Another milestone in the robotics space. I hinted at this in a preview a week ago:

I've written several articles mentioning Unitree Robotics before. It's finally going public (10 Aug 2026).

The IPO price of 150.80 yuan per share puts its valuation at 61 billion RMB (roughly 9 billion USD). Given its 2025 net profit of 278–288 million RMB, the PE ratio at issue is 219x. But that number includes a one-time share-based compensation charge of around 350 million RMB, so the adjusted PE is probably closer to 103x. Still high compared to general equipment manufacturing (~38x). Or is it even fair to make that comparison?

This is the fastest IPO approval on the SSE STAR Market, in just 104 days. It's the first publicly listed, profitable, pure-play humanoid robot company. There's a lot worth digging into in the prospectus. I'll link it at the end (if you read Chinese); I'd encourage you to go through it yourself. Otherwise, I'll break down something I haven't seen anyone else cover.

Short spoiler: Most people who analyzed Unitree's customer structure pointed out that almost none of these robots are doing real work. I had the same thought (if you've read my previous articles). But after listening to a few Chinese-language podcasts and reading through the prospectus, I changed my mind.

I'll focus only on humanoid robots in this post. Quadrupeds, arms, and components are for another time.

Unitree shipped 5,500 humanoid robots in 2025, more than anyone else in the world. But if you remember what I mentioned in my last post, you can't just look at shipping numbers. You also have to look at where they went.

This is the customer breakdown Unitree disclosed in its prospectus:

These numbers line up with what I covered in the previous post. Most were purchased for research and novelty.

One number worth highlighting: humanoid-robot revenue share jumped from 27.6% in 2024 to 51.5% in 2025, overtaking quadrupeds (42.3%) for the first time.

Unitree's revenue grew 335% year over year to 1.71 billion RMB. The average selling price of its humanoid robots dropped from 593,400 RMB to 166,400 RMB, largely driven by cheaper models like the G1 and R1. Despite that, the gross margin for humanoids actually reached ~63%.

The core reason is vertical integration. Unitree designs and manufactures its own motors, reducers, and joint modules in-house. Purchased components make up just 14–18% of its costs. Most other companies outsource their actuators, which account for 40–60% of a humanoid robot's bill of materials. Unitree turned its most expensive component into its biggest cost advantage.

As I mentioned in the intro, most people analyzing this IPO point out the same thing: these robots aren't doing real work. They went to schools and labs. None of them are cleaning houses.

Fair point. I agree.

But none of them thought one step further.

Two quite different products share the same “humanoid robot” label.

The cheaper models are the G1 Basic and R1 Air. They can’t be programmed, and you buy them to use as-is.

The EDU and D series are different. They’re built for secondary development. You get the SDK (Software Development Kit), can run your own models and algorithms, and modify both the hardware and the software.

The price gap is also large between the two. The G1 starts around 85,000 RMB, while the G1 EDU is about 300,000 RMB. Roughly 3.5 times more.

In short, I think the researchers were buying a physical platform to experiment on, not because they wanted it to do something useful today.

If you’re working on locomotion, manipulation, reinforcement learning, sim-to-real, or whatever, you need an actual robot to test on. Building the hardware yourself is expensive and slow.

Unitree has turned the robot itself into a development platform.

I had mentioned this in another post of mine:

Unitree robots are in universities and labs worldwide, including Stanford, MIT, and Carnegie Mellon. Thousands of researchers are using the same relatively accessible hardware to work on different pieces of the stack.

So 74% going to research and education sounds bearish if you treat “research” as a customer category just waiting for the real market to arrive. I’m not sure that’s the right frame. It might mean Unitree has already put its hardware in the hands of the teams building the technology the industry will eventually need.

I picked up this point from a Chinese podcast that I don't think has made it into any English coverage yet.

Han Zheng, co-founder and CEO of Sudo AI, was interviewed on Silicon Valley 101 (one of my favorite podcasts). Sudo builds foundation models for robotic manipulation (brain layer). They recently demonstrated a system that reaches near-perfect pick success rates after training entirely in simulation, with zero real-robot data.

He said the combination most likely to get closest to the end-state answer is Google DeepMind + Boston Dynamics’ electric Atlas. (It’s on my list to write about)

His reasoning was that DeepMind is the most committed big tech company to physical-world AI. Boston Dynamics still has what is probably the most capable humanoid body. Put those two together, and you might get something no single vertically integrated company can match on its own.

