US vs China: The Race for Humanoid Robot Supremacy

Source:https://chatgpt.com/share/67c9f2e3-3f40-800b-adf1-7ab42715832c

Got it! I'll research the most advanced and commercially scalable general-purpose humanoid robot companies in the US and China, focusing on their technical sophistication and commercial potential over the next three years (2025-2028). I'll compare factors such as AI capabilities, dexterity, mobility, funding, production scale, and strategic partnerships. I'll let you know when I have the findings ready.

United States: Leading Humanoid Robot Companies

Tesla – Optimus (Tesla Bot)

Tesla’s Optimus is a general-purpose humanoid robot project unveiled in 2021. Backed by Tesla’s AI prowess and massive manufacturing resources, Optimus is intended to perform a wide range of manual tasks in factories and eventually in daily lifeen.wikipedia.orgwww.thestar.com.my. Tesla has rapidly iterated on Optimus prototypes – by late 2023 the robot could walk, sort objects by color, perform simple yoga poses, and delicately grasp items like eggsen.wikipedia.orgen.wikipedia.org. Key points:

  • AI & Dexterity: Optimus leverages Tesla’s computer vision and autopilot AI technology for perception. It uses human-sized limbs and hands (with improved dexterity in the Generation 2 prototype) to manipulate objects. Recent demos show it identifying and picking up objects autonomously (e.g. sorting colored blocks) and maintaining balance in dynamic posesen.wikipedia.orgen.wikipedia.org. Musk claims the bot will eventually handle tasks “humans don’t want to do,” from household chores to factory worken.wikipedia.orgwww.indiatoday.in.
  • Funding & Backing: Development is funded in-house by Tesla’s multibillion-dollar resources. Elon Musk has stated Optimus could become more significant than Tesla’s vehicle businessen.wikipedia.org, underlining Tesla’s commitment. Partnerships include component suppliers – for example, China’s Sanhua is expected to supply actuators and thermal systems for Optimuswww.allianzgi.com– but Tesla owns the overall R&D.
  • Production & Scalability: Tesla’s scale-up plans are extremely ambitious. Musk aims for limited production in 2025 (on the order of thousands of units for internal use) and then rapid ramp-up in 2026en.wikipedia.org. In earnings calls he projected eventual output exceeding 1 million units per year, once Optimus is maturewww.thestar.com.my. Tesla believes it can leverage its high-volume manufacturing expertise to achieve a cost per robot under **20,000**[thestar.com.my](https://www.thestar.com.my/tech/tech-news/2025/02/05/battle-of-the-bots-china-us-scrap-for-top-of-the-humanoid-heap#:~:text=ramping%20up%20production%20the%20following,year), potentially pricing Optimus around n20–30kwww.thestar.com.myen.wikipedia.org.
  • Use Cases & Roadmap: Initial use will be in Tesla’s own factories – moving materials, assembling parts, or other repetitive tasks. By gathering data in real-world scenarios, Tesla hopes to improve Optimus’s intelligence over timewww.thestar.com.my. Musk has said over 1,000 units will work in Tesla facilities by 2025, and sales to other companies could begin in 2026en.wikipedia.org. The long-term vision (2028 and beyond) is a home assistant capable of cooking, lawn-mowing, eldercare, etc., but near-term focus is on mastering industrial tasks and proving scalability. Tesla’s huge advantage is its existing Gigafactories and supply chain – if the software and robotics engineering fall into place, Tesla can build and deploy robots at a volume no competitor can yet matchwww.thestar.com.mywww.thestar.com.my. The challenge will be ensuring those robots are truly autonomous and dexterous enough for broad use, not just simple pre-programmed routines.

Agility Robotics – Digit

Agility Robotics (Oregon-based) produces Digit, a human-sized bipedal robot aimed at logistics and industrial applications. Digit has two legs and two arms (end in simple grippers) for walking, lifting and carrying objects. While less humanoid in appearance (no head or fingers), it is one of the first commercial general-purpose biped robots available. Key points:

  • AI & Capabilities: Digit is designed to navigate human environments and perform basic manipulation. It can walk forward/backward, up ramps and stairs, balance on uneven terrain, turn in place, crouch, and carry loads like storage toteswww.geekwire.comwww.geekwire.com. An upgraded Digit in 2023 added a “head” with sensors and improved hands for better graspingwww.geekwire.com. Its autonomy is suited for tasks like moving boxes in warehouses or unloading trucks – Agility has demonstrated Digit picking up packages and stacking bins. The robot uses onboard vision and proprioceptive sensors for perception and utilizes dynamic motion planning (Agility’s roots trace to academic research on legged locomotion). While not as acrobatic as some research robots, Digit’s mobility and stability are state-of-the-art for a commercially available humanoid.
  • Funding: Agility Robotics has strong financial backing. It raised $150 million in 2022 led by DCVC and Playground Global, with Amazon’s Industrial Innovation Fund joiningwww.geekwire.com. Amazon is not only an investor but also an early customer testing Digit in its warehouseswww.geekwire.comwww.geekwire.com. Agility’s investors also include robotics-focused VCs and possibly Ford (which partnered with Agility in 2020 to test package delivery)www.geekwire.com. This funding enabled Agility to build out manufacturing capacity and accelerate R&D of next-gen capabilities.
  • Production & Capacity: In 2023 Agility opened “RoboFab,” the world’s first dedicated humanoid robot factory, in Salem, Oregonwww.geekwire.comwww.geekwire.com. The 70,000 sq. ft facility will employ ~500 people and also use Digit robots on the assembly line to help build more robotswww.geekwire.com. Initial output is modest – hundreds of Digits in the first year – but the factory is designed to scale up to 10,000 robots per year at full capacityagilityrobotics.comwww.geekwire.com. This mass production aim (unprecedented for humanoids) positions Agility as a frontrunner in scaling. The first factory-built Digits are being delivered to pilot customers in 2024, with general availability expected in 2025singularityhub.com.
  • Partners & Applications: Agility is laser-focused on warehouse and manufacturing applications. Amazon’s trial involves using Digit for tote handling and material movement alongside human workerswww.geekwire.comwww.geekwire.com. Agility also inked a deal with Ford to explore last-meter delivery (having Digit carry packages from a vehicle to the doorstep)www.geekwire.com. These partnerships validate Digit’s use cases. Going forward, Agility is marketing Digit to logistics firms, retailers, and factories that need flexible automation – essentially, replacing or assisting human labor in physically repetitive tasks like stacking, order fulfillment, and intra-facility deliveries.
  • Roadmap: Agility’s near-term milestone is to successfully deploy dozens of Digits in real customer operations by 2025, proving ROI in tasks like package handling. By 2026–2028, Agility aims to refine Digit’s dexterity (possibly giving it more fingered hands) and cognitive abilities so it can take on a wider range of jobs. The company’s CEO calls the new factory opening “the beginning of mass production of commercial humanoid robots”www.geekwire.comwww.geekwire.com. With production infrastructure in place and a head start in real-world testing, Agility is well positioned to scale. The trade-off is that Digit is intentionally less complex than some rivals (for example, it doesn’t attempt delicate human-like manipulation yet), which makes it easier to produce now but may require upgrades to match the “general-purpose” intelligence and dexterity of others in the long run.

