Physical AI is moving from labs to large-scale use in self-driving cars, delivery drones, and manufacturing robots. This shift brings a new data challenge for AI teams: managing complex real-world multimodal data. Unlike large language models trained on internet data, physical AI requires training on sensor-rich datasets such as video, audio, LiDAR, and telemetry, which older platforms struggle to handle.
San Francisco-based Encord wants to fix this bottleneck. The data infrastructure company recently raised $60 million in a Series C round led by Wellington Management, with support from Y Combinator, CRV, N47, Crane Venture Partners, Harpoon Ventures, and new investors Bright Pixel Capital and Isomer Capital.
This brings Encord’s total funding to $110 million. Ulrik Stig Hansen, co-founder and co-CEO of
Encord shares with TFN, “The company’s valuation following its Series C is $550M.”
Encord’s platform helps AI teams manage, curate, annotate, and align multimodal data at scale, keeping data ready for physical AI development. Over the past year, data on its platform grew from 1 to 5 petabytes, and revenue from physical AI customers increased tenfold. Its clients include Woven by Toyota, Zipline, Skydio, and AXA Financial.
Aiming to be the universal data layer for AI
Encord was founded by co-CEOs Ulrik Stig Hansen and Eric Landau, who saw a growing gap between advanced AI models and the fragmented, manual data infrastructure teams depended on.
Hansen shares with us, “Eric and I worked on big data systems and deep learning research
and saw firsthand how ML teams spent 80%+ of their time on data, not models.
The entire industry was obsessing over model architecture while the real
bottleneck – data quality, curation, and evaluation – was completely underserved.
We started Encord to fix that.”
At its core, Encord is building AI-native data infrastructure, purpose-built for the sensor-heavy, multimodal world of physical AI. Their platform covers the full data lifecycle: capturing, organising, labelling, aligning with human feedback, and redeploying for retraining.
Unlike generalist tools like Labelbox, Scale AI, or SuperAnnotate, Encord is all-in on automation and continuous learning. That’s a game-changer for robotics, automotive, and drone companies, where data is always in flux.
“We handle the full lifecycle at petabyte scale, with an AI-native architecture that uses models to improve the data that trains them. That data flywheel and feedback loop is the core differentiator,” elaborates Hansen.
Encord’s advantage lies in a unified data layer that combines curation, annotation, and evaluation workflows to avoid tool fragmentation. Its AI-native architecture is built specifically for video, LiDAR, sensor, and 3D data. It also provides scalable automation with active learning and human-in-the-loop systems to improve data quality over time.
What about diversity?
On diversity, Hansen notes, “We don’t share specific breakdowns publicly, but we’re a team of 150+ across two countries (US, UK) with 40+ nationalities represented. Diversity of background and thought is something we actively invest in.”
What’s next?
With fresh funding in hand, Encord is gearing up to accelerate product development, expand into new markets, and scale up globally as physical AI adoption takes off.
