China backs embodied AI data growth with $280M PsiBot raise

China’s push into embodied AI just got a high-profile boost, as PsiBot closed a RMB 2 billion ($280 million) financing round to accelerate real-world data collection and logistics deployments for vision-language-action models. The round, spanning an angel and Pre-Series A, signals both deep investor interest and a government-friendly appetite for software-first AI tools that can scale in industrial settings.

PsiBot bills itself as a “small full-stack” embodied AI company, prioritizing the software layer over hardware. Its core technology centers on end-to-end vision-language-action (VLA) models and the data-collection toolchains that train them. In practice, that means building workflows that turn messy real-world observations into clean, labeled signals the models can learn from, then translating those insights into practical robot behavior on the factory floor or in a warehouse. The company has already developed a 21-degree-of-freedom exoskeleton data glove that records human manipulation data via crowdsourced operation, a design choice intended to slash the cost of collecting large-scale training data. In other words, PsiBot is betting that the hardest part of embodied AI isn’t the silicon or the simulations—it’s the data you feed the models and the ways you curate it for real-world use.

The financing arrivals—led by Shanghai state-backed Xuhui Capital—come with strategic support from a constellation of public and private players. Investors include China Development Financial, Guozhong Capital, a media-investment fund backed by China Media Group, and a Yangtze Optical Fibre fund; several existing shareholders also increased their stakes. The round also reflects the broader policy climate in which local governments and state-backed funds are channeling capital into AI-enabled logistics, robotics software, and the data ecosystems that power them. The stated aim is to build large-scale, domestically anchored embodied AI data infrastructure, reducing dependence on foreign data pipelines and speeding deployment in logistics and beyond.

On the ground, the implications are tangible for supply chains eyeing automation without surrendering control over their data. PsiBot’s model-centric approach—emphasizing data collection toolchains and scalable data infrastructure—addresses a core bottleneck: the cost and speed of generating usable, labeled data for AI tasks in dynamic environments like warehouses and loading docks. If successful, the company’s software stack could enable rapid customization of embodied AI solutions across retailers, manufacturers, and third-party logistics providers, with data pipelines that translate a crowded real-world environment into precise robot actions.

Industry practitioners should watch several dynamics. First, data quality matters as much as data volume; crowdsourced data collection can introduce variability, so robust labeling protocols and validation loops are essential to prevent model drift. Second, data ownership and governance will become sharp edges in contracts and partnerships: who owns the raw sensor streams, the annotations, and the trained models when data is generated by human-in-the-loop workflows? Third, interoperability will be critical. A software-first playbook must harmonize with multiple robot platforms and control architectures to avoid vendor lock-in and to sustain scale. Finally, the economics of deployment matter: without sizeable logistics pilots that translate data improvements into tangible ROI, the cost of data collection—even with a glove and crowdsourcing—could outpace gains in accuracy or speed.

In a space where hardware breakthroughs often draw the loudest headlines, PsiBot’s emphasis on end-to-end data ecosystems and practical deployment signals a mature shift: the fastest path to useful embodied AI may lie in the data factories built to train and continuously improve these agents, not just the robots themselves. If the capital rounds translate into real-world scale, the company’s 21-DOF glove and VLA stack could become a blueprint for how China’s logistics and manufacturing operators accelerate automation without losing control of their data assets.

Sources

WUWENAI Launches Physical AI Data Infrastructure Platform “Wuyin” in Zhejiang

PsiBot Raises $280M to Accelerate Embodied AI Data Collection and Logistics Deployment