Apptronik – Apollo

Apptronik is an Austin, Texas-based robotics company (spun out of University of Texas research and a former NASA partnership) that unveiled Apollo, its general-purpose humanoid, in 2023. Apollo is roughly human-sized and designed to be an adaptive worker in industrial and service environments. Apptronik combines advanced robotics engineering with strategic partnerships to accelerate scaling. Key points:

  • AI & Capabilities: Apollo is built for human-like mobility and basic manipulation, intended to work in warehouses and manufacturing plants moving materialswww.reuters.comwww.reuters.com. It has 4° of freedom arms and gripper hands to lift boxes or use tools, and it balances on two legs with a range of motion sufficient for walking, turning, and crouching. Apollo’s control system is influenced by Apptronik’s experience building force-controlled robotic limbs (the team previously worked on NASA’s Valkyrie humanoid and exoskeletons). While detailed specs are not widely published, Apollo’s designers emphasize safe operation around people (it’s fully electric with modest force output per joint to avoid dangerous impacts). The company is also collaborating with Google DeepMind to integrate cutting-edge AI for perception and decision-makingwww.reuters.com. As a result, Apollo’s autonomy and “brain” could rapidly improve by leveraging the latest AI models for vision and language.
  • Funding: Apptronik recently secured a massive **350 million Series A** funding round (Feb 2025) led by B Capital and Capital Factory, with participation from **Alphabet/Google** and other investors[reuters.com](https://www.reuters.com/technology/apptronik-raises-350-million-scale-production-humanoid-robots-2025-02-13/#:~:text=Feb%2013%20%28Reuters%29%20,powered%20humanoid%20robots). This war chest is one of the largest in humanoid robotics and will finance the scaling of Apollo. (Apptronik had earlier raised ~n15M seed and partnered with NASA in 2022 to develop commercial humanoid techwww.iotworldtoday.com.) The new funding is explicitly to scale production of Apollo and expand its capabilities into areas like eldercare and healthcarewww.reuters.com. Such strong financial backing – including big names like Alphabet – underscores confidence in Apptronik’s approach.
  • Production & Manufacturing: Rather than build a factory from scratch, Apptronik has partnered with Jabil, a global electronics manufacturing services company, to produce Apollo at scaleapptronik.com. Jabil will help refine Apollo’s design for mass manufacturing and set up production lines, leveraging its expertise in supply chain and fabrication. This partnership gives Apptronik a clear path to ramp up output quickly without huge upfront infrastructure costs. The company envisions a virtuous cycle where Apollo robots will eventually be building more Apollo robots on the manufacturing line (similar to how Agility uses Digit in its factory)www.robotics247.com. In terms of timeline, Apptronik plans to deploy pilot units of Apollo in partner facilities (like Mercedes-Benz and logistics warehouses) in 2024–2025, and then move to volume production after validating performance.
  • Partnerships & Use Cases: Apptronik has secured commercial agreements with Mercedes-Benz (to explore Apollo in automotive manufacturing) and GXO Logistics (a major warehousing and supply chain firm)www.reuters.com. These partners will test Apollo in real-world scenarios such as automotive assembly lines or distribution centers. Early demonstrations showed Apollo handling warehouse tasks – e.g. moving crates – and the robot is being targeted at any job involving repetitive material handling. Given Apptronik’s NASA roots, Apollo may also find use in hazardous environments or even future space applications (longer-term). The collaboration with Google DeepMind indicates Apollo could integrate advanced vision-language models, potentially enabling more natural interaction (verbal instructions, visual scene understanding) in service roles down the linewww.reuters.com.
  • Roadmap: In CEO Jeff Cardenas’s view, robotics is at an “inflection point” akin to where large language models were in 2023 – he predicts 2025 will see a surge of real-world humanoid deploymentswww.reuters.com. Apptronik’s roadmap for Apollo includes expanded deployment in 2025 and then scaling up production through the Jabil partnership thereafterwww.reuters.com. By 2026–2027, they aim for Apollo to be commercially available to multiple industries, and they are exploring applications beyond factories, e.g. in hospitals or elder care (where a humanoid could lift patients or deliver supplies)www.reuters.com. With substantial funding and strong partners, Apptronik is well positioned to scale while iterating on Apollo’s intelligence. The challenge ahead will be proving Apollo’s reliability and versatility in its pilot programs so that full production can kick into high gear on schedule.

Figure AI – Figure 01

Figure AI is a Silicon Valley startup (founded 2022) that has quickly become one of the most well-funded players in humanoid robotics. Its unnamed humanoid (often referred to as Figure 01) is in development as a general-purpose bipedal robot targeting both industrial and day-to-day tasks. Figure emphasizes advanced AI brains for its robots, aiming to combine human-like intelligence with human-like mobility. Key points:

  • AI & Capabilities: Figure’s robot is planned to be 5’6” tall, 60+ kg, with 40+ degrees of freedom – comparable to an adult human in size and articulationwww.reuters.com. The goal is a high level of dexterity and autonomous decision-making. Figure is partnering with OpenAI to integrate state-of-the-art generative AI into the robot’s “brain”www.reuters.comwww.reuters.com. This means the humanoid will use large language models (based on GPT-4 and beyond) fine-tuned on robotics data, enabling it to converse naturally, understand complex instructions, and reason about taskswww.reuters.com. In essence, Figure is giving embodied AI a physical form – the robot should be able to see (computer vision), talk, and adapt to new tasks without explicit reprogramming. Early prototypes and simulations (revealed in mid-2023) showed the robot walking and balancing; with the infusion of AI, the expectation is that Figure’s humanoid will eventually handle diverse jobs from warehouse lifting to retail customer service or household chores.
  • Funding: Figure AI has attracted an astonishing **675 million** in funding as of early 2024[reuters.com](https://www.reuters.com/technology/robotics-startup-figure-raises-675-mln-microsoft-nvidia-other-big-techs-2024-02-29/#:~:text=Feb%2029%20%28Reuters%29%20,6%20billion). The round included tech giants **Nvidia and Microsoft**, the **OpenAI Startup Fund**, **Intel Capital**, and even Amazon’s fund and Tesla-founder Elon Musk’s rivals like Jeff Bezos (through his venture arm)[reuters.com](https://www.reuters.com/technology/robotics-startup-figure-raises-675-mln-microsoft-nvidia-other-big-techs-2024-02-29/#:~:text=Feb%2029%20%28Reuters%29%20,6%20billion)[reuters.com](https://www.reuters.com/technology/robotics-startup-figure-raises-675-mln-microsoft-nvidia-other-big-techs-2024-02-29/#:~:text=investing%20million%20of%20dollars%20to,cash%20in%20on%20the%20trend). This gave Figure a valuation of n2.6 billion pre-revenuewww.reuters.com– reflecting huge investor belief in the team and vision. Notably, Nvidia’s involvement suggests Figure will leverage Nvidia’s hardware (such as powerful AI chips) for the robot’s computing, and Microsoft’s involvement comes with a switch to Azure for Figure’s AI training needswww.reuters.comwww.reuters.com. The CEO, Brett Adcock, has stated the funds will go toward developing large AI models for the robot, ramping up manufacturing, and hiring talentwww.reuters.comwww.reuters.com. Earlier in 2023, Figure had already raised a $70M seed round and was operating in stealth, recruiting engineers from companies like Boston Dynamics, Tesla, and IHMC Robotics.
  • Production & Scaling: With its enormous funding, Figure is now gearing up for hardware manufacturing. The company in 2024 partnered with BMW to test and deploy humanoid robots in a BMW production facilitywww.reuters.com. This partnership likely serves as a pilot manufacturing use-case and could eventually lead to BMW or other automakers buying robots if tests go well. Figure’s strategy seems to be securing flagship customers (like BMW) early, while using investor capital to build out a production line. They have not publicly announced a dedicated factory yet, but given the resources, Figure could set up a manufacturing operation or contract an electronics/auto manufacturer by 2025. The presence of investors like Foxconn (rumored via venture funds) or use of an existing automotive line via BMW could expedite scaling.
  • Partnerships & Vision: Aside from OpenAI (for AI software) and BMW (for pilot deployment), Figure is working closely with NVIDIA – in fact, it was highlighted by NVIDIA as a key partner using its technologyblogs.nvidia.com. This likely gives Figure early access to the latest humanoid robot simulation tools, GPU-accelerated AI training, and perhaps Omniverse environments to train the robot in virtual scenarios. By training its robot’s AI in simulation, Figure hopes to imbue a kind of general intelligence that transfers to real-world versatility. The OpenAI collaboration indicates the robot could utilize voice and vision AI to interact in a human-like way. Potential commercial applications mentioned by Figure include warehouse labor, manufacturing, retail assistance, and eventually home services – essentially a multi-purpose robot worker.
  • Roadmap: In late 2023, Figure had a working prototype and was refining bipedal locomotion and grasping in the lab. By 2024, the focus shifted to the AI brain (with OpenAI’s help) and preparing for real-world trials (hence the BMW partnership). The company’s massive Feb 2024 funding will enable it to attempt simultaneous progress on AI and hardware. Figure’s stated goal is to have its humanoids begin working alongside humans within a couple of years. The partnership with BMW suggests a target to deploy robots on a factory floor by 2025. From 2025 to 2028, Figure aims to scale up manufacturing and field a variety of pilot programs in logistics, manufacturing, and possibly service sectors. Thanks to the funding, Figure can afford to iterate and even fail fast if needed; however, it faces stiff competition from more established players. If Figure successfully merges cutting-edge AI with reliable humanoid hardware, it could leapfrog others in the “intelligence” dimension, giving it an edge as robots become more than just brute labor. In summary, Figure is very well positioned financially and technologically, but it is still in development – the next 2–3 years will prove if their highly ambitious approach can deliver a real, working general-purpose humanoid at scalewww.reuters.comwww.reuters.com.

