Huang Jinping, founding partner and chairman of Winreal Investment, has spent 15 years investing in hard technology. Before becoming an investor, he worked in China’s wireless communications industry, joining ZTE in the 1990s and spending 15 years there as the sector grew from an emerging market into a global force.
His investments reflect that technical background, spanning telecommunications, smart vehicles, robotics, electric vertical takeoff and landing aircraft, and extended reality eyewear. Winreal’s portfolio companies include Unitree Robotics, Galbot, RayNeo, Vertaxi, Lante Optics, and Jiangnan New Material Technology.
Unitree is among its most prominent investments. Winreal joined the robotics company’s Series A round in 2021 and invested again in subsequent rounds. Yet as capital has poured into embodied intelligence, Winreal has taken a more cautious view of the broader sector.
Huang said many embodied intelligence startups have quickly reached valuations above RMB 10 billion (USD 1.5 billion), with some preparing listings on the Star Market or in Hong Kong.
His framework for evaluating them comes down to three questions:
- Can the technology work?
- Can it be engineered into a product?
- And can it support a viable business?
His concern is that the current rush into embodied intelligence places too much emphasis on short-term valuation gains and too little on the long cycles required to refine hardware, supply chains, and real-world deployment.
He offers a simple example. If 20 robotics companies eventually divide a RMB 500 billion (USD 74.1 billion) market, perhaps three will perform particularly well, companies ranked fourth through tenth may remain viable, and the rest may struggle to survive.
Easy access to capital can create another problem, Huang said. Founders who raise too much money too quickly may overestimate their capabilities, while investor expectations can push operating decisions away from a company’s long-term path.
The following transcript has been edited and consolidated for brevity and clarity.
36Kr: Why did Winreal invest in Unitree early?
Huang Jinping (HJ): When we invested in Unitree, the boom surrounding large models had not started. We simply believed in the long-term value of robotics and spent time reviewing most of the relevant projects on the market.
We first asked whether the sector itself had long-term value. Commercialization was still unclear, but we believed in the underlying technology and future market. Once that premise held, we compared companies across multiple dimensions.
Two things about Wang Xingxing stood out. First was his near-obsession with robotics technology. Whether quadrupeds at the beginning or humanoids later, developing the robot body carries a high technical threshold. Wang approached it like an enthusiast, willing to keep refining the product and push performance.
At the time, the supply chain was immature. Major suppliers were reluctant to serve niche robotics manufacturers, so many components and debugging tasks had to be handled in-house. That level of technical obsession mattered.
Second was unusually strong cost discipline. He scrutinized hardware production costs as well as the company’s overall operating expenses. That is uncommon among young founders.
So our approach was: select the sector, assess whether the technical route was viable, compare teams horizontally, and then back Wang. There was also luck involved. The later surge in large models gave Unitree a major boost.
36Kr: Unitree was still focused on quadruped robots at the time. Was there much disagreement inside Winreal about the investment?
HJ: There was little disagreement. We not only invested ourselves, but also introduced Unitree to more than 20 institutions we knew. None followed us into the round. By the time some wanted to invest later, the window had closed.
We did not look only at whether the technology worked. We also asked where the market was. Unitree’s electric actuation approach could materially reduce hardware costs compared with Boston Dynamics’ expensive hydraulic route, and the technology was transferable to humanoid robots. That gave it a clearer path toward engineering and commercialization.
But the near-term market for quadruped robots was admittedly small. Inspection tasks were one use case, while heavy-duty and industrial applications were still being explored.
Many financially oriented investors focused mainly on current order volume and lacked a framework for judging the long-term value of the technical route. That was the fundamental difference in our decision-making.
36Kr: How has Unitree’s competitive position changed in recent years?
HJ: In the early stage, a company relies on its technical route, product engineering, and market capabilities, and gradually builds a brand. Unitree followed that pattern.
What we have discussed with Unitree is that the next stage is about turning the best suppliers, frontier technologies, and high-quality industrial resources into partners. The competitive barrier gradually shifts toward an ecosystem-level advantage.
