London-headquartered Encord raised €50 mn in Series C funding to expand its AI-native data infrastructure as physical AI shifts from pilot programmes into production.
Wellington Management led the round, lifting total funding to €93 mn. Existing backers Y Combinator, CRV, N47, Crane Venture Partners and Harpoon Ventures joined alongside Bright Pixel Capital and Isomer Capital.
Co-founder and co-CEO Ulrik Stig Hansen argues model scale no longer defines progress in physical AI. Data readiness does.
Teams build advanced models, yet performance breaks down when training data lacks consistency or fails to mirror field conditions. Encord focuses on cleaning, structuring and aligning data before models touch it.
Bill Tinney, Senior Director of AI Product Management and Partnerships at Vantor, describes operational data control as a competitive edge.
Vantor builds AI for critical infrastructure and national security, and uses Encord to manage geospatial workflows spanning curation, annotation and evaluation.
Fragmented tooling slows production teams. A unified data layer reduces friction and scales with workflow complexity.
Founded in 2021, Encord positions itself as a universal data layer for AI teams. The platform manages, curates, annotates and aligns data across the full lifecycle. More than 300 AI teams use the system, including Woven by Toyota, Zipline, AXA and Skydio.
Physical AI systems depend on multimodal data such as audio, video, images, sensor streams and 3D point clouds. Legacy data platforms struggle with those formats.
Storage and processing requirements exceed text-based workloads, which strains infrastructure budgets and forces technical redesigns.
Encord reports platform data volumes rising from 1 petabyte to over 5 petabytes within twelve months. That scale surpasses three times the data used to train GPT-4.
Revenue from physical AI customers increased 10x over the same period, reflecting production deployment rather than lab experimentation.
Industry projections estimate more than 400 mn AI-driven robots will come online within four years, with the physical AI sector surpassing €25 bn in market size over the same window.
Growth shifts focus toward proprietary datasets captured through sensor feeds, robotic telemetry and field edge cases.
Unlike large language models trained on open internet data, physical systems rely on controlled, domain-specific inputs.
Co-founder and co-CEO Eric Landau states the new capital will fund product development and geographic expansion. He frames the opportunity in operational terms.
Models degrade without disciplined data pipelines. Continuous improvement in deployed systems depends on infrastructure built for iteration, not one-off training cycles. Encord targets that layer as physical AI scales into commercial environments.