Boston Dynamics – Atlas (R&D Pathfinder)

Boston Dynamics, based in Massachusetts, is renowned for building the world’s most advanced humanoid research robot, Atlas. While Boston Dynamics does not yet sell a general-purpose humanoid, Atlas serves as a cutting-edge platform to push the boundaries of bipedal locomotion and manipulation. BD’s work – backed by Hyundai Motor Group – positions it as a technical leader, and potentially a future entrant to the commercial humanoid market. Key points:

  • AI & Capabilities: Atlas is often dubbed “the world’s most dynamic humanoid.” It can run at human-like speeds, jump and even do backflips, and balance on uneven terrain with agility unmatched by any other humanoid robotbostondynamics.combostondynamics.com. In recent demos, Atlas showed dynamic manipulation: using both hands to pick up heavy objects, climb scaffolding, and even throw objects accuratelybostondynamics.comnewatlas.com. Its control system tightly integrates perception with motion – using depth sensors, Atlas can adapt its steps in real-time to obstaclesbostondynamics.com. Boston Dynamics has essentially solved many core challenges of locomotion and is now working on giving Atlas more useful task skills. However, Atlas is largely teleoperated or pre-programmed for demos (it’s a testbed, not an autonomous AI robot yet)en.wikipedia.org. The company’s focus is on perfecting the hardware and low-level control: custom high-torque actuators, lightweight 3D-printed structures (Atlas has a titanium/aluminum frame)bostondynamics.combostondynamics.com, and whole-body dynamics algorithms. This makes Atlas extremely advanced in dexterity and athletic ability – an excellent foundation if paired in the future with higher-level AI for autonomy.
  • Funding & Team: Boston Dynamics has an unusual history – it was born out of MIT, then acquired by Google, then SoftBank, and now Hyundai Motor Group owns 80% of it (acquisition valued at $1.1B in 2021)www.therobotreport.com. Hyundai’s backing gives BD strong financial support and manufacturing expertise, as well as strategic interest in using robots for smart factories and mobility. While BD operates more like an R&D company, Hyundai’s influence has shifted it toward practical products (e.g. Boston Dynamics commercialized the Spot quadruped and Stretch warehouse robot under Hyundai). The Atlas project itself is well-funded internally for research. Boston Dynamics also benefits from an exceptionally talented robotics team and over a decade of humanoid research, giving it intellectual capital that few can rival.
  • Commercial Plans: At present, Boston Dynamics has no announced plan to mass-produce Atlas. Instead, its first commercial robots are Spot (the quadrupedal robot dog) and Stretch (a warehouse robotics system for moving boxes). These are being sold in limited quantities – for instance, a few hundred Spot robots have been sold to date for inspection and research uses. Atlas remains a research platform, but its technology is directly influencing Boston Dynamics’ future products. It’s expected that Boston Dynamics will eventually apply Atlas’s innovations to commercial humanoids or advanced logistics robots. Analysts speculate that by later in this decade (2026–2028), BD could introduce a humanoid specifically for factory work or disaster response, leveraging Atlas’s hardware innovations with a more cost-engineered design.
  • Partnerships & Applications: Because of Hyundai’s involvement, one likely application for a future BD humanoid would be in automotive factories (Hyundai has even shown concept videos of factory service robots and wearables). Boston Dynamics also collaborates with DARpa and academic labs, meaning Atlas’s descendants could be suited for defense or emergency response tasks (e.g. navigating rubble, operating tools in human-designed environments). In terms of AI, BD is somewhat more conservative – it focuses on robust motion first, then layers task-specific programming. This contrasts with AI-heavy approaches like Tesla or Figure. However, as AI models become more plug-and-play, BD could integrate them when needed. Notably, competitors like Figure have cited how “getting the hardware working is the easy part” and the true challenge is higher-level autonomynewatlas.com– Boston Dynamics has essentially mastered the hardest hardware part already, which could make them formidable if they solve the “brain” aspect.
  • Outlook: In the next three years, Boston Dynamics will likely continue refining Atlas (improving its hands, efficiency, and autonomy) and possibly announce a path to commercialization if the market demand for humanoids grows. Even if they don’t sell Atlas directly, BD’s technology may quietly underpin other efforts – for instance, they could license or partner (Hyundai could integrate Atlas tech into its factories or products like service robots). Boston Dynamics brings unmatched technical sophistication – any eventual commercial humanoid they produce would likely be extremely capable in dexterity and mobility. The question is scaling: can they make Atlas-like robots at a reasonable cost and volume? With Hyundai’s manufacturing might, it’s plausible in a few years. For now, Boston Dynamics remains the industry’s “benchmark” for what humanoids can physically do, rather than a commercial leader – but this could change quickly if they pivot from R&D to mass product, making them a dark horse in the scale race. (Aside from these, other U.S.-based efforts include Sanctuary AI (Canada/U.S.) and startups like Unlimited Robotics, but the four above are the most prominent with near-term scaling prospects. The U.S. also benefits from top academic labs feeding talent into these companies, and big tech firms (Google, Meta, etc.) investing in robotics research that could spill over into humanoid development.)

China: Leading Humanoid Robot Companies

UBTECH Robotics – Walker Series

Shenzhen-based UBTECH is one of China’s most prominent humanoid robot companies, known for its Walker bipedal robots. UBTECH has years of experience (founded 2012) in humanoids – from small consumer robots to full-size bipeds – and is now aggressively pushing into industrial applications. Its latest Walker S series robots are human-sized, AI-powered humanoids built for factory and service tasks. Key points:

