How Shenzhen Actually Works: A Ground-Level Primer

So you've watched so many western founders come to Shenzhen to build their dream products. So what is the most common mistake that you've seen them? We try and convince people not to build hardware from the service. Shenzhen's competitive advantage is cost. But beyond cost, what is the true advantage of Shenzhen's capability? Eric Pan started Seeed Studio in 2008 selling hardware from his apartments. Today, the company behind Reachy Mini's production and NVIDIA's robotics partner and the AI hardware partner banner all over Shenzhen is telling the next generation of Western founders to stop making hardware. We're in Shenzhen to find out why. For a startup to build a trust of over like hundreds of different suppliers is nightmare.

Hi Eric. So from the inside, how would you actually describe the business today in 2026? So we started with open source hardware and we still do so but it's evolving all the way from modules to devices now more into solution systems and we see ourselves as a platform to incubate more hardware ideas into real products and you've described the business as two-thirds of prototyping for early stage startups and a third for scale up production. So can you tell me more how that scale up production part actually work and how would a founder work with a team like yourself? Sure, I think it's actually a loop. We provide the open source modules and devices devices as the reference design modules as the core components. So people will buy them and do the prototypings to try them out to infuse your ideas, software, data into the creation. Then the best option is come back to us and to scale that because we have all the supply chains, we have all the expertise. we just need to like change design files, remove some components for them or remix several projects together to make the into their creations. So it's very naturally growing from the two-thirds of prototyping into one sort of scaling up.

So the Reachy Mini story where the Pollen Robotics' demand spiked five times more than they expected. So how did you help the team and how did you work with the founder to Yeah, we know Pollen Robotics for many years and they have been using our like speaker which is microphone array from the very beginning. Okay. Yeah. So it's very organically they evolve into a device that they want to manufacture. They ask us for help and we respect this project very much. So why not and we host them as of our team. So they come to Shenzhen stay here for 3 months and in this office or in this office and we put our engineers to work them together like side by side. Basically we help them to re-engineer many part of the design because we know the manufacturing better and uh it's more like a engineering process. They do the EVT. We help them to like fine-tune the designs. We talked to the supply chain knew the restrictions and we had a lot of injection moldings or stampings for different kind of the parts because they have very clear goal how many they going to ship or how fast they should be doing that and we put our project management team engineers in full speed to support this happen and it was kind of magic because they are the right team to work with. They know us they trust us. So even it's two companies we work as one team.

In terms of the supply chain partners, are they all over Shenzhen or do you have like a set you know list of people that you go to for like various components? Of course we have uh SRM like a supply chain relation management and we have the majority of them are around Shenzhen and Dongguan but we still buy parts from like east of China from all over the world. Mhm. What would have happened to Pollen Robotics if they didn't work with you guys? Like how would they meet that demand? I don't know. There can be many faults. They can be like uh if they go to all the supply chain themselves because to any of the supplier they need a long time to trust the customers as well. Yeah. Not not only vice versa but they need to understand you this is a startup but they are not nobody. They can pay on time.

They know the manufacturer and respect the process. So it's taking a long time to build a trust but if for a startup to build a trust over like hundreds of different suppliers it's nightmare. It's not only about the budget it's also about the understanding of each other. understanding of what's limitations on manufacturing. Why we cannot do this way? It's also on the quality management. How do I set the quality to make sure the factories knows what to deliver? Also, it's about the lead time. Lead time is actually most of the lead time is about waiting because the manufacturer cannot be empty until you come. They always have some cues. But why would they prioritize your projects? So, it's a lot of uh like orchestration is in between and we are doing this job for 18 years. We have our own product.

We have own supply chains. We have own factories. So that's make this life much easier. Mhm. Yeah. And you have this like close relationship with various factories that you've worked with over the many years. And so it kind of builds that trust over time and when the startups go to you guys and then to the factories then there's one layer of support. Exactly. We are more like a future or middle ground of the trust because we need to understand the customers as well. Like we have selection of our customers very much. We don't do consumer electronics because we think that's going to be very fierce competitions. Yeah. Has to be very large scale that the corporations might be taking over. So we focusing on emerging technology.

We focusing on more vertical or industry oriented stuff. Mhm. For sure. So you've watched so many Western founders come to Shenzhen to build their dream products. So what is the most common mistake that you've seen them make? Well, uh the mistakes are very uncommon. Each team have their like restrictions. Some of the like Apple experts they working for Apple or Microsoft for hardware they come to Shenzhen and they are demanding very much like a corporation but they have very small volume and they have a very small supplier which is previously they are working for Apple. So their methodology will fail. They have to change the methodology to more agile more collaborative relationship based not like contract based with a lot of demanding way. Mhm.