Building that ecosystem requires capital, especially corporate venture capital, to help mobilize resources. The moat increasingly depends on whether you can bring the best partners together.
36Kr: How has Wang Xingxing changed as an entrepreneur?
HJ: Over the years, he has developed his own business philosophy. As Unitree gained attention, outside distractions increased sharply, which created a real management challenge.
There were benefits too. Unitree spent very little on advertising but received a huge brand dividend, with resources and partners approaching the company proactively.
The team has handled that period well. Management remained disciplined, the company stayed focused on its core business, and it was not pulled off course by outside noise. By the time the technology and products matured and demand arrived, Unitree was prepared and well funded.
Ultimately, it comes back to the founder’s ability to manage changes in the industry and the company’s growth pace. On that front, Wang’s discipline and judgment are unusually strong among comparable founders.
36Kr: Did Unitree grow faster than you expected in terms of investment return?
HJ: Yes. Much of the upside came from external factors. We did not expect embodied intelligence to develop this quickly, and advances in large models accelerated the sector.
But outside observers tend to overestimate how quickly technology can be deployed. Autonomous driving has taken years even though driving is a relatively standardized task with limited control dimensions and comparatively clear rules.
Robotics is much more complex. To work in real environments, the industry still lacks enough data, and model capabilities have a long way to go. These problems can be solved, but they require patience and time.
When enthusiasm and capital flood into a sector in a short period, the market tends to assume development will happen faster than it actually can. Rapid valuation increases are one expression of those inflated expectations.
36Kr: How do you judge whether a robotics company can survive and keep building over time?
HJ: Technical feasibility and commercial competitiveness have to be validated over a long cycle.
We want to invest in leaders. Another common strategy is to invest in several companies outside the leaders on the assumption that, as long as the sector stays hot, all their valuations will rise and investors can exit later at a profit.
That is essentially an arbitrage approach imported from public markets. It assumes capital enthusiasm will keep spilling over and lifting nonleaders as well.
From an industrial perspective, that assumption does not hold. Leaders tend to become stronger, while midtier and lower-tier companies diverge over three to five years. Many companies can reach high valuations without their businesses keeping pace, and their valuations eventually have to fall.
Before 2023, the robotics path was less clear, and few teams had both a workable technical route and strong engineering capabilities. We believed Unitree’s leading position would strengthen.
After embodied intelligence took off in 2023, more teams and more capital entered. Our view of the industry’s eventual structure did not change. We have invested in some vertical companies around the upstream and downstream supply chain, but the principle remains to back leaders.
This is not a simple either-or choice. The short-term market tests whether a company can survive; the long-term sector determines how much room it has to grow. Sustainable companies need both.
36Kr: Do high valuations hide the disadvantages of young embodied intelligence startups compared with companies such as Unitree that have lived through a full industry cycle?
HJ: A short-term valuation is not the same thing as a company’s true operating condition or technical strength.
We often compare company building with a child growing up. You cannot skip the necessary stages. Even exceptional teams develop weaknesses if they bypass stages that normally build capability.
Founders who find fundraising too easy can overestimate themselves and make decisions that deviate from their long-term path. Once they raise large amounts, they carry investors’ performance expectations and may distort operations to produce attractive short-term results.
Unitree entered the field in 2016 and spent years refining its technology and products. That accumulated capability came from repeated trial, error, and iteration; it cannot be compressed into a short period.
In the early years, the upstream supply chain was immature, key components lacked established suppliers, and manufacturers were reluctant to cooperate. Unitree had to develop components itself, which also helped it control costs.
A team that has existed for only a year or two may not even have fully solved the complete robot. It is difficult to see how it could simultaneously develop critical components in-house.
Many new teams are assembled quickly and look strong on paper, but they have not had enough time to learn how to work together or survive hard setbacks. When market conditions shift or supply chains break, hidden problems emerge at once.
