Depth-Wit (深度机智) lands hundreds of millions in Series A for physical AI base model in two months
The AMW Read
Introduces a new Chinese full-stack physical AI entrant with a novel 'Human Learning' data approach, updating the player map and exemplifying the fastest-ARR-ramp recurring pattern; funding is substantial but not at billion-dollar scale.
Depth-Wit (深度机智) lands hundreds of millions in Series A for physical AI base model in two months
Depth-Wit, a Beijing-based physical AI startup, has closed a Series A round of "hundreds of millions of renminbi" (approximately several tens of millions of USD) within two months, following a prior round of similar size. The latest tranche was led by Guoshou Yangtze River Delta Science & Innovation Fund, with participation from existing backers Puhua Capital, ChengTong Science & Innovation Fund, and a consortium of market-oriented and industrial investors including Lanhu Capital, Beyondsoft Technology, Pangu Venture Capital, Zhaohui Capital, Caixin Capital, Daohe Long-Term Investment, Yigao Capital, and Mingde Capital. The company defines an original "Human Learning" technical route, building a full-stack data-model-hardware system for physical AI. Depth-Wit claims its first-generation head-mounted data-collection device predates a similar product from Scale AI by nearly a year, and its PhysBrain 1.0 foundation model was released before Generalist AI's GEN-1. The company has accumulated tens of thousands of hours of first-person multimodal data across hundreds of real-world scenes, validated zero-shot generalization from human data to robot execution, and shipped full product lines—including full-size humanoid robots, wheeled dual-arm robots, and 3D-printed educational platforms—generating cumulative signed orders totaling tens of millions of RMB in under one year of operation.
Why it matters: Depth-Wit illustrates an emerging force in the physical AI segment—a China-native player pursuing a full-stack, domestically-controlled base model for embodied intelligence. This fits the "fastest-ARR-ramp" recurring pattern seen in AI coding tools, now extending into robotics, as well as the "context-engineering moat" pattern where proprietary first-person human data rather than simulation data becomes the defensible training asset. The funding also signals growing sovereign-capital appetite for domestic physical AI infrastructure, with state-linked funds (Guoshou, ChengTong) anchoring the round. However, the company's total disclosed funding of "hundreds of millions" RMB remains orders of magnitude below the billion-dollar rounds commanded by Physical Intelligence or Bezos Prometheus, underscoring a capital-compression arc for Chinese robotics startups relative to U.S. peers.
Grounded expert take: Depth-Wit's claim of zero-shot generalization from pure human data to robot execution, if independently reproducible, would mark a real step toward breaking the data-cost bottleneck that has limited physical AI scaling. The company's integration with Insta360 (影石创新) for spatial perception hardware and its coverage by state media (CCTV's "News Simulcast") point to a strategic alignment with China's push for autonomous physical AI supply chains. Yet the article is a funding press release: milestones like "leading international benchmarks" and "design ahead of Scale AI" are self-reported without third-party audit. The key open question is whether Depth-Wit's human-first data approach can achieve the scale (millions of hours) and generalization breadth necessary to compete with simulation-driven or hybrid approaches from Physical Intelligence, Google DeepMind, or Tesla. For now, Depth-Wit is a company to watch for its technological narrative and capital momentum, but its market position remains pre-revenue at scale.