So this is one of common mistakes for the hardware background people. Yeah. But for the software background people they don't know manufacturing at all. So they will think okay if it should be like a coding I compile in five minutes and I can change it. But for hardware and changes might take weeks. So you have to be have a lot of simulations understandings of the process. We call DFM (design for manufacturing). But for small teams it's not designed from manufacturer. It's designed from manufacturer. If you're Apple they the factories should be you care for them. you change according to the needs. But if you are a startup, the best way is not to design from scratch. It's design from what's already out there. Interesting. For example, if you want to design a tablet, you go to a tablet factory, see what's they have been working around and to change from that. Give it a motor so you can move around or give it different extensions.

But don't build from scratch. Yeah. So that's I think it should be a key lesson for the software team is look what's already out there and design from that. So it seems like a lot of people tend to come to Shenzhen with a certain expectation of okay this is the prototype that is from the 3D modeling and then like make it happen and so it seems like it doesn't really work like that way where it's better to see on the ground what's available and then do the design. I think it's not a common mistake but it's a common takeaway is people should come to center with openness. They should let go of their ego. Mhm. They should respect more of what's already out here. The engineers are very expertise. Even the workers they have the experts. Yeah. So, respect for their way they doing things.

Yeah. By default, they're not doing wrong. By default, the creators might be wrong. Mhm. Yeah. I think that's the best way to avoid the mistake is competition more often to see already what's happening. And another way to overpass the mistakes are don't do hardware. Mhm. From 10 years ago, we trying to have people to build their hardware, but 10 years later, we're trying to convince people not to build hardware. Tell me more about that. Yeah. Because you are software team. You don't need to build the hardware from scratch. Mhm. You should look at because why we do open source hardware is for people to take care them as a reference design. Mhm. So you have the reference design, you have the 3D printed enclosure. That's good enough. You don't need to compete with the world on the hardware. The software team usually have their expertise on the like functions on the user experiences on the vertical knowledges. That's their best competency.

Yeah. So they should focus on that and using a more standardized hardware. Don't put too much efforts like to make this hardware good enough as Apple. it's not going to happen. So they should f really understand who they are and come to Senzhen to find the right small enough complimentaries to their business. Yeah. And why do you think that there are so many hardware founders that are from software background that is trying to innovate on the hardware side? Because the open source hardware is not a secret, right? Like it's everyone knows that they exist and why don't more people use that in your opinion? You're saying that you're trying to make people make hardware less and focus on the software. So what do you think is stopping these people from focusing on the software and still go to the hardware side and try to innovate there?

I think it's because most of the venture capitalist because VC wants to invest in hardware and for more than 10 years people think oh hardware might be the new software and they collapse. Yeah. And now I think for this trend we see a lot of drive is because of AI is eating software. Yeah. Yeah, you panic and want to find a standing ground where hardware might be that but uh I think it's not about software and hardware. It's more about public domain technology or like vertical domain technology. So people should have the little one. Yeah. If you on the public domain technology, you should be working for Apple or like Meta or Anthropic. Can you share more about what you mean by public domain knowledge?

Public domain knowledge is you want to make a device that everyone knows how to make and everyone wants to use that. I see. It's like iPhone the next iPhone. I think most of the team who wants to build the next iPhone will fail. Most of them most of the companies even most of hardware companies will fail on that. But for the verticals for the more standard longtail market, there are so many interesting things need to be renovated. Mhm. Can you tell me a specific story, you don't have to share the name of a western founder that came to Shenzhen and failed. What did they think they understood and what did they actually miss? Yeah, I too many of them. Yeah, but a common story is like a western founder they have software knowledge they have a lot of budget they come to send they find a design house and they want to build this into scale very fast but they don't know what will be the limitations and so they go to the factory point at all variet of things because they pay up good money everybody listen to them but they are not giving the right guidances in the end the manufacturing will be slowed down will be too costly they don't know how to set up the expectations about the quality. So the team will be deliver something like without clear understanding of the requirements. They don't know how to command the requirements and they don't listen.

Yeah. So this is a very high failure risk. People will have very ideal expectation because they injen so they punch the factory too hard and they push to the limitations sometimes beyond the limitations. Yeah. So the hardware cannot be made. In the end they don't ship. They have a lot of investment or back orders but they cannot ship and they failed and they broke their reputations. Yeah, this happens a lot but this happens often in the consumer electronics because for consumer electronics everybody is trusting the wave and everybody is fighting for the wave. They have to be fast. Yeah. They have to be differentiating. Mhm. But I think most of them are failing. So we don't want to have them. Yeah. You mentioned a point earlier about building trust. You know let's say I'm someone that's coming to Janjun to build my dream product. How would you advise me to build trust with the partners that I'm working with like locally like a factories and basically the whole part of the supply chain?