Company building cannot be rushed. Teams that have not lived through a cycle often struggle when conditions become genuinely difficult.
36Kr: Is the shift toward high valuations and IPOs another sign that capital is distorting company behavior?
HJ: To some extent, yes.
36Kr: What would a healthier capital cycle look like?
HJ: In a normal situation, the best sequence is straightforward: a company develops good technology, turns it into a good product that the market accepts, raises a reasonable amount of money to support iteration and expansion, and then builds the business.
Reality often deviates in two directions. One is that a good company is misunderstood by capital and cannot raise enough, so it develops very slowly. The other is that money arrives too quickly and in excessive amounts, especially across many companies at once, and companies start competing with one another in distorted ways.
Different investors also have different expectations. Some think like industrial investors and are prepared to accompany a company for the long term. Others think more like financial traders: when a theme rises, they rush in and hope to sell quickly at a profit.
That can backfire. A market bubble may push valuations high without companies creating equivalent value. If an overheated sector cannot produce business performance, valuations fall, competition compresses gross margins, profits disappear, and growth becomes harder to sustain.
Consider a hypothetical RMB 500 billion robotics market shared by 20 companies. Ideally, everyone would do well. Realistically, perhaps three would do extremely well, companies four through ten would survive reasonably well, and the remainder might not.
The problem is that investors often cannot tell today who will finish in the top three or top ten. Even if they can, they may not be able to buy into those companies. So they place bets on companies ranked 11th through 20th, partly because market pressure makes it uncomfortable to have no exposure to embodied intelligence.
36Kr: Why do institutions still invest in midtier companies when the sector looks promising but access to leaders is limited?
HJ: There are two practical constraints.
First, allocations in leading companies are often locked up by early investors, leaving later institutions no way in.
Second, in the previous two years, early valuations for midtier companies were low enough to leave room for a tactical trade.
Professional institutions can often estimate the long-term ceiling of most companies reasonably well. The problem is not always that they cannot see the limits. Sometimes they see them clearly but cannot invest in the leaders, so they settle for companies lower down the rankings.
As more capital keeps pouring in, even valuations at the tail end of the market are rising rapidly, which makes the bubble larger.
Once several leaders are public, the market should become clearer. The timing will depend on capital markets, policy, and technical deployment. When leading companies prove their business models and generate revenue at scale, capital will concentrate further in them and financing will become much harder for the middle of the field.
I expect that divergence to become increasingly visible over the next two to three years.
36Kr: Does that help explain why the leading embodied intelligence companies keep raising more money?
HJ: Yes. Investors see brand, capital, and a solid team, so they view the company as safer and better positioned.
36Kr: Where is the market bubble most obvious?
HJ: One characteristic of embodied intelligence is that some companies valued at more than RMB 10 billion may have only tens of millions of RMB in actual revenue. That gap is one reason the sector is criticized for having a large bubble.
But if a company has chosen the right direction and eventually creates very large value, a short-term bubble can gradually be absorbed.
36Kr: How long might that absorption take, and when could the market become more rational?
HJ: Many companies may never find the right market entry point. They can run out of capital without establishing a sustainable commercial loop. Others will outperform expectations. Polarization will inevitably set in over the next several years.
In terms of asset allocation, suppose there are eventually one billion robots globally with an average price of RMB 10,000 (USD 1,482.3). That implies a market measured in the tens of trillions of RMB, so current investment can make sense over a long horizon.
But that value will not be distributed evenly. Some players will break out, while others will be eliminated.
Capital does not yet have enough information to identify the ultimate winners precisely, especially from fourth place through 20th. That is a probability game, not a deterministic forecast.
Even companies or technical routes that fail will leave behind engineers and operators with real experience. Those people can contribute to the next cycle of innovation.
36Kr: Which developments in embodied intelligence do you see as inevitable, and which remain uncertain?
HJ: We often discuss one internal view: a general model for embodied intelligence will eventually break through. The uncertainty is timing.
