NomadicML just closed an $8.4 million seed round to build what it calls an “agentic reasoning system” for the mountains of video data that autonomous vehicle and robotics companies can’t process fast enough. TechCrunch AI reports the round values the company at $50 million post-money.
The round was led by TQ Ventures, with participation from Pear VC and Google’s Jeff Dean. The funding will go toward onboarding more customers and refining the platform.
The Problem Is Massive
Companies building self-driving cars, robots, and autonomous construction equipment collect thousands (sometimes millions) of hours of video. Right now, organizing and cataloging that footage is a human job. Someone has to watch it. Even on fast-forward, that doesn’t scale.
Worse, the most valuable data is the rarest. Edge cases (a police officer waving a car through a red light, vehicles driving under a specific type of bridge) are exactly what physical AI models need to learn from, and exactly what’s hardest to find in a sea of routine footage.
What Nomadic Actually Does
Nomadic’s platform turns raw footage into a structured, searchable dataset using a collection of vision language models. Think of it as a search engine for real-world driving and robotics footage.
But co-founder and CTO Varun Krishnan (who also happens to be an international chess master ranked 1,549th in the world) argues it goes beyond simple labeling. “It’s an agentic reasoning system: you describe what it needs and it figures out how to find it,” he told TechCrunch AI. The platform uses multiple models to understand actions in context, not just tag objects in frames.
This means customers can:
- Find specific edge cases across massive fleets for training data
- Monitor fleet behavior for compliance
- Feed curated datasets directly into reinforcement learning pipelines
- Iterate faster on model development
Who’s Using It
The customer list already includes Zoox, Mitsubishi Electric, Natix Network, and Zendar. Antonio Puglielli, VP of Engineering at Zendar, said Nomadic let his company scale up “much faster than the alternative of outsourcing” and praised its domain expertise over competitors.
Nomadic also won first prize at Nvidia GTC’s pitch contest last month, which is certainly not a bad signal for a seed-stage company.
Why This Matters
Auto-annotation and model-based data tooling is becoming critical infrastructure for physical AI. Established players like Scale, Kognic, and Encord are building AI tools for similar work. Nvidia released its open-source Alpamayo model family that can be adapted to the problem.
What stands out here is Nomadic’s positioning. It’s not trying to be a general-purpose labeling shop. It’s betting that AV and robotics companies will outsource this specific infrastructure the same way Netflix outsources content delivery.
“The second an autonomous vehicle company tries to build Nomadic internally, they’re distracted from what makes them win, which is the robot itself,” Schuster Tanger, the TQ Ventures partner who led the round, told TechCrunch AI.
What Comes Next
The immediate roadmap includes tools that understand the physics of lane changes from camera footage and tools that derive precise gripper locations for robotics video. The bigger challenge: expanding beyond visual data into lidar sensor readings and multi-modal sensor fusion.
“Juggling around terabytes of video, slamming that against hundreds of 100 billion-plus parameter models, and then extracting their accurate insights, is really insanely difficult,” CEO Mustafa Bal said.
At $50 million valuation on a seed round, investors are clearly betting that Nomadic can crack it. With 95% of fleet data reportedly sitting unused in archives across the industry, the opportunity is enormous if the tech delivers.
Full details are available in the original report on TechCrunch AI.