Encord Raises $60M for 'Physical AI' Data Infrastructure

Founded in 2021 by Ulrik Stig Hansen and Eric Landau, Encord has now raised a total of approximately $110 million. The company is addressing the critical data infrastructure needs of "physical AI," which involves complex, multi-modal data from sources like video, LiDAR, and other sensors used in robotics and autonomous vehicles. This focus on the data layer, rather than the models themselves, targets a significant bottleneck where ML teams often spend over 80% of their time. The push into physical AI highlights the growing debate between using synthetic data and human-in-the-loop (HITL) validation. While synthetic data offers speed and scalability, human labelers are crucial for tasks requiring nuance, context, and the ability to identify and mitigate biases that AI-generated data might perpetuate. This distinction is vital for agentic models that must operate reliably in the messy, unpredictable real world. Model alignment techniques like Reinforcement Learning from Human Feedback (RLHF) are central to this process. RLHF uses human preferences to train a reward model, which then guides the AI's learning process to produce outputs that are more helpful and harmless. This method is more complex than simple supervised fine-tuning but is considered highly effective for aligning models with human values. A newer approach, Constitutional AI, aims to make alignment more scalable by providing the model with a set of explicit principles—a "constitution"—to follow. The AI then learns to critique and revise its own outputs based on these rules, reducing the reliance on constant, granular human feedback for every decision. This method is being applied in areas like content moderation and other systems where safety and ethical adherence are critical. Evaluating agentic AI requires new benchmarks beyond traditional metrics. Instead of just measuring text quality, new evaluations like AgentBench, WebArena, and GAIA test an agent's ability to perform multi-step tasks, use tools, and navigate complex environments. Enterprises are increasingly focused on holistic evaluations that include cost-efficiency and reliability, as some agents can have 50x cost variations for similar accuracy levels. The fundraising climate for AI infrastructure remains robust, with investors reallocating capital from other sectors into AI. In early 2026, seventeen U.S.-based AI startups have already secured funding rounds exceeding $100 million each. However, the market is concentrating, with fewer startups receiving larger funding amounts, and investors are increasingly scrutinizing compliance and a clear path to profitability. This focus on high-quality, specialized data is transforming the data labeling workforce. The demand is shifting away from low-skill, repetitive tasks toward experts in fields like medicine and law who can provide the nuanced, domain-specific feedback required to train frontier models. This trend points to a future where data labeling becomes a more specialized and highly-skilled profession, essential for building trustworthy and capable AI systems.