With apprentice-level capabilities developed domestically yet exorbitant prices charged overseas: As physical AI ventures into global markets, why is the relentless drive for efficiency turning out to be increasingly risky?
In recent intensive exchanges with multiple enterprises, the anxieties and reflections of all parties have ultimately pointed to the same proposition: Where exactly should the value realization of Physical AI be completed?
Combining these first-hand insights, I have drawn three judgments that are completely different from the inertia of hardware going overseas over the past decade:
First, spatial mismatch: the domestic market is a training ground, while overseas is the real business market;
Second, anchor switching: bid farewell to cost-effectiveness based on material costs, and shift to smart premium based on overseas labor costs;
Third, benefit transfer: shift from simply exporting products overseas to deep local integration, and transferring benefits and equities to local stakeholders.
This article will analyze around these three judgments.
Training Ground and Business Market
Let's first look at a somewhat fragmented industrial phenomenon.
As a much-watched branch of Physical AI, embodied intelligence is experiencing unprecedented capital enthusiasm in China, with intensive financing coming one after another, and dynamic demonstrations at various exhibitions emerging in endlessly. However, if we turn our attention to the real front line of the manufacturing industry, factory decision-makers have an exceptionally calm attitude towards procurement.
An entrepreneur once calculated an account for me: the hourly wage of production line workers in second-tier cities is about 18 to 22 yuan; factories generally require a payback period of 8 months for purchasing automation equipment, and almost no one accepts a payback period exceeding 24 months. According to rough calculations based on his figures, if a 300,000-yuan robot wants to pay for itself within 24 months, it needs to replace 12,500 yuan of labor cost per month. Converted at an hourly wage of 20 yuan, that is 625 working hours, equivalent to the single-shift workload of three and a half workers, not counting operation and maintenance costs and electricity bills.
His conclusion is straightforward: if embodied intelligence only targets the domestic market, it will be difficult to achieve positive gross profit within five years, unless the revenue mainly comes from government subsidies.
The same robot placed in Japan, Europe or the United States will have a completely different account book. The local labor cost is several times that of China, and labor shortage is a normal state. The payback period of a robot that can work stably will be greatly shortened.
As a training ground, China is almost irreplaceable. It has dense manufacturing scenarios, a complete supply chain, and a large number of engineers capable of rapid iteration. Problems found at one workstation today can be modified and retested next week.
For robots to move from being demo-ready to delivery-ready, repeated trials and tests in real scenarios are required, and such opportunities are most easily obtained in China.
In a batch of early-stage projects I saw recently, a phenomenon is very illustrative: the companies that take the lead in generating revenue are often the ones that help robots "learn to work", focusing on pilot-scale verification, data collection, testing and evaluation, and core sensors, while robot manufacturers lag behind.
The logic is very simple: robot companies have money and are in a hurry to implement their products, so they are willing to pay to make robots learn to work. I once called this type of work apprenticeship working hours in my article: robots work for manufacturers and laboratories, producing data and capabilities, and the paying party is the one that wants the robot to learn to work. What the domestic market buys the most today is exactly the apprenticeship working hours.
Thus, a new division of labor is taking shape: train in China, and collect revenue overseas.
The path for the previous generation of hardware to go overseas was to verify in the domestic market, then replicate and expand overseas. For embodied intelligence, overseas may be the "only" profitable market, and the statement of "the second market outside China" needs to be updated.
This means that for ontology robot companies, going overseas is no longer an option after growing up, but a prerequisite that must be written into the company's design from day one. I have already seen entrepreneurs who hardly do any domestic business.
Cost-effectiveness and High Premium
After clarifying that overseas is the real business market, the core proposition that follows is pricing. After going global, what price should we set for our products?
Over the past two decades, the perception label of Chinese hardware in overseas markets has been highly solidified: focusing on "cost-effectiveness". To put it bluntly, it means using similar product quality to exchange for market share by lowering prices.
This label is loosening. A founder who has been working hard overseas for many years has an observation: Few people believe that Chinese hardware has excellent quality when you say that overseas; almost no one objects when you say it is more intelligent.
Electric vehicles, sweeping robots, drones... have made intelligence a striking label for Chinese hardware for the first time, and AI will continue to amplify this label.
The label has changed, but many founders' pricing habits still remain in the last era.
He gave an example of lawn mowers. This is a mature category, the mainstream products of international brands sell for about 900 US dollars, and the typical practice of the previous generation of domestic enterprises is to make products with similar quality and sell them for 400 US dollars.
They did the opposite, introducing the mature lidar technology used in automobiles into lawn mowers, solving the borderless problem that has plagued the industry for many years, that is, no need to pre-buried boundary lines under the lawn, and then set the price to more than twice that of international brands. According to him, the products sell very well.
