China-based Spirit AI has raised $280 million in growth funding to further develop and deploy its general-purpose embodied AI models. The company announced the funding on February 25, as the robotics sector increasingly shifts toward Vision-Language-Action (VLA) architectures that integrate perception, language, and physical control within a single model.
Headquartered in Beijing, Spirit AI is building a universal control system for robots using large-scale datasets derived from human video footage and wearable sensor data. According to the company, broad and less heavily curated datasets are essential to achieving robust generalization in real-world environments.
The approach reflects broader international efforts by organizations such as Google DeepMind and Physical Intelligence, which are also developing large-scale embodied AI systems.
Focus on Large-Scale, Diverse Data
Spirit AI argues that highly curated “clean” datasets can limit model performance at scale. Instead, the company prioritizes what it describes as “dirty data” — large volumes of unscripted interaction data collected in realistic settings.
Co-founder and Chief Scientist Yang Gao, who is also an assistant professor at Tsinghua University, states that this strategy is necessary to develop models capable of operating reliably across diverse real-world scenarios.
Spirit AI reports that it has accumulated more than 200,000 hours of interaction data to date, with a target of exceeding one million hours by the end of 2026. Through proprietary wearable data-collection devices, the company says it has reduced data acquisition costs by approximately 90 percent compared with traditional teleoperation methods.
In January 2026, the company’s Spirit v1.5 model ranked first on the global RoboChallenge leaderboard, according to Spirit AI.
Deployment in Battery Manufacturing
The technology has been deployed in production lines at CATL, one of the world’s largest battery manufacturers. There, the models control robots handling flexible wire harnesses — a task that has historically been difficult to automate due to material variability.
Spirit AI reports that its systems achieve a success rate of more than 99 percent in this application, operating with precision and cycle times comparable to skilled human workers.
With the new funding, the company aims to accelerate further development of its general-purpose models and expand their use across additional industrial settings.
