The $4,900 Humanoid — or, why we need to wake up and win the AI robotics race

You can soon buy a walking, dancing, handstand-ing humanoid robot on AliExpress for $4,900. That's almost as little as a spec'd-out pair of Apple Vision Pros. It's probably less than your car.

The robot, called the R1, is produced by Unitree, a Hangzhou-based robotics company that until recently was known primarily for quadrupeds — and selling said quadrupeds to various militaries. To put the price into context, Tesla has suggested that Optimus may eventually reach roughly $20,000¹. Figure AI and Apptronik are generally estimated to sit higher still. Boston Dynamics sells their flagship Atlas humanoids for around $150,000.

So how did Unitree manage to decimate their production costs? The answer isn't cheap Chinese labour, anymore.

80% of the Frontier

For years, the dominant narrative around humanoids focused on frontier capability. Viral demonstrations from Boston Dynamics or research laboratories framed the sector as an advanced engineering problem still some distance from commercial viability. The idea was: once we develop the tech to build well-functioning, generalisable robots, we can then transition to commoditising them.

Unitree clearly doesn't believe in that analogy and is giving humanoids the Fordian treatment — relentless supply chain optimisation to produce a good-enough army of Synths.

The obvious assumption is that lower prices reflect correspondingly lower quality. To some extent, they do. The R1 is not near the capability frontier. Its hands are comparatively primitive and cannot perform delicate manipulation tasks reliably. The motors produce substantially lower torque than systems such as Atlas or Optimus. Battery life remains constrained.

Boston Dynamics demonstrations remain technically extraordinary, and certainly superior to what Unitree has shown. But a few of BD's most famous skills are produced at the edge of reliability: Marc Raibert, BD's CEO, himself acknowledged this fragility, noting that some of their viral videos took many takes in controlled environments. Meanwhile, Unitree regularly stages synchronised live demonstrations involving dozens of robots operating simultaneously in public settings. Factory operators care less about whether a humanoid can perform a perfect acrobatic manoeuvre once than whether it can complete the same movement ten thousand times without service interruption.

So Unitree robots' capabilities are not even so much worse, and the delta is definitely not to scale with the price difference.

Learnings from DJI

Unitree was founded in 2016 by Shanghai University graduate Wang Xingxing, briefly after working at DJI. The career path itself says something about modern Chinese hardware: DJI became the global drone leader by compressing the distance between prototype and mass production. The company took technologies that had largely emerged from Western research labs and hobbyist communities, integrated them aggressively into consumer products, then drove prices down through manufacturing scale and vertical integration.

The launch of the Phantom in 2013 for just $629 was truly a revolution. Within a decade, drones went from specialist equipment costing thousands of dollars to commodity hardware. DJI has ended up controlling more than 90% of the global consumer drone market.

Unitree increasingly looks like the same playbook applied to humanoids. The company began by building quadrupeds, releasing increasingly inexpensive alternatives to systems such as Boston Dynamics' Spot. Over time, Unitree developed a reputation for compressing the cost of advanced robotics hardware far faster than many Western competitors expected. Its move into humanoids followed the same trajectory: H1 launched around $90,000, the G1 is roughly $16,000, and the R1 now sits near $4,900 — 21× less than the H1, 34× less than Atlas.

Supply chain engineering as a moat

A significant portion of Unitree's advantage comes from brilliant supply chain engineering. The company relies heavily on rotary joints rather than more complex and expensive linear systems. Rotary systems sacrifice some performance, but they are substantially cheaper, easier to manufacture, and easier to maintain at scale. Unitree also designs many of its own permanent magnet synchronous motors in-house rather than purchasing expensive actuator assemblies externally.

The broader design philosophy strongly resembles the actuator architecture developed for MIT's Mini Cheetah project. That research demonstrated that relatively inexpensive, 3D-printed, hand-wound, quasi-direct-drive electric actuators could achieve remarkable dynamic locomotion when paired with good control software. One of these actuators might cost as little as $25–$150 to produce, making an R1's 12 motors per bot as cheap as ~$300, compared to thousands for linear actuators.

