Xiechuang Data Builds Full-Chain Computing Power Services for Physical AI, FCloud OmniBot Platform Completes Upgrade

As large-model technology gradually extends toward physical robots, physical AI is becoming the key link connecting virtual simulation and physical manufacturing, and the pace of industrial deployment depends heavily on dedicated computing power and high-quality interaction data. Addressing the industry-wide challenges of data scarcity and insufficient full-chain computing power supply, Xiechuang Data (300857.SZ) has deeply invested in building computing power infrastructure to provide infrastructure support for robot training, simulation, and policy optimization.

Looking at business implementation progress, this strategic positioning is rapidly translating into substantive commercialization results. In its investor relations activity record dated October 29, the company disclosed that it is already deeply cooperating with leading overseas enterprises on end-to-end model development and computing power optimization, that domestic customers continue to expand, and that in the future, as the robotics industry scales up, the computing power services business is expected to steadily deliver performance.

Currently, the large-scale deployment of physical AI is facing the dual core bottlenecks of data and computing power. Calculations by Sinolink Securities show that the real interaction data available for embodied models is less than one ten-thousandth of that for large language models, while embodied intelligence reaching a usable level requires at least 10 million hours of multimodal data. There are constraints on the computing power side as well: training world models and physical simulation require support from clusters of a thousand or even ten thousand GPUs, while robot bodies are strictly limited by power consumption, heat dissipation, and real-time latency, making it difficult for a single form of computing power to meet the full-chain needs of both training and inference. On this basis, EO Intelligence judges that computing power is shifting from a general-purpose foundation to a dedicated track for embodied intelligence, becoming a key variable determining the pace of industrial deployment.

In response to common industry pain points, Xiechuang Data has chosen to avoid the robot body manufacturing track and focus on building the underlying computing power and data foundation. Its FCloud OmniBot platform completed an upgrade this year. Relying on NVIDIA OSMO cloud-native orchestration capabilities, customers can freely, on demand, link Isaac Sim simulation, Cosmos data generation, Isaac Lab training, and GROOT model development into a complete pipeline that can be directly invoked, greatly lowering the threshold for developers in environment setup and scheduling.

The supporting construction of data capabilities and computing power clusters further consolidates the company's foundation advantages. At the World Robot Conference in August, FCloud, together with Tashan Technology, released a visual-tactile multimodal data solution, completing data collection through self-developed Ego visual devices and TS-ECHO tactile finger sleeves, achieving unified governance in the cloud and supporting cross-model reuse. In terms of computing power cluster construction, the company has established a dedicated training cluster focused on robot physical AI in Zhangjiang, Shanghai; in terms of business model, it takes computing power optimization capability as the core and achieves profitability by providing robot enterprises with full-chain Token services for training and inference. At present, the company has launched ecosystem cooperation with institutions such as the National and Local Center and the National Advanced Center, continuously expanding the boundaries of industrial deployment.

As the industry moves from technical validation toward large-scale industrial deployment, the computing power foundation, as the core hub connecting simulation, data, training, and real-machine deployment, is seeing its industrial value become increasingly prominent. Relying on its computing power optimization capability and full-chain service layout, Xiechuang Data is positioned at the core link of the industry, and its technological and ecosystem value will continue to be released as the industry matures, providing underlying support for the deep integration of physical manufacturing and the robotics industry. (Fang Yuan)