Spirit AI was founded in 2024 in Beijing to build embodied intelligence: foundation models that connect perception, language and physical action so robots can operate in the real world. The company develops vision-language-action (VLA) models trained at scale and pairs them with its own humanoid hardware, betting that tightly integrating model and embodiment is the fastest route to capable, general-purpose robots.
Its flagship robot, Moz1, is a force-controlled humanoid with 26 degrees of freedom designed for high payload efficiency, and it is driven by Spirit's in-house embodied-AI model. A distinguishing element of the company's strategy is its emphasis on training with large quantities of real-world, imperfect 'dirty data' rather than relying solely on clean, curated demonstrations, an approach the company argues better reflects the messiness robots face in deployment. In January 2026, Spirit reported that its Spirit v1.5 embodied-intelligence foundation model topped the RoboChallenge global leaderboard, signaling performance comparable to leading international embodied-AI systems.
The founding team is drawn from UC Berkeley, Tsinghua University and Peking University, and skews notably young, with many members under age 30. Co-founder and chief scientist Yang Gao is an assistant professor at Tsinghua and holds a PhD from UC Berkeley, anchoring the company's research credibility.
Spirit AI raised approximately $290 million in a Series A led by Chaos Investment and YF Capital, valuing the two-year-old company at about $1.5 billion, part of a broader surge of Chinese embodied-AI funding. The company competes in the global humanoid and robot-foundation-model race alongside firms like Physical Intelligence, FieldAI and X Square Robot, differentiating through its combination of in-house VLA models, force-controlled humanoid hardware, and a data strategy built around real-world operational data.