I think the best way is first to come as a student as if you are student. You are like a role climber but you are the first time to the mountain. You have to trust your coach, trust your guides. go to the factories to see what's happening to gain as much knowledge as possible and go to Hang Bay to see what's already been built, understand the methodologies, understand the material they processor use and come back and to rethink your do I really want to build this? Can I just tolerate what's already out there? It's not perfect, but should I just change a little bit to be different? Yeah. Don't build it from scratch, but rebuild. You are like shopping in the whole center area. Yeah.

Find the right ground to start with. So be a student and go on the ground and pick from what's available and try not to change everything from the start. Exactly. So there's been a lot of talks in the west about reshoring. So bringing back manufacturing facilities and capabilities back to the west. And from the surface we all know that Shenzhen's competitive advantage is you know one of them is cost right? Everything is a fraction or a hundth of what it will cost in the west. But beyond cost, what is the true advantage of Shenzhen's capability? We usually see it as the three parts. Quality, cost, and lead time. Mhm.

So, and this impossible triangle. If you need a cost, usually you need to have a very n time to prepare for that. Senzen, I think people are talking about cost, but under the cost is all the engineering complexities. Mhm. It's a full like a tree of supplies that it needs to be responsive. So is like a jungle with enough biodiversityities. You can find engineer with all kinds of backgrounds, all kinds of skill sets. So if you want hardware, you can find the right spaces to remingle to whatever you want. I think that's first reflecting to speed. Secondly, cost really matters to quality and volume. If you have small volume, you can expect the cost to be low anywhere.

But in Shenzhen, you can get a relatively much lower cost on the small volume and bigger volume because there's enough competitions. There's enough demand and supplies. I think really matters about resharing is can we replicate the demand and supply in United States, Europe or anywhere in we are facing the global demand and we are trying to supply according to that needs. Mhm. So I think in short term you cannot just remove the whole jungle and to redeploy that on the Mars. That's not going to happen. Of course the restoring can happen on some stages for semi assembled products. You redo that to recombine them into the finished product. That can happen in Vietnam in the states in Japan. We're trying to do that as well.

Mhm. So the global demand coming to Shenzhen is enabling that speed, that lead time, the cost advantage. So the question is how to replicate that elsewhere anywhere where the cost is so high to make a certain components especially the bespoke ones. So tell me about rebots. You mentioned that you're trying to uh have more people make less hardware and focus on the software side. And tell me how you decide to build rebots and the story behind. Yeah. uh for robot is kind of u how to say unexpected. We didn't plan to do robots in the very beginning.

We focus on the open source hardware is on at the AI part. Mhm. But gradually we made the uh edge computing like JSON. So people come back to us looking for more parts around the computations. Mhm. But we like robot so much. Our team say okay how about we make a robot arm. I say no we don't we're not good at robot arms. But in the end we it's like internal alpha projects. It's like for hobists for internal explorations. After a few months we we share that on GitHub say okay would you guys like to build a open source robber armor with us and we see a lot of interests. Nice. And this is new market to us. We look into that. Okay there are so many robber arms but most of them are closed sourced.

The professionals are all closed sourced and they're super expensive because they were from the old industry way. They were not beautiful for the AI part. They have beautiful automations. Is it like the open arm or you know it's before that it's like kuka it's like they were like the traditional. Yeah those are like huge scales. Huge scales. Then we have the like education arms like do boards like elephants they have different kind of servo based right arm and then we okay with all the robot industry popping up in China the motors the parts are becoming cheap enough affordable enough. So we will on behalf of all the ecosystem we customize from this supply chain to build open source arm and evolve with everyone. Is that also a servo motor?

No this is brushless. Okay. Brushless. Nice. Yeah. And with all the drivers all the the P controls but we will open source everything. We already open source everything. So I think it's becoming a trust ground, a reference design for people to build to different sizes, different activations and we are more like platform to incubate more embodied or physical AI parts from this. What are some projects that are built building upon this that stood out to you? It's a very interesting evolvement. Previously with a expensive like a $20,000 robot arm, you can only do heavy industry professional ones. But with the cheaping out like $1,000 robot arm, you can think of a lot of domains cheap enough to replace human neighbors or a dangerous environment that not suitable before. We call them the how say pan industry. It's like not professional industry but it's more like for retail for agriculture. Not industry industry but it's like for everyday usages.

