Encord raises 50M to build the data layer for physical AI

"Physical AI systems generate and depend on extremely rich data. Unlike large language models trained on static text, autonomous machines rely on streams of video, sensor feeds, lidar scans, audio and telemetry captured in dynamic settings. Legacy data tools, designed for labelled images or text, struggle to handle this scale and diversity."

"Encord's platform automates the full data lifecycle, from ingestion and organisation to annotation, curation and evaluation, helping development teams build and retrain models with traceability and performance insight. Co-founder and co-CEO Ulrik Stig Hansen has repeatedly emphasised that for physical AI, the bottleneck is not the size of the models but the quality and readiness of the data feeding them."

"Even sophisticated algorithms will underperform if the underlying datasets are inconsistent or misaligned with real-world conditions. Encord says its platform currently manages more than 5 petabytes of multimodal data, roughly triple the volume from a year ago, and that revenue from physical AI customers has increased by an order of magnitude."

Encord, a London-based data infrastructure startup, secured €50 million in Series C funding led by Wellington Management, bringing total capital raised to €93 million. The company addresses the growing demand for data infrastructure as artificial intelligence expands into physical applications like autonomous vehicles, robots, and drones. Physical AI systems require complex, multimodal datasets including video, sensor feeds, lidar scans, audio, and telemetry—far exceeding the capabilities of legacy data tools designed for static text or labeled images. Encord's platform automates the complete data lifecycle from ingestion and organization through annotation, curation, and evaluation. The company manages over 5 petabytes of multimodal data, triple the volume from a year prior, with physical AI customer revenue increasing significantly.

Read at TNW | Startups-Technology