Overview
DeepSeek founder Liang Wengfeng is intensively entering embodied AI, per Tencent News and NetEase Tech: he plans to lead the seed round of Winverse AI—founded by ex-Tencent Hunyuan multimodal lead Hu Han—with tens of millions of dollars at a ~$400M post-money, targeting the robot “brain” and model-acceleration infrastructure. This extends DeepSeek’s “embodied turn” after its ~¥141M strategic stake in Unitree’s IPO.
Earlier, at the October 15 China International Industry Fair, Chinese robot developers, chipmakers and industrial leaders built a commercial supply chain for embodied AI, signaling a shift from “trade-show prototypes” to “factory-floor mass deployment.” Zhang Hao of the Shanghai AI Society noted embodied AI has moved past exploration into a unified ecosystem of chips, foundation models and robots; the bottleneck shifted from single-hardware to fleet-level coordination. Five developers showed production-ready hardware; enterprise-grade costs fell sharply.
Another signal: domestic AI-chip funding is heating up as chips move from “catching up” to “scaling,” with upstream players like ChangXin posting profit pre-announcements on AI demand. But risks accumulate—AI-cloud firm CoreWeave fell 15% in a day with >$50B debt/lease obligations, as leverage risk spreads from credit to equities.
Background & Interpretation
1. Technical Background and Evolution
Liang once ranked AGI as “LLM → chain-of-thought → Agent → continual learning → self-iteration → embodied,” calling embodiment the “hard job.” His early entry reflects the ceiling of the LLM narrative and physical-world interaction as AGI’s next gate. From investing in Unitree (body) to betting on Winverse (brain), DeepSeek positions across the “embodied + LLM” chain, echoing OpenAI’s robotics move.
2. Core Drivers and Underlying Mechanisms
Two capital signals stand out: giants now fund compute via low-cost bonds rather than equity dilution, showing balance-sheet strength; and money flows along the entire embodied chain—body, brain and chips alike. The mechanism is “industry resonance”: upstream (memory, packaging, power) already books AI gains, lifting mid/downstream valuations and funding.
3. Market Structure and Industrial Chain Impact
For private markets, embodied AI becomes the post-LLM funding theme, lifting early-stage valuations; for compute chips, demand extends from training to edge and robot inference, opening domestic-substitution windows. But highly leveraged AI-infra solvency risk warns that when demand slows, heavy-asset players feel it first. Value capture is reconstructing—“pick-and-shovel” sellers may cash in before “miners.”
Implications & Outlook
Short-to-medium term, embodied-AI funding and mass production accelerate in parallel; edge-model and robot-inference chip demand surges. Watch the pace of DeepSeek-led deals, CoreWeave-style debt leverage’s drag on AI-infra valuations, and post-scale ecosystem maturity of domestic AI chips.
Takeaways for Industry Participants
AI and robot firms should fill the chain (body + brain + compute) rather than single-point break through. Investors can favor upstream certainty and early-stage valuation discipline. Industrial players must guard against debt-leveraged compute expansion when demand slows. Developers should build edge/robot model-deployment capability as a new moat.
This column compiles industry information and shares technical perspectives; it does not constitute investment advice.