Unitree's IPO filings confirm the company "self-developed all core components such as motors, reducers (gearboxes), and lidars". This vertical integration means Unitree isn't paying a supplier's markup on the single most expensive pieces.

Density matters as much as cost. A robotics company in Hangzhou sits inside the same industrial network that already manufactures drones, batteries, smartphones, electric vehicles, and rare-earth magnet assemblies at enormous scale. A recent analysis notes that Chinese factories can assemble a robotic arm for "less than half" the material cost of a US equivalent. A startup in Boston or Munich may spend weeks sourcing prototype components from multiple suppliers across continents. A comparable firm in Shenzhen or Hangzhou can often iterate hardware locally within days.

Made in China 2025

Industrial policy has reinforced these advantages. Beijing identified robotics, advanced manufacturing, batteries, and automation as strategic sectors years ago. In late 2024 alone, $20+ billion was directed into robotics firms via grants, state VC funds, and local technology programs. State-owned buyers even bulk-purchased domestically built robots — Unitree's own filings mention a $17 million order in 2025 — to ensure factories are scaling up.

China is also ageing rapidly, labour costs continue to rise, and the old economic model built around abundant low-cost manufacturing labour is weakening. In that context, the country's investment into automation increasingly resembles an attempt to build its future workforce rather than birth it.

What makes Unitree particularly interesting is that humanoids are not yet primarily being sold into industrial deployment. In 2025, over 5,500 of its humanoid robots were shipped globally — about 37× the volume of the next-largest competitor. But ~74% of that humanoid revenue came from research, education and startup customers, while actual industrial deployment remains relatively limited.

Unitree is selling "picks and shovels," not worker-hours. It doesn't have to solve the difficult engineering problem of shipping deployable bots — it just has to build the platform for the Western engineers who will.

Thousands of developers and companies built workflows around DJI systems because the hardware was available, affordable, and sufficiently reliable. Unitree wants to do the same, but by open-sourcing the system they're building a moat deeper than DJI's ever was.

Why this matters beyond a price war

Having the majority of drones flying in your airspace produced in China doesn't bode well for a lot of people — the FCC included. It recently added DJI to its Covered List, effectively banning the import of new DJI drone models into the US. But what does that mean for the thousands of US companies that built their products on the DJI platform? According to the WSJ, around 40% of pilots forecast an "extremely negative" or business-ending impact.

The same risk profile becomes more acute with humanoids. The concern is not just that China might lead in unit cost, but that it could become the default substrate for global robotics experimentation. If thousands of research labs, startups, and industrial users are iterating on Chinese hardware, then software and system-level improvements also accumulate there.

Robotics is also structurally different from earlier waves of globalisation. It's not only about manufacturing anymore — it's manufacturing plus sensing, autonomy, software iteration, and tight hardware-software coupling. When you change up a robot's hardware, training is expensive and non-trivial, so having manufacturing coupled with development for the control system platform is absolutely crucial. Once a country controls the physical production base at scale, it becomes much easier to iterate the entire system. The feedback loop between factory floor, field deployment, and product redesign becomes geographically concentrated.

What the West should do

My two cents:

1. Fight fire with fire. The US and Europe need to aggressively invest capital in robot supply chains. While we're structurally constrained by certain raw materials, we still have the technological upper hand. Even though less efficient, we're still quite adept at manufacturing, especially when it comes to large, precise, and mobile things — rockets, planes, combustion cars, high-precision machinery. Precision German engineering doesn't have to be a relic of the past — but it'll have to refer to mecha-suits instead of cars.
2. Ship short of perfection. The European disease is to spend three years building the perfect prototype while Unitree ships v0.8 to a thousand researchers and lets them debug it for free. Ship before you're ready, price for volume, get your hardware into every damn grad office in the world, and let the global robotics community write your software for you. We need to build a few Unitrees of our own.

Both things are easier said than done. But what's clear is that we need more urgency. Right now, we're too busy debating whether we should be scared of accidentally building Skynet. The real question is — do we want Skynet to be Americo-European or Sino-Dictatorial?

Got a different opinion? Leave us a comment on Substack. We're always excited to duel strangers on the internet.