Yeah. More like commercial use. Commercial use in a shop or in a laboratory. Yeah. In a lab. It can it can be very pervasive. It's like personal computer. Yeah, in the beginning we don't think personal computer is not a must but in the end personal computer people think of all the usages we benefit from that I think it's a similar stage for personal robots but it's not necessarily to be a robot that work for you like a human noise that's too expensive in the beginning maybe you can start with a personal robot to help you give you extra hand on many things it can be a place from salty machine it can be a coffee maker you could be working as your chef in the restaurants for barbecue Mhm. So we see this possibilities and uh we are pushing that very much into cooking because cooking is talking about but cooking is also a very diversity of actuations and sensing.

Yeah different food different question have their different process but how can autonomous machine do that so it's a very interesting challenge and we don't have a fixed one answer that's where we like it. Yeah, with all diversities. You might be making noodles, you might be making barbecue, you might be making sushi, but people can challenge us with different models, different methodologies, but in the end, we hope that people can really use it in the practical situation. Do you think this will go together at some point? Yeah, sure. These actually they go together, but just not on this model. This is a very cheap robot hand. Mhm. And with limited activations of all freedom, but if we How many steel? 12 12ish.

12. Okay. Yeah. Yeah. And what's the cost for that like process? Uh I think the cost is $100ish. That's really cheap. Yeah. Yeah. Yeah. Wow. How's that possible? Like I mean just small servers. I see. Okay. They're very easily engineered a very clear smartly engineered. Mhm. So I believe these guys are not for venture capitalists but they are more for small business to start humbly to do something for real. Yeah. That's great. It's a different approach. Yeah. So humanoids, everyone's talking about them and so in the US a lot of these humanoid companies are vertically integrated. So platform first as well and in China I see a lot of just start small a narrow use case and then deploy it and expand. So is that how you see it and what's the winning strategy?

I think there there's no winning and losing between these two kinds because they are different creatures. Mhm. So they are different in it's like a bio system. they are in a different position for the humanoids once dominates the world have all the human egos I don't know them I don't know how they are going to win or lose they have their mythologies at least they have enough capital to pursue that possibility but what we are looking at is we believe the majority of the robots are not humanoids they are like your working cleaners they should be like nav that can corresponding to your actions even we look at the this room it's like robot we're in the belly y it's called smart home it can be more active, proactive. The AI should be responding to you. It should be agentic.

Mhm. So I think in the future like most of the devices we have today will be a robot. Yeah. Even the camera it should be moving automatically. It should be understanding your like podcast needs and to adjust itself. It's a robot as well. Yeah. But it does not needs to be a humanoid. Yeah. So our our passion and I think through the open source and maker culture is we can enable more startups who wants to build more vertical things that the big company don't care about. So you reopened the San Jose office earlier this year and you hosted this embodied AI hackathon in around GTC. So tell me more about your thoughts on the US market for seed and what are you thinking?

Yep. Uh we have the US office in Bay Area 10 years ago. We closed down because COVID because nobody is doing hardware for a few years but now AI is back. People talking about physical AI people building robots. So we want to build a a channel a portal between Bay Area and Bay Area. So I think the because the economy the talents are very complimentary the software is very strong application the venture capitalist is very strong in San Francisco Cisco area but in Shenzhen we have all the engineering solid ground supply chains. So it's very naturally complimentary to each other. But how do we shorten the distance? Not only hosting people flying from San Francisco, but we should go to the Bay Area to really sharing what's already out there to help people to build the hardware easier. So they don't need to raise a lot of big chunk of money to build something. We can go to the front end and then one button to get things made.

So that's our strategy to do that. And US is still the biggest market and we have a lot of friends relationships but we need to have a mechanism to fasten their Christians. Yeah. Save a lot of flights for for your customers. Yeah. Exactly. 20 years from now when someone writes the history of how physical AI is done and what role do you think Shenzhen would play in that history? I think Senzen should be the the biggest incubator for all the hardwares but it's not incubating for Senzhen teams only. It should be incubator for the global teams. So what are we envisioning 20 years from now is not a one gen company in China come across the world with their products. I don't think that's the most reasonable way. It should be team from all over the world. They come to Shenzhen maybe work with us maybe not to build their stuff for their own like a business for their own people. They control all the like data all the models all the supply chains. So they they should be use as a service and we we from 15 years ago we keep advocating is innovate with China.

It's not made in China in the states. It's a it's already like a hybrid very one piece of uh the industry. Yeah. It's a close partnership. It's closed partnership, but we should be more open. Mhm. And at least the city is doing all the opennesses and we are just trying to contribute our parts and for people to choose I need this or not. Well, thank you, Eric. This is great.