Over the past 26 months, a humanoid robotics company has released four generations of hardware. Its gross margin on robots is 63.2%. It has shipped over 5,500 units—more than any other humanoid robot maker. This is not a crypto startup. But the way Yuzhu Technology is building its business mirrors the data flywheel I first saw in DeFi yield farming back in 2020. Back then, I watched protocols like Compound bootstrap liquidity by offering yield. Users deposited tokens, the protocol collected interest data, and smart contracts iterated. Yuzhu is doing the same thing, but with physical hardware: low-cost robots generate real-world interaction data, which trains the next generation of models. It’s a flywheel, and it’s working. But is it sustainable? And what does this have to do with blockchain? Everything, if you believe that the future of trust is decentralized.
Yuzhu Technology, based in China, was recently covered by Nomura Securities with a “Buy” rating. The report highlights a handful of mind-bending numbers: 10-20% of robot components are sourced externally—everything else is built in-house. That vertical integration delivers a gross margin of around 60%, a level most consumer hardware companies can only dream of. The company’s humanoid robots span four product lines: the G1 for consumers, H1 for research, R1 and H2 for industrial use. In 2025, Nomura expects Yuzhu to ship 5,500 units, making it the global leader in humanoid robot shipments. The revenue forecast is equally aggressive: from 26.87 billion yuan in 2026 to 53.96 billion in 2027, then 131.84 billion in 2028—a compound annual growth rate of 122%. Those numbers would make any crypto project’s TVL look tame.
Here’s where the blockchain analogy hits hard. Yuzhu’s strategy is not about AI breakthroughs. It’s about scale and data. The thesis goes: build robots cheap → ship many → collect massive amounts of real-world physical interaction data → use that data to train better models → produce even better robots. This is exactly the same loop that made Bitcoin’s proof-of-work secure: more miners → more hash power → more security → more miners. In crypto, the value accrues to the network. In Yuzhu’s case, the value accrues to the company’s proprietary data. But the mechanism is identical: a virtuous cycle that becomes harder to disrupt the larger it grows.
During my years auditing early Ethereum projects, I saw many whitepapers promise a “data flywheel” but deliver nothing more than a token sale. Yuzhu, by contrast, has actual hardware in the field. Its external components are only 10-20% of the bill of materials, meaning the company controls its supply chain with a level of self-reliance that blockchain purists admire. Democracy isn’t a transaction where every voice holds weight. But here, the “voice” is the physical data from each robot. The more robots Yuzhu sells, the more data it collects, and the more it can improve its models. In a world where AI models are trained on static datasets, Yuzhu is creating a dynamic, self-improving system.
But this is where the contrarian voice needs to speak up. The data flywheel only works if the data is actually useful for the next stage of the technology. Right now, most of Yuzhu’s shipments go to research labs, university education, and government procurement. Those are “demo” and “experiment” use cases, not industrial productivity. The robots are performing tasks like walking, balancing, and simple object manipulation. The data they generate is rich but limited in diversity. Compare that to the industrial environments where robots need to weld, assemble, or navigate chaotic warehouses. The data from a university lab may not translate to a factory floor. Code is the new conscience. But if the code is trained on the wrong data, it becomes a liability.
Nomura’s revenue forecast of 122% CAGR assumes a massive leap into industrial repeat orders. The report shows a curious jump: 2027 revenue growth accelerates to 101% from 58% in 2026. That kind of hockey-stick shape usually signals a single catalyst—a large customer, a new product, or a regulatory change. The report does not disclose what that catalyst is. As someone who has seen the 2021 DeFi bubble inflate and pop, I recognize the pattern. Back then, protocols promised “future revenue” from lending fees, but the actual TVL was concentrated in a few whales. Yuzhu’s industrial adoption might be concentrated in a few early adopters, not mass market.
Decentralization is a verb, not a noun. Yuzhu’s vertical integration is the opposite of decentralization. It owns the entire stack—motors, sensors, software, assembly. That gives it cost advantages but also makes it a single point of failure. If the company’s self-driving algorithms fail to generalize to industrial tasks, the entire flywheel stalls. In blockchain, we learned that permissionless innovation wins in the long run. Yuzhu’s closed architecture might be fast now, but it could be outpaced by a consortium of smaller players using open-source hardware and decentralized data markets.
The report also sidesteps the elephant in the room: Chinese competitors. Nomura claims Yuzhu is “global first” in shipments, but it does not compare with domestic rivals like Zhiyuan Robotics or UBTECH. These companies are also shipping humanoid robots, and they may be closing the cost gap. In crypto, we saw how Ethereum’s first-mover advantage was eroded by faster, cheaper L2s. The same could happen here if Yuzhu’s cost advantage is not structural but temporary.
My takeaway, after 28 years watching technology cycles: Yuzhu is a fascinating case study in hardware flywheels, but it is not a “buy the destination” story. It’s a “buy the transition” story. The company is proving that humanoid robots can be built, sold, and improved at scale. That alone is remarkable. But the next step—turning those robots into industrial workhorses—is where the real value lies. And that step requires a leap of faith that the data from labs will generalize to factories. If you believe in the flywheel, you can buy the vision. But remember: even in the most robust blockchains, the data must be validated by the network. Yuzhu’s data is validated by its own balance sheet. That’s a different kind of trust.