesting is that it's happening at a Swedish startup, not at a Silicon Valley lab. The Nordic angle isn't incidental. Sweden's retail chains and security companies represent a concentrated, accessible market for early deployment.

Any company putting AI behind surveillance cameras is going to face scrutiny. Antrino Labs says its system doesn't store identifiable footage and focuses on behavior patterns rather than facial recognition. The founders are clearly aware of the optics. Their website includes a dedicated section on AI safety and privacy methodology.

Sweden's ISP, the inspectorate for strategic products, still needs to approve the Pomonagruppen investment. Salman expects that to clear by late April. The regulatory review is reportedly procedural rather than substantive, but it's a reminder that AI surveillance companies operate in a more complex regulatory environment than your average SaaS startup.

The broader European AI Act will also affect how companies like Antrino deploy their technology. Real-time biometric surveillance in public spaces faces strict restrictions under the Act. Antrino's retail focus keeps it in safer regulatory territory for now, but any expansion into public space monitoring would require careful navigation.

The technology works. Antrino's existing customers control roughly half of Sweden's security camera market, according to the company. The challenge now is deployment speed. Going from 200 to 5,000 cameras means onboarding dozens of new customer sites, each with different camera hardware, network configurations, and privacy requirements.

Pomonagruppen's capital should cover the sales and implementation team needed for that expansion. The cloud credits from Microsoft and Nvidia handle the compute costs. What's left is execution: can Salman and Latif build a deployment machine as impressive as their AI model?

Twelve months from now, the camera count will tell the story.