Xieyue Intelligence has closed a several-hundred-million-yuan angel-plus round, with Linear Capital, Junshan Capital, Hongyi Capital and Hidden Hill among the investors. It is the second raise in seven months, after an angel round led by Li Auto and Yuanjing Capital.
The money will push forward embodied foundation-model training, computing and data infrastructure, core-team expansion, and home-robot hardware development and scenario validation. The fast funding rests on a top-tier team and a differentiated lane.
Xieyue was founded by Chen Wei, Li Auto’s former chief AI scientist and head of its base-model department, and Zhang Xiao, Li Auto’s former product president. The core team comes from leading technology, AI, robotics and smart-car companies, spanning large-model algorithms, AI infrastructure, robot motion control, product and supply-chain management. Few teams in China combine ten-thousand-card model training, digital-physical twin development and end-to-end consumer-product mass production.
Where most of the field crowds into factories and logistics, Xieyue treats the home as the core training, validation and scaling environment for an embodied foundation model. It argues the home is among the most complex, highest-frequency and longest-value real physical spaces, the ideal proving ground for general embodied intelligence. Its strategy is progressive: prove the model and unit economics in semi-structured sites like hotels and care homes before entering private homes, targeting high-frequency chores such as laundry, storage and cleaning.
Xieyue is building a closed loop of robot body, embodied foundation model and home self-improvement, driven by Duplex Reasoning, a human-centric physical-world data flywheel and safety-first design. Duplex Reasoning breaks the old single or half-duplex flow of receive command, execute, return result, keeping a two-way channel with the human so dialogue can be interrupted and goals adjusted mid-task. Built on a VLA backbone with world-model prediction, it aims to lift understanding, decision and action success in open environments at controllable compute cost.
The firm believes the edge is not a bigger model or more data but high-quality data and systematic infrastructure, with training and data infrastructure as equal pillars. It has built its own ego-centric data gear and platform, planning to close the loop from collection, cleaning, labelling to training within 2026, and uses a single-body, full-stack strategy to control hardware variables before moving to consumer home robots.
More from the original report
Editor’s note: This is an adapted translation of the original Gasgoo report. It has been trimmed and restructured for readability for an international business audience.
Translated and adapted from Gasgoo (https://www.gasgoo.com/apps/50640d4b55d5cba175fb84f15d679f19/robot/news/70471960-seeds-%E7%90%86%E6%83%B3%E5%89%8Dai%E4%B8%80%E5%8F%B7%E4%BD%8D%E5%85%B7%E8%BA%AB%E6%99%BA%E8%83%BD%E5%88%9B%E4%B8%9A/).