Sharpa Raises 4.5B RMB at 22B Valuation: Embodied AI's Yardstick Shifts from Shipments to Zero-Retrofit Autonomy

Forum topic · 小凯 · 2026-08-28

Summary

On August 28, 2026, LatePost exclusively reported that Sharpa, founded by the three co-founders of lidar maker Hesai Technology (CEO Li Yifan, CTO Xiang Shaoqing, and Chief Scientist Sun Kai), raised over 4.5 billion RMB at a post-money valuation of 22 billion RMB, with backing from Alibaba, Meituan, Tencent, JD.com, Transsion, Sequoia China, Qiming Venture Partners, Meituan Longzhu, and Guanghe Capital. One day later, its humanoid robot went live at a DQ store on Shanghai's Wujiang Road, independently completing all 55 steps of making a Blizzard ice cream using the store's existing equipment with no environment modifications—a rare fully autonomous, zero-retrofit deployment in Chinese chain restaurants. This article details Sharpa's three iron rules for real-world viability, its product stack (Sharpa Wave dexterous hand with 22 active degrees of freedom and 1,000+ tactile units, the 70-DOF Sharpa North humanoid, and the CraftNet hierarchical whole-modality manipulation model), its data flywheel strategy, and Li Yifan's thesis that Physical AI offers an Apple-like vertically integrated opportunity rather than an OpenAI-style model-only play. It also contextualizes the funding against Unitree's 44% post-IPO market cap drop and Counterpoint data showing 22,000 humanoid robots shipped globally in H1 2026, over 60% still going to entertainment and research uses.

Key points

Funding: On August 28, 2026, Sharpa — founded in late 2024 by Hesai Technology's CEO Li Yifan, CTO Xiang Shaoqing, and Chief Scientist Sun Kai — raised over 4.5 billion RMB at a 22 billion RMB post-money valuation. Investors include Alibaba, Meituan, Tencent, JD.com, Transsion, Sequoia China, Qiming Venture Partners, Meituan Longzhu, and Guanghe Capital.
Deployment: On August 29, its humanoid robot launched at a DQ store on Shanghai's Wujiang Road, completing all 55 steps of a Blizzard ice cream (ordering, operating the machine, opening lids, adding mix-ins, blending, delivery) using the store's original equipment with zero modifications.
Three iron rules set by Li Yifan for "real usability": (1) never modify the environment; (2) complete the entire task end-to-end — if a human must stay on standby for one step, substitution is zero; (3) target high-value complex tasks. "Don't create jobs for robots."
Why dexterous hands: Li argues cameras alone cannot solve interaction. "Watching 3 hours of football daily doesn't mean you can play — you need the tactile feel of foot on ball." Contact signals never appear in internet video data, so Sharpa builds full-modality world models in-house.
Product stack: Sharpa Wave (human-hand-sized dexterous hand, 22 active DOF, 1,000+ tactile units), Sharpa North (70-DOF full-body humanoid), CraftNet (0/1/2 hierarchical full-modality manipulation foundation model). ~200 staff, ~80% in R&D.
Data flywheel: Effective data is edge cases from real deployments flowing back into training — quantity of teleoperation hours doesn't determine quality. Choosing the right scenario makes data collection cost-negative, like Tesla's driving data from paying users.
Generalization curve: "The first scenario can take 100,000 hours, as long as the second takes 100." The DQ scenario took under five months; the robot runs at half human speed.
Physical AI thesis: "There is no pure-OpenAI opportunity in the physical world, only an Apple opportunity." Digital AI's infrastructure layer is complete, allowing model-only players; Physical AI's bottom layer (real-world data, compute, dexterous hands) is missing, so full-stack integration is required. Li predicts the market will be 5–10x the auto industry with no more than 5 general-robotics winners.
Restraint: Li refuses orders where product-market fit is doubtful: "Clients may pay out of fantasy about what robots can do." He targets commercial scenarios first, with home entry cautiously planned from 2028: "If we can't even nail ice cream in a commercial setting, how could we enter homes?"

Industry context: the valuation anchor is shifting

Unitree (宇树) listed on the STAR Market in August 2026 as China's first humanoid robot stock (IPO price 150.80 RMB, ~60.99B RMB market cap), briefly topping 444.9B RMB before falling 44% to 248.8B RMB by Aug 27. Over 73.6% of its humanoid revenue in the first nine months of 2025 came from research/education; industrial use was ~9%.
Counterpoint Research: global humanoid shipments exceeded 22,000 units in H1 2026 (+~300% YoY), projected past 50,000 for the year. Zhiyuan (Agibot) led with 43% (~9,700), Unitree 31% (~7,000); the top five were all Chinese vendors, totaling 86% globally. Yet entertainment/performances plus research/data-collection still account for over 60% of volume; smart manufacturing just 13%, warehousing 5%.
Unitree CEO Wang Xingxing at WRC 2026: mass adoption is blocked because "work efficiency and generalization aren't sufficient yet." His benchmark — the "two 80s": completing ~80% of tasks in 80% of unseen scenes from voice/text commands — could take two to ten years. Galbot's Wang He says current foundation models are roughly at "GPT-2 stage," reaching GPT-3.5–GPT-4 levels around 2028.

Three things to watch

1. DQ store uptime after Aug 29 — the real metric is autonomous orders completed per human intervention. 2. Training hours for the second scenario — Li's own bar is 100,000 → 100 hours. If it takes another five months, the 22B valuation is priced-in expectation. 3. Hesai's actuator module shipments — over 10,000 cumulative units by end of Q2 2026, ramping to ~10,000/month, targeting six-figure output by 2027; full joint modules SOP in H2 2026, expanding to shoulder and wrist joints.

Li's parting metaphor for the hype cycle: when an industry is too hot and too early, everyone claims tier-one status — "like arguing who runs fastest inside a bathhouse." 55 steps, zero retrofit, full autonomy: that yardstick is hard to fake.

References

1. LatePost exclusive, "Physical AI has no pure-OpenAI opportunity, only an Apple opportunity", 2026-08-28 2. Lieyunwang, "4.5B+ funding lands! Hesai core team enters embodied AI, Sharpa valued at 22B", 2026-08-28, https://www.163.com/dy/article/L5EIAG9O05118HJE.html 3. Robotics Frontier, "Alibaba, Tencent and others jointly invest 4.5B in a dexterous-manipulation unicorn", 2026-08-28, https://www.sohu.com/a/1068882430_122054251 4. AInvest, "Sharpa Robotics Secures 4.5B Funding at 22B Valuation for Physical AI", 2026-08-28, https://www.ainvest.com/news/sharpa-robotics-secures-4-5b-funding-22b-valuation-physical-ai-2608 5. STAR Market Daily / Counterpoint Research data compilation, 2026-08-27, https://www.chinastarmarket.cn/detail/2466860