Encord has just raised a $60m Series C: An interview with Co-CEO and Co-founder Eric Landau

Encord raises $60m to build AI's physical data layer

Encord is building the data layer for physical AI and just raised $60m to scale it. The bet is that models commoditize while the scarce input becomes the data that takes a system from 90% to 99.999% accuracy, and that physical AI has no internet to scrape for the sensor and video it needs. Encord already moves petabytes for customers such as Toyota's self-driving unit.

Takeaways

• Encord raises $60m on physical AI demand. Wellington Management led the round, taking total funding to $110m. Landau says market pull drove it, and that the surprise was investors asking why Encord had not burned more of its Series B.
• Data will outlast models and compute. Landau says AI needs three things, and data is the one he thinks will matter most. Models built multi-hundred-billion-dollar companies and NVIDIA is worth north of a trillion, while Scale AI, the largest data company, sold for $14bn.
• Physical AI drives a 10x revenue jump. Revenue from physical AI work has grown 10x, Landau says. Customers include Woven by Toyota, Toyota's self-driving unit, and Pickle Robot, which uses Encord as its data engine with agentic annotation and human review before retraining.
• The physical world cannot be scraped. Text on the internet seeds language models, but nothing equivalent exists for robots and factories. Physical data is multimodal, mixing sensor, audio, text and visual. Companies gather it ad hoc, so the smartest data strategy wins as models commoditize.
• Encord moves petabytes, 50 times GPT training data. Landau says data flux through the platform has grown an order of magnitude and now runs at multiple petabytes, roughly 50 times all the data used to train GPT-3 or GPT-4.
• Human imagination limits AI adoption. Landau says nobody has found the equilibrium state for AI inside companies. Encord runs Claude in Slack and Notion and held a Cursor and Claude Code hackathon. He is bearish on tools whose problem has not appeared.

What Landau expects next

• AI already completes tasks that take a human two hours, and that doubles every seven months on the METR Institute's measure. Landau says model progress is not slowing.
• Several model providers can win because no single metric matters. He expects Anthropic to lead in enterprise and coding, and OpenAI in consumer, with Google eventually winning on its cash machine.
• He expects hundreds of thousands of AIs in different shapes and sizes rather than one monolithic system, some open source and some closed, and compares models to people with distinct personalities.
• His everyday picture has a robot cleaning the house overnight, drones delivering packages, autonomous commutes and health scans.