  • AI & Capabilities: The UBTECH Walker S is about 1.7 m tall, fully electric, with 26+ degrees of freedom including dual armswww.micweekly.comwww.tweaktown.com. It features advanced joint actuators (force-controlled joints with “rigid-flexible coupling” for smooth motion) and a multi-modal AI “brain.” UBTECH integrates a multimodal large AI model for general task intelligence, semantic SLAM for navigation, and learning-based motion controlwww.micweekly.com. In practical terms, Walker can walk at ~4–5 km/h, balance dynamically (even perform brief jogs or jumping in some versions), and use its arms to grasp and manipulate objects. UBTECH has demonstrated Walker robots carrying out factory tasks: e.g. working alongside autonomous forklifts on automotive assembly lineswww.micweekly.com. Walker S can coordinate with other robots and factory management systems – in one case, a Walker moved materials and pressed machine buttons in sync with automated guided vehicles (AGVs)www.micweekly.com. UBTECH has also given Walker conversational and vision capabilities (facial and gesture recognition, and the ability to perceive human emotions) from its earlier consumer robotics workwww.micweekly.com. This blend of physical dexterity and interactive AI makes Walker a versatile platform for industry and potentially public-facing services.
  • Funding & Backing: UBTECH is well-funded, having raised hundreds of millions in venture capital over the past decade (investors include Tencent and incubator funds; it was a “unicorn” valued around $5B in 2018). The company also benefits from government support – it’s a key player in China’s national humanoid robot initiativesen.tmtpost.comen.tmtpost.com. UBTECH is part of the Embodied Intelligence Robot Innovation Center in Beijing, sharing research and standards with other firms and institutionsen.tmtpost.comen.tmtpost.com. This consortium, supported by China’s Ministry of Industry and IT, provides resources and an open-source platform (called “Tiangong”) for core humanoid technologies, which UBTECH helped developen.tmtpost.com. Such backing accelerates UBTECH’s R&D and helps align it with national strategy (e.g. access to academic research and government pilot programs).
  • Production & Scalability: UBTECH is gearing up for mass production by 2025. The company plans to deliver 500–1000 units of Walker S to clients by late 2025www.tweaktown.comwww.tweaktown.com. It has already deployed an initial batch (“the first humanoid team”) to a car factory in China in 2024, as a trialwww.scmp.comwww.yahoo.com. To scale, UBTECH is partnering with major manufacturers: it will be supplying Walker robots to Foxconn (the Apple iPhone assembler) for electronics factory automation, and to SF Express (a logistics giant) for warehouse and delivery taskswww.tweaktown.comwww.tweaktown.com. These are huge potential customers that could each absorb many robots if trials succeed. UBTECH’s manufacturing base in Shenzhen and ties to electronics supply chains allow it to ramp up production relatively quickly, leveraging China’s component cost advantages. Cost-wise, UBTECH hasn’t published unit prices, but Chinese experts note that to succeed, domestic humanoids like Walker must target under ¥200,000 (~$28k) per unit in coming yearswww.thestar.com.my. UBTECH’s scale and integrations suggest it aims to hit those cost targets through volume and local sourcing.
  • Applications & Partnerships: UBTECH’s strategy focuses on industrial and enterprise applications first. Automotive assembly is a key use-case: Walker robots can perform repetitive assembly or inspection tasks, augmenting the workforce in a car plant. The partnership with FAW-Volkswagen (a major auto joint venture) has UBTECH’s robots working on unmanned production lines to improve car manufacturing efficiencyqviro.com. In logistics, UBTECH envisions its humanoids loading/unloading trucks, sorting packages (in coordination with conveyor systems), or even last-mile delivery in warehouses – SF Express’s interest hints at these scenarios. Another domain is smart services in public venues: UBTECH has shown its smaller Walker X model greeting people and handling reception duties at events (and even dancing in coordinated performances). With AI-driven interaction, Walker could serve in malls, airports, or hospitals for customer service and security patrols. The company’s breadth of technology (from past educational robots to cutting-edge humanoids) means it can customize the “personality” and interface of Walker for different clients.
  • Roadmap: By end of 2025, UBTECH intends to have mass production in full swingwww.tweaktown.com. It is launching iterative improvements – e.g. Walker S2 in mid-2025, and a Walker S3 by late 2025www.tweaktown.com– each version more capable, showing a fast evolution cycle. This reflects a “ship and improve” approach: get robots out to real users, then refine hardware/software in subsequent generations. Looking at 2026–2028, UBTECH could have a fleet of several thousand humanoids deployed across manufacturing, logistics, and perhaps public service sectors in China. This would make it one of the largest humanoid operators globally. The company will likely also start expanding abroad, potentially offering humanoid workforce solutions in other Asia-Pacific countries or partnering with Western firms via its investors. In terms of maintaining high intelligence and dexterity, UBTECH’s approach is to incorporate the latest AI models (they mention large models and end-to-end skill learningwww.micweekly.com). Combined with their proven hardware, this positions UBTECH as a top contender balancing technical sophistication with immediate scalability – indeed, it may achieve the highest near-term volume of humanoid deployment among all playerswww.tweaktown.comwww.thestar.com.my.

Unitree Robotics – H1/G1

Unitree Robotics, based in Hangzhou, is known for making affordable legged robots (they gained fame for robotic “dog” quadrupeds). Now Unitree has entered the humanoid arena with its H1 and G1 humanoid robots. Unitree’s philosophy is to push costs down and production up, making humanoids as accessible as their smaller robots. They aim to leverage China’s supply chain to flood the market with capable yet inexpensive humanoids in the next few years. Key points:

  • AI & Capabilities: Unitree’s H1 is billed as a “universal humanoid” – a bipedal robot capable of human-like movements including dancing, walking, and basic interactionswww.thestar.com.my. A group of 16 H1 robots famously performed a synchronized folk dance at China’s Spring Festival Gala in early 2025, showcasing their balance and coordinated motion controlwww.thestar.com.mywww.thestar.com.my. Technically, H1 stands around adult human height and has a straightforward, functional design (with a screen for a face). The commercial model G1 is a slightly smaller humanoid (1.27 m tall, 35 kg) optimized for mass productionwww.micweekly.com. Depending on configuration, G1 has 23–43 degrees of freedom, allowing it flexibility in movement. It can walk at up to 2 m/s (human jogging speed) and perform dynamic actions like getting up from a fall, squatting, and even doing routines like “stick dancing”www.micweekly.comwww.micweekly.com. Unitree equips its robots with advanced balance algorithms (honed from quadruped development) and decent perception – likely using cameras and possibly lidar for navigation (though detailed AI specs are scant, they mention using AI for interactive rolesqviro.com). The emphasis is on reliable locomotion and basic manipulation; current Unitree humanoids have relatively simple gripper hands. Their demonstrated use-cases include inspection patrols, greeting people, carrying items, and entertainment performances.
  • Funding & Business Model: Unitree is a private company but has scaled largely through sales revenue of its smaller robots (like the Unitree Go1 robot dog, which sells for a few thousand dollars and has been shipped worldwide). Founder Wang Xingxing has a vision of making humanoids ubiquitous, and the company likely benefits from local government support and China’s investor enthusiasm for robotics. While not as flush with VC cash as some competitors, Unitree’s existing robotics business provides a revenue stream and manufacturing experience. Importantly, Unitree has achieved economies of scale in production of robot dogs – and it’s applying similar methods to humanoids. By designing for simplicity and using off-the-shelf components (where possible), Unitree keeps costs low. The company also adopts a consumer electronics style sales strategy (direct online sales of standard models), which is novel in the humanoid field.
  • Production & Cost Leadership: Unitree’s standout achievement is price. The new Unitree G1 is priced at just ¥99,000 (around **13,800**) for the base model[thestar.com.my](https://www.thestar.com.my/tech/tech-news/2025/02/05/battle-of-the-bots-china-us-scrap-for-top-of-the-humanoid-heap#:~:text=Unitree%2C%20based%20in%20the%20eastern,RM304%2C073%29%20its%20competitors%20charge)[micweekly.com](https://www.micweekly.com/top-humanoid-robot-companies-in-china/#:~:text=of%20joint%20movement%20angles%20that,G1%20robot%20is%2099%2C000%20RMB). This is an order of magnitude cheaper than most competitors (many humanoids cost n70k–$150k or more). Unitree has essentially begun mass production of the G1 – it announced it is taking orders and plans to produce large quantities. Founder Wang stated the price may drop further as they scale, and he projects annual shipments in the hundreds of thousands within a few yearswww.thestar.com.my. While that figure may be optimistic, it signals Unitree’s intent to capture market share through volume. In fact, Unitree sees a 3–5 year horizon (2025–2028) for clear commercial use cases to emerge and wants to be ready with massive production by thenwww.thestar.com.my. The company has built out production lines in Hangzhou and leverages China’s dense network of robotics parts suppliers (motors, reducers, sensors) to crank out units at low cost. By 2024, Unitree was already shipping prototypes and claiming the ability to fulfill significant orders. If it can maintain even basic functionality, Unitree could saturate certain niches (like education, entertainment, simple security tasks) with affordable humanoids before others scale up.
  • Applications & Target Market: Initially, Unitree’s humanoids are aimed at interactive service roles and light tasks. The H1’s public dance was a statement that these robots can be entertainers or attractors in events, theme parks, or marketing displays. Unitree also mentions uses like inspections and monitoring – a humanoid that can walk around a facility, read gauges, or carry a camera could serve in security or routine inspection for power plants and suchqviro.com. Another target is education and research: a low-cost humanoid is valuable for universities or labs that can’t afford a $100k robot – Unitree could sell dozens to schools to tinker with AI and robotics. In the consumer realm, down the line, a Unitree robot might act as a household helper or simply a tech toy for enthusiasts, given the relatively accessible price. Unitree is essentially trying to democratize humanoid robots by making them as common as personal drones or 3D printers became when their price dropped. That said, because of cost constraints, the current models likely have limited autonomy – they might rely on teleoperation or cloud assistance for complex tasks. It’s expected Unitree will continuously improve the AI (perhaps via partnering with Chinese AI firms) to make their cheap robots smarter over time.
  • Roadmap: Unitree envisions a breakthrough in humanoid capabilities by 2026, believing that more advanced perception and decision-making will come together with their refined hardware by that time (as the founder told media)www.scmp.comwww.scmp.com. In 2025, the focus is on ramping production of G1 and seeding the market. They have set an aggressive goal: if they truly aim for “hundreds of thousands” of units annually, they may release consumer-oriented versions or partner with a major consumer electronics firm to distribute them. For example, it wouldn’t be surprising if Unitree collaborates with a company like Xiaomi (which has its own robot initiative) to co-brand or integrate smart home features. Technically, we can expect Unitree to upgrade their humanoids with better hands and smarter AI gradually, but keep the core design stable to churn out units. By 2028, if Unitree’s plan succeeds, they could have one of the largest humanoid fleets in operation – albeit performing relatively simple tasks. They are best positioned to win on scale and cost, which can drive learning: as one Chinese expert noted, “the more units you sell, the more data you gather…that data helps make the robots smarter”www.thestar.com.mywww.thestar.com.my. Unitree is embracing this fast iteration, which could, in time, narrow the capability gap with more expensive robots.