For robotics companies to open a real market, they still have to return to the product. Start with simple tasks, solve real problems in specific environments, and then move gradually toward generalization. Otherwise, customers have no reason to pay.
Which scenarios become the biggest opportunities is still changing. That is where founders and investors need judgment and a longer time horizon.
36Kr: Which application scenarios look most attractive to Winreal?
HJ: We pay more attention to vertical applications.
Solving a specific problem in a vertical market and creating verifiable commercial value is a clearer route and can create a positive feedback loop.
Autonomous driving offers a useful comparison. A vehicle is also an embodied intelligent device, but driving is more standardized, so companies can collect large amounts of effective data and progress more quickly.
In robotics, models are improving but data remains scarce, and collecting robot data is much harder than collecting vehicle data. That creates an opening for vertical companies.
A vertical player can enter a specific environment, accumulate data there, and use an appropriate model to make the robot work in a limited setting first. General-purpose embodied intelligence companies may struggle to go that deep.
A vertical company that spends years inside one market can build a real barrier through scenario data, customer understanding, and a stable sales network.
36Kr: How does the investment approach differ between vertical and general-purpose embodied intelligence?
HJ: The competitive dynamics are completely different.
General-purpose robots compete on model generalization and hardware capability. The advantage goes to whoever can make a robot do a wider range of tasks sooner.
Vertical applications compete on specialization. A robot designed to clean bathrooms, for example, may have a different form factor from a general-purpose humanoid, collect data differently, and use different sales channels.
It may need to be sold to property management or cleaning companies rather than through consumer electronics channels. That kind of know-how takes years to accumulate.
The deeper a company goes into the scenario, the higher the barrier can become. Once it has enough scenario data, understands customers’ actual needs, and has a stable sales network, it becomes difficult for late entrants to compete.
The vertical model has more in common with consumer businesses than with a purely technology-driven general platform.
36Kr: Unitree has listed, while several others are preparing IPOs. Will companies that miss the current listing window be at a meaningful disadvantage?
HJ: Broadly, yes, but it’s more complicated than the phrase “listing window” suggests.
An IPO is an important risk control milestone. Once a company lists, one of the biggest pressures on founders and early investors, repurchase obligations, is removed. Regardless of the share price, they no longer face the same risk of having to repay investors.
That is one reason companies may rush to list even when revenue is still limited. They are securing a kind of safety cushion.
But a company that has spent heavily without achieving a substantive business breakthrough will still face major operating pressure after listing. The pressure simply shifts from repurchase obligations to market capitalization management.
For investors, an IPO provides liquidity. It does not prove that the investment thesis has worked. The real determinant of company value is still whether the business itself works.
So the listing window matters, but getting through it does not solve the fundamental problem. The real dividing line is whether a company can remain competitive after it reaches the public market.
36Kr: What advice would you give investors venturing into robotics from other areas.
HJ: Return to industry fundamentals. Ask what problem the company solves and whether solving that problem has commercial value. That does not change.
If you can remain patient when everyone else loses patience, that can create differentiation. But once a direction becomes consensus, be more cautious. Consensus brings more money, higher valuations, and often more risk of developing into a market bubble.
Public markets make this easy to see. When memory chips and optical modules rallied sharply in the first half of the year, people who bought them felt brilliant and those who did not felt they had missed out. When the market reversed, it moved quickly.
Private markets are not fundamentally different. When everyone chases the same direction, prices are high and expectations are already aggressive. If you are pushed into backing the third- or fourth-ranked company, the eventual return may still be poor.
It is possible to pick a dark horse. But you need a clear reason why that company can become the exception and why it can win. Picking a dark horse is much harder than picking the leader.
KrASIA features translated and adapted content that was originally published by 36Kr. This article was written by Huang Nan for 36Kr.
Note: RMB figures are converted to USD at rates of RMB 6.75 = USD 1 based on estimates as of August 20, 2026, unless otherwise stated. USD conversions are presented for ease of reference and may not fully match prevailing exchange rates.