Behind this is a flywheel: only sufficiently innovative products can be priced high; with high pricing, there is enough gross profit to expand the global channel; scale and profit will in turn support the next round of R&D.
The starting point of the flywheel is precisely the step that founders are most unaccustomed to: daring to set the highest price in the world.
We are too used to exchanging share with diligence. First, we relied on the supply chain dividend, and then on the engineer dividend. But now engineers are also getting more expensive, and the share obtained only by low price cannot support sustainable innovation.
Whether it is an intelligent lawn mower on the lawn or embodied intelligence on the production line, in essence, they are all Physical AI that integrates AI into hardware and works in the real world.
The pricing reference for a robot overseas should be the local wage it replaces. Pricing for the overseas market with the domestic cost logic is equivalent to giving up the future of sustainable development.
I analyzed in a previous article that the valuation unit of embodied intelligence should be changed from the number of units shipped to the working hours on duty, and the same logic applies to pricing: Pricing by unit takes materials as the anchor; pricing by working hours takes wages as the anchor.
In my opinion, switching the anchor point and logic of pricing is the most daring innovation that domestic enterprises are reluctant to make.
High Efficiency and Safety Cushion
After switching the pricing anchor, another deeper challenge emerges: how to make the overseas market accept the enterprise's commercial success in the long run, so that this high gross profit can be realized continuously?
There is a story that impressed me deeply. A foreign investment promotion official in Southeast Asia said to visiting entrepreneurs: "Your engineering team built the longest bridge in our local area, with good quality and fast speed. But from design to construction, even down to every nail, you use your own resources, and even the lunch boxes for workers are shipped from China. After the bridge is completed, nothing is left to our local community." For this reason, countries are now tightening their investment attraction policies, no longer welcoming this kind of "all-inclusive" Chinese projects, but forcing the bundling of local interests.
The poignant part of this story is that these enterprises did nothing wrong. Priority to efficiency is the very force that pushed Chinese manufacturing to where it is today, but overseas, it may turn into the biggest risk.
The reason is not complicated. In 2025, China's trade in goods surplus has exceeded 1 trillion US dollars. Many countries want to import Chinese products, but are increasingly unable to find products they can sell to China.
When a foreigner only brings products and does not leave jobs and industries, the more economically successful it is, the more politically vulnerable it becomes. Elected governments have to account to their voters, who care about jobs, while price and efficiency rank behind.
A founder who has deployed business in Europe gave another solution. He cooperated with a local century-old established factory for production, used the sub-brand of the local factory, helped them hire new workers, and made use of their existing service network.
His summary is very incisive: Faced with increasingly high localization thresholds, proving your innocence on your own is always passive. A better situation is to have local partners willing to stand up and speak for you.
Another entrepreneur thought more thoroughly. He is already prepared that when he enters Europe and the United States to carry out large-scale business in the future, he may need to set up a joint venture with local enterprises that have deep local connections, with the local party holding the controlling stake. He said this is very similar to the arrangement when foreign-funded car companies entered China in the past: they could only form joint ventures with local enterprises, and the foreign shareholding ratio could not exceed 50%. If one day the local shareholding ratio can gradually decrease, it will be a gain for him.
This makes me realize that the end point of the new generation of enterprises going overseas may change from occupying the market to designing benefit distribution: which profits are left to local partners, which positions are reserved for local employees, which links are handed over to local suppliers... Turning part of the profits into local people's wages and shares is a more lasting protection than any tariff negotiation.
The core of the previous generation of overseas expansion is one-way product export; this generation of overseas expansion must learn to transfer equity and build a community of shared interests with local stakeholders.
Final Notes
To sum up, the above three judgments are actually three progressive stages of the same commercialization path: take the domestic market as the training ground to solve the engineering problem of "whether the product can be delivered"; take the overseas market as the business market to solve the profitability problem of "how to realize the value of technology"; build a safety cushion with equity and profits to solve the survival problem of "whether the business model can be sustainable".
After more than ten years of industrial research, I have seen too many companies with leading technology that eventually collapsed due to mismatched commercial structures.
In this wave of Physical AI, Chinese enterprises have excellent cards in their hands: dense manufacturing scenarios, the globally recognized "intelligent" label, and extreme engineering capabilities. But good cards do not mean ultimate victory.
The key to winning the next stage may lie in how entrepreneurs answer three more essential business questions: where to find the real paying customers, whether they dare to set the highest price in the world, and whether they are willing to transfer benefits to achieve symbiosis with local communities.
This article is from the WeChat official account "IoT Think Tank" (ID: iot101), written by Peng Zhao, published with authorization from 36Kr.