Fourier Intelligence – GR-1

Fourier Intelligence is a Shanghai-based robotics firm that pivoted from rehabilitation devices to humanoid robots. In 2023 it launched the GR-1, a general-purpose humanoid aimed at healthcare and research applications. Fourier’s GR-1 stands out for its strength and collaboration strategy – it’s designed to assist in physical tasks like patient lifting, and Fourier is distributing units to partners to co-develop its capabilities. Key points:

  • AI & Capabilities: The GR-1 is a human-sized humanoid (around 165 cm tall, weighing 50+ kg) with a focus on high payload capacity. Its powerful actuators (notably in the hips) can deliver up to 300 Nm of torque, enabling the robot to carry up to 50 kg – nearly its own weightnewatlas.comnewatlas.com. This is an impressive feat: for comparison, Boston Dynamics’ Atlas can only carry ~11 kgnewatlas.com. Such strength means GR-1 could potentially pick up an adult human in a caregiving scenario. Fourier specifically envisions GR-1 as a caregiver or mobility assistant: helping immobile patients transfer between beds and wheelchairs, for examplenewatlas.com. The robot’s locomotion and balance are also well-developed. Fourier has shown GR-1 walking steadily, navigating around obstacles, and withstanding some shoves (a gentle version of the “broomstick test” where the robot regains balance after a push)newatlas.comnewatlas.com. GR-1 has human-like arms and hands (with fingers) to perform tasks like picking up delicate objects (it can grasp a glass bottle without breaking it)newatlas.com. While its movements were somewhat slow and cautious in demos, they were precise – indicating good control systems. On the “brain” side, Fourier is incorporating conversational AI and emotional interaction systems (the robot has a screen face to express and display infonewatlas.com). It likely uses a combination of onboard and cloud AI for vision and dialogue, given Fourier’s background in “cloud intelligent robotics.” Overall, GR-1’s technical sophistication is high, though in 2023 it was still in a prototype/software-tuning stage (not yet performing complex unplanned tasks autonomously).
  • Funding & Origins: Fourier Intelligence was founded in 2015 and made rehabilitation robots and exoskeletons prior to the humanoid venturenewatlas.com. This gave them relevant expertise – for instance, a lower-body exoskeleton uses many similar components and control algorithms as a bipedal robot’s legsnewatlas.com. The company has likely been funded by Chinese tech investors and has partnerships in research (e.g. with universities and the Shanghai government). While not as high-profile as some startups, Fourier has substantial resources from its medical robotics business and in 2023 it received a boost by aligning with the national push on humanoids. CEO Gu Jie has spoken about mass production of humanoids being closer than ever, indicating fundraising efforts to scale GR-1kr-asia.com. NVIDIA has spotlighted Fourier in its developer programsdeveloper.nvidia.com, suggesting access to cutting-edge AI hardware. In short, Fourier isn’t as cash-rich as, say, Figure, but it’s well-connected and financially stable from its established product lines.
  • Production & Strategy: Rather than immediately selling GR-1 to end-users, Fourier’s approach in 2023–2024 was to produce an initial batch of 100 GR-1 units and distribute them to research labs and partnersnewatlas.com. These units (manufactured in a pilot production run) effectively seed an ecosystem of developers working on humanoid applicationsnewatlas.com. By having 100 different teams experiment with GR-1, Fourier hopes to crowdsource improvements – a “hyper-collaborative” approach to accelerate the robot’s learning and software developmentnewatlas.comnewatlas.com. It’s a bit like how computer companies seeded PCs to developers in early days. This strategy may yield better AI and more use-cases for GR-1 before a wide commercial release. In terms of manufacturing proper, Fourier signaled it plans to scale up after this initial run. They claimed “world’s first mass-produced humanoid robot” for GR-1, which might be more marketing than reality, but they are indeed among the first to build dozens of a advanced humanoid at oncenewatlas.com. With additional funding, Fourier could set up a production facility to make hundreds annually, especially if there’s demand from hospitals or overseas labs (Fourier has offices in Singapore and partners globally due to its rehab products).
  • Applications & Partnerships: The immediate application focus is healthcare and eldercare. China (like many countries) faces an aging population, and robots like GR-1 could help in nursing homes or private homes by monitoring seniors, fetching items, or physically supporting patientsnewatlas.comnewatlas.com. Fourier’s co-founder Zen Koh specifically mentioned GR-1 could act as a therapy assistant or home companion for the elderlynewatlas.com. This niche differentiates Fourier from others targeting factories. Additionally, Fourier’s robots could be used in rehab centers – for example, helping patients practice walking (acting as a safety harness or even as a trainer by physically moving limbs). Outside healthcare, Fourier is eyeing general service tasks: one can imagine GR-1 guiding visitors in a mall, or doing light housework, given its strength to carry groceries or move furniture. The partnership with NVIDIA (through the Inception program) and others implies they are embedding strong AI capabilities (e.g., using vision to detect if an elderly person has fallen, and then physically helping them up). By giving GR-1 to universities (some likely went to Chinese universities and possibly international ones), Fourier is also fostering research that could expand its capabilities – e.g., someone might program GR-1 for firefighting or lab research tasks.
  • Roadmap: In 2024, Fourier’s goal was to get those 100 robots into the field and incorporate feedback. By 2025, we can expect Fourier to launch a more polished GR-1 v2, perhaps with improved hands or faster operation, ready for limited commercial deployments (such as in a few hospitals or clinics as trial caregivers). If those go well, a larger production run could follow. The company’s founder has spoken about mass production coming “closer than ever,” and being confident that technical challenges can be overcome now by focusing on systems integration rather than isolated problemswww.thestar.com.mywww.thestar.com.my. This suggests Fourier is in the stage of engineering optimizations and scaling, rather than fundamental R&D. By 2026–2028, Fourier might partner with a major healthcare provider or government program to roll out GR-1 robots in eldercare facilities across some regions, which would represent scaling with purpose. Compared to others, Fourier might not ship tens of thousands of robots in that timeframe, but the depth of capability in each GR-1 could be very high (a single GR-1 might replace several human caregivers worth of lifting/monitoring tasks). In sum, Fourier Intelligence is best positioned as a specialty humanoid provider – slightly slower on mass scale, but maintaining high dexterity and strength for critical applications, and leveraging partnerships to steadily improve their robot’s intelligencenewatlas.comnewatlas.com.

