NomadicML Secures $8.4 Million to Transform How Autonomous Vehicles Handle Massive Video Data

NomadicML Lands $8.4 Million to Tame the Data Overload Facing Autonomous Vehicle Builders

Building the next generation of autonomous machines requires more than just cutting-edge hardware — it demands an intelligent way to make sense of the enormous volumes of data those machines produce. That's exactly the problem NomadicML is setting out to solve, and investors are taking notice.

The startup announced a seed funding round of $8.4 million on Tuesday, reaching a post-money valuation of $50 million. The round was led by TQ Ventures, with additional participation from Pear VC and Google's AI chief Jeff Dean. The fresh capital will fuel customer acquisition and further development of the company's core platform.

The Problem: A Sea of Unsearchable Video

Companies building self-driving cars, warehouse robots, and autonomous construction equipment routinely accumulate millions of hours of video footage for training and evaluation purposes. The traditional approach to organizing this content? Hire humans to watch it. Even at fast-forward speed, that process simply doesn't hold up at scale.

NomadicML estimates that roughly 95% of fleet data collected by its target customers sits untouched in archives — valuable intelligence that goes completely unused.

The challenge intensifies when teams need to locate so-called edge cases — rare but critically important events that expose weaknesses in physical AI models. Finding a needle in a million-hour haystack is precisely where NomadicML's technology steps in.

How NomadicML's Platform Works

Founded by CEO Mustafa Bal and CTO Varun Krishnan — who first crossed paths as computer science undergraduates at Harvard — NomadicML has built a platform that converts raw video footage into structured, searchable datasets using a suite of vision language models.

The system doesn't just label footage; according to Krishnan, it functions as an "agentic reasoning system" capable of understanding the context of actions taking place on screen. Teams can describe what they're looking for in natural language, and the platform figures out how to locate it across vast libraries of video.

Practical applications include:

• Identifying every instance in which an autonomous vehicle crosses an intersection under police direction
• Pinpointing moments when vehicles pass beneath a specific type of overpass
• Extracting precise gripper positions from robotic manipulation footage
• Analyzing the physics of lane changes from camera data

These findings can then be fed directly into training pipelines, enabling faster iteration and more reliable reinforcement learning.

A Team Built for the Problem

Bal explained that he and Krishnan developed the concept after repeatedly encountering the same data management headaches during their time at companies including Lyft and Snowflake.

"We are providing folks insight on their own footage, whatever drives their own AVs and robots," Bal told TechCrunch. "That is what moves these autonomous systems builders forward, not random data."

The team behind the platform is notable in its own right. Krishnan holds the title of international chess master, currently ranked 1,549th in the world. Meanwhile, every one of the company's roughly dozen engineers has published a peer-reviewed scientific paper.

NomadicML also claimed first prize at Nvidia GTC's pitch competition last month, adding further validation to its technical approach.

Growing Demand and Early Customers

The platform is already being used by a handful of high-profile clients, including Zoox, Mitsubishi Electric, Natix Network, and Zendar.

Antonio Puglielli, VP of Engineering at Zendar, noted that Nomadic's tool allowed his team to accelerate development significantly faster than outsourcing would have permitted, and highlighted the company's domain expertise as a clear differentiator.

This category of AI-powered auto-annotation is quickly becoming essential infrastructure for physical AI development. Established players like Scale, Kognic, and Encord are all building similar capabilities, while Nvidia has released an open-source model family called Alpamayo aimed at the same challenge.

Why Specialization Is the Winning Strategy

Schuster Tanger, the TQ Ventures partner who led the round, framed the investment thesis around the value of focused infrastructure.

"It's the same reason Salesforce doesn't build its own cloud and Netflix doesn't build its own content distribution facilities," Tanger said. "The second an autonomous vehicle company tries to build Nomadic internally, they're distracted from what makes them win, which is the robot itself."

What Comes Next

Looking ahead, NomadicML is expanding its capabilities beyond visual data. The next frontier involves processing non-visual sensor inputs such as lidar readings, as well as fusing data across multiple sensor modalities simultaneously.

"Juggling terabytes of video, running it against hundreds of hundred-billion-parameter models, and accurately extracting insights from all of that — it's genuinely, insanely difficult," Bal acknowledged.

With fresh funding, an impressive technical team, and a growing roster of enterprise customers, NomadicML appears well-positioned to become a foundational layer in the autonomous systems stack.