Here’s the podcast link for whoever understands Mandarin (or you can use auto-translated subtitles):

That framing changes how Unitree should be looked at.

If the future is the best brain running on the best body, and those don’t need to come from the same company, then I do see Unitree’s value proposition shift. It’s not about being the company that does everything. It’s about being the body that everyone else builds on.

Which loops straight back to the 74%.

The researchers buying EDU-series robots aren’t waiting for Unitree to invent the perfect AI. They’re building their own. I think Unitree just has to keep shipping the most accessible, open, and affordable humanoid hardware. The intelligence will come from everywhere else.

There will be companies that do what Apple did (hardware + software, potentially Figure AI and Tesla Optimus), but I truly believe that’s a way harder route to go. On a completely different scale than a smartphone OS. Well, probably because I am more of an open-source person. I might be a little biased here, but yeah, I think someone will focus on building the best brain and plug it into the most accessible body.

I’m going to push back on Unitree’s own story.

Of the 4.2 billion RMB they planned to raise, roughly 85% is earmarked for R&D. The largest piece is the “intelligent robot model R&D” project aimed at building their own embodied intelligence models (half of the total). The message they are trying to convey is that they’re not just a hardware company, but brain and body all under one roof.

I understand why they’re telling it that way. Hardware companies get hardware multiples. A platform company with its own AI gets a much higher valuation. At 103x adjusted PE, they need something more than hardware.

But what I do see is researchers paying three times more for the right to build their own intelligence on Unitree’s body. A price point almost no one else can touch.

If Unitree keeps the hardware open and affordable, I don't think it has to win the AI race itself. It just has to stay the body that every AI team reaches for first. Thousands of teams, each trying different approaches to locomotion, manipulation, and navigation. That’s a lot more shots on goal than one internal R&D department, no matter how well-funded.

If they lock the ecosystem down to push their own models, the open-platform story might start to crack. If they stay open, the full-stack valuation story gets harder to defend. You can’t credibly tell investors you’re building a vertically integrated AI robotics company and at the same time be the neutral hardware platform everyone else develops on. At some point they have to choose.

My bet (and this is just my read, not investment advice) is that the open path has the higher ceiling (same goes for the brain; I’ll write about the AI models in the future). One company building behind closed doors is fast. A thousand labs building on the same body is faster, and it makes me more optimistic about the pace of this industry than I was before reading this prospectus (I mean, I knew this before but didn’t think of it).

One week before Unitree priced its IPO, on July 28, the US FCC (Federal Communications Commission) added foreign-produced advanced robotic devices to its Covered List.

This doesn’t immediately ban Unitree’s existing robots. The G1, H2, R1, Go2, B2, and A2 already have FCC authorization, so they can still be sold in the US. But the future models will need a specific exemption to enter the American market.

That’s not a small issue for Unitree. Without an exemption or authorization, it means no sales in the US.

The US accounts for roughly 13–19% of its revenue, depending on the period. And American universities are among the biggest overseas buyers of its humanoid robots. If my reading of that number is right, access to the US research market matters quite a bit.

A Stanford lab that already owns a G1 can keep using it. A researcher who bought one last year doesn’t suddenly lose access. The problem is the next robot. The one with better actuators, sensors, hands, and maybe compute.

If Unitree can’t keep shipping those next-generation platforms into the US, the effect probably won’t show up right away in the numbers. It would show up gradually, as researchers start buying something else (if a better option comes along, of course).

Unitree’s advantage has always been relatively simple as I see it: build capable robots, make them affordable, and get them into as many hands as possible.

The more researchers use the hardware, the more people build software around it. The more software gets built, the more useful the platform becomes. Then more researchers buy it. I think it’s closer to an ecosystem rather than a hardware flywheel.

I don’t know whether Unitree is worth $9 billion, or whether its full-stack AI strategy will work. But after reading the prospectus, I do think the usual interpretation of Unitree is missing something.

I believe Unitree has already found a market for an affordable, relatively open humanoid platform that thousands of researchers can build on.

And I think that distinction matters. I believe robotics will eventually look more like an ecosystem than a single vertically integrated company. The hardest problems — manipulation, navigation, reasoning in unstructured environments — are too broad for one company to solve alone.

That said, I'll be watching how the FCC situation plays out.

Here’s the prospectus: https://static.sse.com.cn/stock/disclosure/announcement/c/202603/002178_20260320_QY8F.pdf

I read a 363-page Chinese prospectus so you don't have to. If that's the kind of coverage you find useful, subscribe so that the next piece lands smoothly in your inbox.