AgiBot (Shanghai) – YuanDong and Lingxi Series

AgiBot (short for “Agile Robot”) is a newer Chinese startup that has burst onto the scene by already producing hundreds of humanoid robots. Based in Shanghai, AgiBot unveiled multiple humanoid models in 2024 under two families: “YuanDong” (for interactive service) and “Lingxi” (for flexible manufacturing)www.micweekly.com. AgiBot exemplifies the fast-paced, product-focused approach of Chinese startups in this sector. Key points:

  • Products & Capabilities: At its 2024 launch event, AgiBot showcased five commercial humanoid robotswww.micweekly.com. For example, the YuanDong A2 is an interactive service humanoid – likely used for customer engagement, with friendly appearances and smooth arm movements to greet, serve coffee, or play games with people. In contrast, the YuanDong A2-W (W for “Work”? or a variant for manufacturing) is built for factory tasks, demonstrated by lifting 30 kg weights and unpacking delivery boxes on stagewww.micweekly.com. AgiBot’s demos even included robots playing mahjong, underscoring fine manipulation and coordination. These units appear to be human-sized, bipedal robots with decent dexterity (perhaps 5-fingered hands or at least multi-finger grippers to handle mahjong tiles). The performance of lifting 30 kg indicates strong actuators and balance – putting AgiBot’s bots in a similar class to Fourier’s in terms of payload. Their motion seemed competent in controlled environments, though it’s unclear how advanced their autonomy is. Likely, AgiBot uses a lot of cloud AI and teleoperation for complex tasks initially, focusing on getting physical hardware out in real scenarios quickly.
  • Production Achieved: Remarkably, AgiBot announced that by end of 2024 it had produced 962 robot units, with about 700 already shipped to clients and 200+ used internallywww.thestar.com.my. These numbers include both bipedal humanoids and wheeled service robots, but it signals significant manufacturing activity. Few companies in the world can claim to have built that many humanoid-form robots. AgiBot’s ability to do so likely comes from leveraging existing supply chains (perhaps using components from industrial robots and drones) and a modular design. The fact that over 700 are shipped means there are customers actively deploying them – possibly shopping malls, tech parks, or local government projects where service robots are in demand. This immediate traction gives AgiBot real-world feedback and some revenue. They’ve essentially gone straight into selling “beta” robots to early adopters, a bold strategy that can help them refine quickly.
  • Market Focus & Support: AgiBot’s robots are intended for interactive services and manufacturing assistancewww.thestar.com.my. In China, there’s a big market for humanoid greeters, entertainers, and guides – think robot receptionists at banks or robot teachers for kids’ STEM classes. AgiBot is tapping into that with YuanDong A2. For manufacturing, they aim to have humanoids that can be quickly trained to do different assembly or handling tasks, providing flexibility where fixed automation is not suitable. The startup likely has local government backing, as Shanghai has been keen on establishing itself as a robotics hub. Being Shanghai-based, AgiBot might also collaborate with manufacturing giants in the Yangtze Delta region. The mention in media of an impending “mass production” suggests authorities and investors are watching it as a potential success story. Notably, Chinese experts warned that if companies can’t keep prices under ¥200k in 2025, it’ll be hard to push productswww.thestar.com.my– AgiBot seems aware and is probably pricing aggressively (though not publicly disclosed, their shipments imply a competitive price point).
  • Scaling & Competition: AgiBot operates in a crowded domestic market, so it’s driving hard to capture share. A professor He Liang (who runs another humanoid firm) noted a bit of a price war already in China, with razor-thin margins and a “fast fish eats slow fish” mentalitywww.thestar.com.mywww.thestar.com.my. AgiBot exemplifies this – it’s moving fast to deploy robots even if each unit isn’t hugely profitable, aiming to establish brand and get volume. By achieving nearly 1000 units in a short time, AgiBot is arguably ahead of many Western companies in pure numbers. Their challenge will be ensuring quality and functionality keep up – mass production of complex robots is non-trivial. But if they succeed, AgiBot could emerge by 2025–2026 as one of the highest-volume humanoid producers globally, especially in the service robot segment.
  • Future Plans: AgiBot has signaled it is starting mass production imminently (as of early 2025)www.thestar.com.my. This suggests they are transitioning from small-batch builds to an actual assembly line for their humanoids. They might target thousands of units in 2025. International expansion could also be on the horizon – perhaps selling service robots to other Asian countries or the Middle East, where demand for novelty robots in malls and expos is high. Technically, we can expect AgiBot to enhance the AI of its robots (more natural language processing for the service model, more autonomy for the industrial model). Given their rapid model rollout (five models at once in 2024), they’ll likely consolidate to a few core designs that can be manufactured in scale. Over 2026–2028, AgiBot will either scale up dramatically (if early customers are happy and reorder in larger quantities) or risk being outpaced if their robots don’t perform as hoped. Right now, they appear well-positioned in China’s ecosystem – they have momentum. Their bet is that being first to market in volume will give them a data advantage and learning curve benefits, which in turn will improve their robots quickly, creating a tough-to-catch lead. It’s an approach mirroring how some Chinese tech firms operate (fast, iterative, volume-driven), which could indeed make AgiBot a major player if the quality holds up.

Xiaomi – CyberOne

Electronics giant Xiaomi introduced CyberOne in 2022 as its first full-size humanoid robot. While more of a high-profile demo than a fully commercial project, CyberOne underscores China’s big tech interest in humanoids and Xiaomi’s potential to scale hardware. Key points:

  • Concept & Features: CyberOne stands 177 cm tall and weighs 52 kg, built in a standard human form. Xiaomi endowed it with a suite of sensors and AI for interaction – it is said to recognize 85 environmental sounds and 45 human emotions via its vision and hearing systemsqviro.com. It uses Xiaomi’s in-house Mi-Sense depth vision module for 3D environment mappingwww.micweekly.com, and has an “advanced cerebellum” for balance, allowing bipedal walking and posture control. Its limbs can generate up to 300 Nm torque (comparable to Fourier’s and Tesla’s robots in strength)www.micweekly.com. In demos, CyberOne walked across the stage, gave flowers to Xiaomi’s CEO, and responded to some voice commands – showing basic mobility and social interaction. The robot’s head has a curved LED display to show facial expressions. Essentially, CyberOne is a showcase of Xiaomi’s ability to integrate sensors, motors, and AI algorithms into a humanoid package.
  • Xiaomi’s Approach: As a top consumer electronics manufacturer (smartphones, appliances, etc.), Xiaomi has the supply chain and capital to mass-produce complex electronics cheaply. CyberOne is likely a research platform for now, developed by the Xiaomi Robotics Lab. Xiaomi also has a smaller quadruped robot (CyberDog) which it did release in limited quantities. If Xiaomi sees a viable market, it could leverage its massive manufacturing capabilities to produce humanoids at scale. CyberOne is still early-stage – after 2022 we haven’t seen a mass rollout – but Xiaomi continues to invest in robotics and AI. Notably, Xiaomi is one of the stakeholders in China’s Embodied Intelligence Robot Center (with UBTECH)en.tmtpost.comen.tmtpost.com, indicating it’s collaborating on core tech.
  • Status & Plans: There’s no sign yet of Xiaomi selling CyberOne units; it appears to be in the R&D phase. However, Xiaomi’s CEO Lei Jun hinted that this is a long-term development and that iterations will continue. If Xiaomi can refine the robot’s capabilities (especially autonomy and battery life) sufficiently, it might pilot them in its Xiaomi stores or for marketing events to gauge public interest. A feasible path for Xiaomi could be a “Xiaomi home robot” that ties into its smart home ecosystem – a humanoid that can control Xiaomi IoT devices, carry objects, and provide companionship or security. Given Xiaomi’s brand, if they do commercialize, they would likely aim for a relatively affordable price point (perhaps by limiting the robot’s actuators or using some toy-grade components as they did with CyberDog).
  • Impact: Even if CyberOne itself isn’t mass-produced yet, Xiaomi’s involvement has symbolic importance. It signals that not only startups but also tech giants are in the humanoid race in China. This brings hefty resources and competitive pressure. Xiaomi can also contribute key technologies – for example, efficient batteries, cost-optimized servo motors, and consumer-grade AI assistants – which could trickle down to other players or raise the bar for user experience (Xiaomi excels in user-centric design, which could make its humanoid more user-friendly in interactions than some engineer-focused rivals). In the 2025–2028 timeframe, if Xiaomi decides to seriously enter the market, it could quickly scale to thousands of units given its manufacturing muscle. It might also acquire or partner with a startup (much like how Tesla acquired tech for its projects) to jump-start development. In summary, Xiaomi’s CyberOne is an advanced prototype with human-interaction flair, and while not yet commercial, Xiaomi’s potential to scale and popularize humanoid robots in daily life is huge should they choose to pursue it.

Others in China’s Humanoid Ecosystem

China’s humanoid robotics scene is extremely vibrant, with numerous startups and research institutions contributing. A few notable ones include:

  • Leju Robotics: A Shenzhen-based firm that launched the Leju Kuavo humanoid in 2023. It runs on Huawei’s Harmony OS and is capable of dynamic motions like jumping. Weighing ~45 kg with 26 DOF, it can jump 20+ cm and walk 4.6 km/hwww.micweekly.com. Leju emphasizes an open-source approach and aims at research and entertainment applications.
  • EX Robots (EXROBOTS): Focuses on realistic “human-like” robots with expressive faces and dexterous hands. CEO Li Boyang says their edge is in emotive interaction – their humanoids have lifelike facial expressions, voice conversation, and gesture controlwww.micweekly.com. EXROBOTS claims it has 200+ bionic humanoids already and is gearing up to mass-produce 500+ units, targeting use in museums, education, retail and eldercare as companionswww.micweekly.comwww.micweekly.com. These are more android-like (emulating human looks) than utility robots.
  • Qingdao Qingbo (Qingbao) Robot: A startup (founded 2022) developing cloud-driven humanoids. Qingbao’s Beta series humanoids combine on-board motors with a cloud “brain” system for AI and have modules for micro-facial expression and voice interactionwww.micweekly.comwww.micweekly.com. They sold “dozens of units” in 2024 according to their chairmanwww.thestar.com.my. Used in venues like airports, shopping centers, and coffee shops as greeters and service staff.
  • RobotEra (Hangzhou): Maker of the “Star One” humanoid, focusing on strong joint actuators and advanced materials (carbon fiber, etc.) for a light yet strong robotwww.micweekly.comwww.micweekly.com. Integrating large language models and force control, their “Xiao Xing” robot family aims for both agility and AI understandingwww.micweekly.com.
  • Horizon Robotics / Cambricon: These aren’t robot makers but chip designers. They provide the AI processors that many Chinese humanoids use for edge computing. Cambricon, for instance, offers high-performance vision chips that can be used to run neural networks on the robotwww.allianzgi.com. This domestic semiconductor support helps Chinese firms reduce reliance on foreign AI chips, which is crucial under geopolitical tech restrictions.
  • National Research Initiatives: China’s government has set explicit goals (e.g. develop 100 humanoid applications by 2025)en.tmtpost.comand established innovation hubsen.tmtpost.comen.tmtpost.com. Academic labs like those at Tsinghua, Beihang, and Northwestern Polytechnical University are deeply involved. They often collaborate with companies – for example, Prof. He Liang at NPU leads a humanoid startup and contributes to industry conferences with insights on system engineeringwww.thestar.com.mywww.thestar.com.my. This tight academia-industry linkage means a steady pipeline of talent and breakthroughs feeding the companies mentioned above. In short, beyond the headline companies (UBTECH, Unitree, Fourier, AgiBot, Xiaomi), there is a swarm of innovative Chinese startups tackling pieces of the humanoid puzzle – whether it’s low-cost actuators (e.g. companies like LimX Dynamics with a stair-climbing botwww.micweekly.comwww.micweekly.com) or full humanoids for niche markets (like Booster Robotics’ small humanoid that can do push-ups and play soccerwww.micweekly.com). This ecosystem approach, backed by large state and private investments, gives China a robust foundation to scale humanoid production in the coming years.

Comparison and Outlook (2025–2028)

Both the US and China are racing to develop intelligent humanoid robots at commercial scale, but they play to different strengths:

  • Technical Sophistication: US companies often lead in cutting-edge AI integration and software. For example, Tesla and Figure are heavily leveraging AI brains (LLMs, vision networks) to give robots higher-level autonomy and human-like understandingwww.reuters.comwww.reuters.com. Boston Dynamics sets the bar in raw agility and dexterity with Atlasbostondynamics.combostondynamics.com, and its know-how influences the whole field. Chinese companies, on the other hand, excel in rapid hardware iteration and mechatronics – UBTECH, Unitree, and others have developed robust actuators and control systems (often on par with Western designs) and focus on integrating proven technologies into full robots for practical taskswww.thestar.com.mywww.thestar.com.my. China’s robots are increasingly matching global peers in mobility and adding their own innovations (e.g. Unitree’s ultra-low-cost design, or UBTECH’s multi-robot coordination in factorieswww.micweekly.com). In short, US robots might be “smarter” out-of-the-box, while Chinese robots might be more optimized for doing specific jobs reliably. However, this gap is narrowing as know-how crosses borders (often through open research and conferences).

  • Scaling Approach: US players tend toward a centralized scale strategy – a few well-funded companies aiming to achieve massive production. Tesla is the clearest example: if Optimus succeeds, Tesla alone could pump out millions of units, essentially becoming for humanoids what Ford was for carswww.thestar.com.mywww.thestar.com.my. Similarly, Agility Robotics building the first humanoid factory signals a centralized scaling in the US (with tens of thousands of Digits a year from one site)www.geekwire.comwww.geekwire.com. China’s scaling is more distributed among many companies. Individually, Chinese firms might produce at lower rates than Tesla’s ambitionswww.thestar.com.mywww.thestar.com.my, but cumulatively, dozens of Chinese manufacturers could equal or surpass the output – a “many small rockets vs one big rocket” analogywww.thestar.com.mywww.thestar.com.my. Chinese firms also benefit from supply chain maturity and cost advantages by sourcing locally in volumewww.allianzgi.comwww.thestar.com.my, which helps when scaling up. The downside for China’s fragmented approach is potential duplication of effort and varying standards, but government coordination is improving that via standards centersen.tmtpost.comen.tmtpost.com.

  • Funding & Investment: The US humanoid sector has seen massive injections of capital from private investors and Big Tech. Figure’s 675M round (with Nvidia, Microsoft, Bezos, etc.)[reuters.com](https://www.reuters.com/technology/robotics-startup-figure-raises-675-mln-microsoft-nvidia-other-big-techs-2024-02-29/#:~:text=Feb%2029%20%28Reuters%29%20,6%20billion)[reuters.com](https://www.reuters.com/technology/robotics-startup-figure-raises-675-mln-microsoft-nvidia-other-big-techs-2024-02-29/#:~:text=Other%20companies%20involved%20in%20the,Align%20Ventures%20and%20ARK%20Invest)and Apptronik’s n350M (with Alphabet participation)www.reuters.comwww.reuters.comgive those startups huge war chests. This means they can hire top talent, invest in long-term R&D, and absorb initial losses to refine their robots. Tesla, of course, leverages its (multi)billionbudgetandgeneratesitsowncashtofundOptimus.InChina,fundingisamixof∗∗privatecapitalandstatesupport∗∗.CompanieslikeUBTECHhadlargeVCrounds(e.g.(multi)billion budget and generates its own cash to fund Optimus. In China, funding is a mix of **private capital and state support**. Companies like UBTECH had large VC rounds (e.g. n820M in 2018) and continue to attract investors, but also align with state programs that provide grants and incentivesen.tmtpost.comen.tmtpost.com. There is also a nationalistic drive to own the humanoid technology stack (especially as the US restricts exports of certain AI chipswww.thestar.com.mywww.thestar.com.my). This may insulate Chinese companies from some global investment flows, but domestic funding is ample. Bottom line: both countries have significant capital pouring in – the US from tech giants and venture capital, China from a combination of venture, corporate (e.g. Xiaomi, Huawei investing in the space), and government channels. The best-positioned companies are those who use funding not just to build robots, but to build the manufacturing and support ecosystem around them (which Tesla, Agility, and UBTECH are clearly doing with factories and partnerships).

  • Strategic Partnerships: Partnerships often indicate who can accelerate deployment. In the US, we see tech-manufacturing tie-ups: e.g. Apptronik with Jabil for manufacturing and with Mercedes/GXO for early adoptionapptronik.comwww.reuters.com; Figure with OpenAI and Microsoft for AI and with BMW for testingwww.reuters.comwww.reuters.com; Agility with Amazon and Fordwww.geekwire.comwww.geekwire.com. These give US companies access to domain expertise and trial environments (factories, warehouses) to prove their robots. In China, partnerships often involve end-users and local governments: UBTECH securing Foxconn and SF Express dealswww.tweaktown.comwww.tweaktown.com, or city governments deploying robots in public spaces (a common arrangement). Additionally, Chinese firms partner with research institutes – e.g. many are spin-offs from universities (like He Liang’s startup from NPU). This academic partnership ensures a pipeline of new ideas and skilled engineerswww.thestar.com.mywww.thestar.com.my. The company best poised to scale will have strong alliances: Tesla, again, is unique in that its “partnership” is internal (it will be its own first big customer, automating Tesla factories). That vertical integration is powerful. Meanwhile, a company like UBTECH, by partnering with Foxconn (which could deploy robots across dozens of factories if pilots go well), might suddenly scale out hundreds of units quickly in a real setting – a huge advantage in learning and refinement.

  • Commercial Readiness (2025–2028): By 2025, multiple players plan to begin mass productionwww.thestar.com.my. It’s a critical milestone where prototypes turn into products. Those who hit this window with a reliable product will gain early market share and feedback. Agility will have Digits in warehouses; UBTECH will have Walker in factories; Tesla aims to have Optimus in limited production; Apptronik and Figure plan initial deployments. We will likely see a divergence in use-cases: US robots more often in logistics and manufacturing (where labor shortages and high value tasks align with early robot capabilities), and Chinese robots also visible in public-facing service roles (due to societal acceptance and novelty factor, you might see humanoids as receptionists or restaurant servers in China sooner). Over 2026–2028, if things go well, these roles will start to blur as each side expands (e.g., US robots entering retail and hospitality, Chinese robots entering more factory settings). The best-positioned companies are those that can maintain or improve their robot’s intelligence and dexterity while scaling up production – a tricky balance. It’s not enough to ship thousands of robots if they’re too “dumb” to perform useful work; conversely, having the smartest robot in the lab is moot if you can’t produce it affordably. Who is best positioned to scale widely with high capability? Based on current insights:

  • Tesla has an unmatched scaling capacity and a clear plan to utilize itwww.thestar.com.my. If Optimus meets even part of its promises, Tesla could leap ahead in sheer volume. Tesla’s strength is integrating advanced AI (vision, navigation) with cost-efficient design – however, Optimus is still proving its dexterity and generality. Should Tesla solve those technical hurdles by 2026, its million-per-year vision makes it a likely dominant playerwww.thestar.com.my. Tesla could effectively commoditize humanoid hardware like it did electric vehicles. The risk is that Optimus underperforms intellectually (early hints show it’s still catching up to others in fine manipulationen.wikipedia.orgen.wikipedia.org). Yet, with Tesla’s AI team and Dojo supercomputer resources, one can’t count them out on intelligence.

  • UBTECH appears best poised on the Chinese side to deploy widely. It combines strong tech (a capable biped with arms) and concrete manufacturing partnerships nowwww.tweaktown.com. By targeting known industrial needs (where robots don’t have to be geniuses, just competent and consistent), UBTECH can scale while iterating its AI with each generation. Delivering 1,000 robots to real clients by 2025 will give it a huge dataset and experience advantagewww.tweaktown.comwww.tweaktown.com. UBTECH might not match Tesla in AI research, but its robots are getting smarter through practical use and it’s actively incorporating large modelswww.micweekly.com. This pragmatic path of “learn by doing” at scale could yield robots that are both reasonably dexterous and proven in the field – a strong combination.

  • Agility Robotics is in an excellent position in the US for a specific market (warehousing). By 2028, Agility could have thousands of Digits working across Amazon warehouses and other logistics centers. Its robots are deliberately limited in “general” intelligence, focusing on a narrower set of tasks very effectively. This means high reliability and a simpler path to scale production. As long as demand for warehouse automation stays high, Agility can scale widely. They may gradually improve Digit’s manipulation to expand its use cases, but even as-is, Digit doesn’t need human-level dexterity to be valuable (simple hands suffice for tote handling). So Agility may become the first to deploy humanoid-ish robots at scale in a profitable, sustainable waywww.geekwire.comwww.geekwire.com. They might not have the flashiest AI, but their integration with industrial workflow and proven ROI could beat out more experimental rivals in the near term.

  • Apptronik and Figure are wildcards with tremendous momentum. By 2028, each could conceivably catch up or surpass others if they execute well. Apptronik, with 350M and backing from enterprise partners, has the pieces in place: a solid hardware team, manufacturing partner, and AI partnership (Google)[reuters.com](https://www.reuters.com/technology/apptronik-raises-350-million-scale-production-humanoid-robots-2025-02-13/#:~:text=Feb%2013%20%28Reuters%29%20,powered%20humanoid%20robots)[reuters.com](https://www.reuters.com/technology/apptronik-raises-350-million-scale-production-humanoid-robots-2025-02-13/#:~:text=The%20company%2C%20which%20said%20it,including%20elder%20care%20and%20healthcare). If Apollo performs in Mercedes factories and GXO warehouses, orders will follow. Apptronik’s focus on an “Apollo for every industry” (including eldercare later) shows they aim for versatility – maintaining dexterity and intelligence while scaling. Their use of proven manufacturing (Jabil) reduces risk of supply issues, meaning they can iterate on Apollo’s design without worrying about factory setup delays. **Figure**, with its tech titan-backed n675M, arguably has the highest ceiling if it can realize its visionwww.reuters.comwww.reuters.com. Partnering with OpenAI and having Nvidia/Microsoft onboard suggests Figure’s humanoid might achieve very advanced cognitive and interactive abilities, effectively giving it a supercharged “brain”www.reuters.comwww.reuters.com. If their hardware matches the AI, Figure’s robots could be among the smartest and still be built at scale thanks to all that capital (and perhaps manufacturing via partners like Foxconn or BMW). The question for Figure is timeline – they started later, so they need to go from prototype to mass production swiftly to compete by 2028. Given their funding, they very well might.

  • Unitree and similar Chinese low-cost contenders might flood certain markets (education, hobbyist, light-duty service). Unitree’s strategy of high volume at low cost means by 2028 one could see tens of thousands of Unitree G-series humanoids in schools, shopping centers, and even homes, especially in Asia. While these may not have top-tier dexterity, they’ll benefit from continual upgrades and data. Unitree explicitly notes that data from mass deployment will improve their AIwww.thestar.com.mywww.thestar.com.my, potentially allowing simple robots to get smarter over time. If Unitree hits its marks, it will have established a de facto standard for entry-level humanoids (much like DJI did for drones). This could pressure everyone on price and force the whole industry to be more accessible. In conclusion, the race is not zero-sum – multiple winners may emerge serving different segments. The US companies best positioned to scale widely with high capability are Tesla (for its manufacturing might and AI integration) and Agility (for its focused, market-ready product), with Figure and Apptronik as strong runners-up likely to surprise with rapid progress. In China, UBTECH leads in marrying sophistication with scale (delivering real industrial value soon), while Unitree leads in scale via affordability, and Fourier/others add depth in capability. China’s overall ecosystem advantage means even if no single firm there matches Tesla’s volume by 2028, collectively they will deploy a huge number of humanoids, ensuring they gather experience and improve fastwww.thestar.com.mywww.thestar.com.my.

Crucially, scaling widely while maintaining high intelligence and dexterity will require continuous innovation. Companies must iterate robots like software – those that can cycle through versions quickly (UBTECH with Walker S1→S2→S3 in two yearswww.tweaktown.com, Tesla with Optimus Gen2 in a yearen.wikipedia.org) have an edge. Additionally, companies that treat deployments as learning opportunities rather than just sales will advance their robots’ intelligence the fastest. In that sense, the best positioned are those combining big production plans with big data/AI loops: Tesla intending to learn from each Optimus in the fieldwww.thestar.com.my, UBTECH and Unitree racing to place robots in many environments to gather diverse data, and Figure explicitly training on robotics datasets to make its AI robustwww.reuters.comwww.reuters.com.

By 2028, we expect to see humanoid robots moving beyond tech demos into everyday work – lifting boxes, fetching tools, guiding visitors, or caring for patients. The companies profiled are at the forefront of this revolution. Those that can scale up without compromising on performance will shape the future of how widely humanoid robots are adopted. Right now, it’s an open field with Tesla and Chinese firms neck-and-neck in this “space race” for robotics supremacywww.thestar.com.mywww.thestar.com.my. The next three years will be pivotal in determining whether the world’s first truly ubiquitous humanoid workforce comes out of an American factory line or a Chinese one – or both. Either way, the progress is accelerating, and both tech superpowers are driving it with their leading companieswww.thestar.com.mywww.thestar.com.my.

Sources: