Chinese Humanoids Go to Work. The US is Starting to Fall Behind.

"Soon, robots will start working alongside humans. That's what engineers and analysts say. It also looks like most of these robots will be "Chinese." The market for AI-powered humanoid machines is already impressive there.

 

Although most of the humanoid robots presented at numerous conferences, trade fairs, and exhibitions are still not efficient or mobile enough to compete with humans, experts predict that robots will have a huge impact on the job market and the economy. Amazon is already testing robots from the American startup Agility and – according to internal memos leaked to the media – expects to replace a significant number of employees with them within a few years. Bank of America analysts predict that by 2035, approximately 10 million humanoid robots will be produced annually. Morgan Stanley predicts that by 2050, one billion such robots will be in use, with almost one-third – 302.3 million – in China, and 77.7 million in the US.

 

Chinese companies are not lagging behind American manufacturers of humanoid robots, and it even looks like they are starting to overtake them.

 

Unitree is ahead of the competition. The Chinese leader in humanoid robots

 

Probably no company producing humanoid robots today has such an advantage as Unitree Robotics from Hangzhou – according to the "Wired" portal. While Elon Musk's Optimus wobbles during demonstrations, Unitree robots sprint, perform kung-fu kicks, and acrobatic backflips.

 

This company's walking robots are also incredibly cheap – costing tens of thousands of dollars or less, which is about one-tenth the price of a typical humanoid in the US. Unitree is the most well-known Chinese robotics startup, a national champion in the technology industry, and is reportedly preparing for a $7 billion IPO in Shanghai, according to Wired.

 

However, even if Unitree fails, there are already over 200 other Chinese companies working on humanoid robots. In the US, there are about 16 significant companies building bipedal robots.

Humanoids are becoming cheaper. Chinese mass production is changing the market.

 

With such numbers, it's hard not to suspect that the first country to have a million humanoids will be China.

 

Unitree, based in Hangzhou, about 180 kilometers from Shanghai, sells a wide range of walking robots. Their offerings also include other models – for example, dog-like robots, four-legged machines similar to those produced by the American company Boston Dynamics.

 

In 2023, Unitree's sales of four-legged robots were ten times higher than Boston Dynamics', according to data from the consulting firm SemiAnalysis. Nearly 24,000 such robots were sold for applications including construction sites, oil rigs, and factories. The robots can climb stairs and navigate rubble to conduct inspections or security patrols.

 

In 2023, Unitree unveiled its first model, the H1 – a bipedal humanoid robot. Less than a year later, the company began selling a more advanced model, the G1, which costs just over $13,500. This summer, the company showed an even cheaper, but less advanced, bipedal robot, the R1, costing only 39,999 yuan ($5,700).

 

The "ChatGPT Moment" for Robots: Artificial Intelligence Learns the Physical World

 

The next breakthrough in the development of humanoid robots is expected to be the so-called "ChatGPT moment," when machines will gain "intelligence." According to analysts, this breakthrough is a matter of three to five years.

 

At the Beijing Academy of Artificial Intelligence (BAAI), a non-profit organization founded in 2018 by the authorities of the Chinese capital to coordinate AI research, work is underway on a new open robotic model, Robobrain 2.0, designed to combine the capabilities of language models with spatial reasoning [1].

 

In the US, it has been alleged that BAAI is attempting to acquire American technology for the modernization of the Chinese military. BAAI employees, however, claim it is simply a research center. Nevertheless, in May 2025, the Department of Commerce added the institute to its sanctions list, prohibiting it from acquiring American technologies.

 

Despite this, according to Wired magazine, many heads of American robotics companies fear competition from China and believe they cannot keep up with the pace set by the Chinese in this field. Chinese companies have the support of their government – ​​financial, legal, and technological – while American manufacturers currently have to manage on their own.”

 

That is not true. Let’s compare in dollars government support for Unitree humanoids and Tesla's Optimus

Based on available reports, both Unitree and Tesla have received significant, though differently structured, support from their respective governments. While Tesla has historically benefited from billions in loans, tax credits, and regulatory support for its electric vehicle business—which now aids the development of the Optimus robot—Unitree is receiving direct, targeted support from the Chinese government as part of a national strategy to lead the humanoid industry by 2027.

Unitree (China) Government Support

 

    Targeted National Strategy: The Chinese government has identified humanoid robots as a key future industry, with specific plans to establish a "world-class" industry by 2027.

    State-Backed Funding: Unitree's funding rounds have included investment from state-backed entities, such as the Shenzhen Investment Holdings Co and the Beijing Robotics Industry Investment Fund.

    Support Ecosystem: Unitree has won public support and is featured in government-backed initiatives, such as the World Humanoid Robot Games in Beijing.

    Valuation & Investment: As of mid-2025, Unitree reached a valuation of approximately 12 billion yuan (approx. $1.7 billion USD), driven by rapid funding from both public and private sources.

 

Tesla Optimus (USA) Government Support

 

    Historical & Indirect Support: Tesla has benefited from at least $38 billion in government funding, including federal, state, and local incentives (rebates, tax credits, loans). While this was largely for electric vehicles, it provided the infrastructure and capital for advanced research.

    Regulatory Credits: Tesla has generated over $11 billion in revenue from selling regulatory credits to other automakers.

    Direct Aid: In 2010, Tesla received a $465 million loan from the Department of Energy to build its Fremont factory, which was fully repaid in 2013.

    Future Market Potential: The US government's focus on reshoring manufacturing may favor Tesla's industrial-focused, "made-in-USA" robots, though specific direct grants for Optimus are not publicized in the same way as Chinese state support.

 

Comparison Table

Feature            Unitree (Unitree Robotics)      Tesla (Optimus)

Location          China                                       USA

Primary Support Type            Targeted, state-backed investment & industrial policy           Regulatory     

                                                                                             credits, state/local tax incentives

Direct Funding Yes (e.g., Beijing Robotics Industry Investment Fund)          Historically $465M

                                                                                        loan (repaid); billions in EV tax credits

Focus  Rapid, low-cost commercialization (e.g., $16k G1)   High-capability industrial/factory

                                                                                                       deployment

Key Advantage           Lower cost, faster time to market       Vertical integration, AI capacity,               

                                                                                                   massive capital base

Summary: While Tesla operates with a much larger total pool of accumulated government and regulatory support ($38B+ over a decade), Unitree is benefiting from more direct and localized state investment aimed specifically at accelerating its development to compete with U.S. firms.

 

1. Is Robobrain 2.0 available in the West?

RoboBrain 2.0, unveiled in June 2025 by the Beijing Academy of Artificial Intelligence (BAAI), is

available globally as an open-source model via platforms like Hugging Face, making it accessible to developers and researchers in the USA and the EU.

While the technology is open-source, it is primarily designed by a Chinese research institute to power the next generation of Chinese humanoid robots, with 20+ Chinese partners already involved in its development.

Here are the key details regarding its availability:

 

    Release & Access: BAAI released the checkpoints for RoboBrain 2.0 (specifically the 3B, 7B, and 32B versions) on Hugging Face in June and July 2025.

    Open-Source Nature: It is accessible to the worldwide AI and robotics research community, meaning developers in the USA and EU can access the code and models.

    Focus: It is aimed at advancing embodied AI and is not currently a consumer product for sale, but rather a tool for robot manufacturers and researchers.

 

It is important to distinguish this BAAI RoboBrain 2.0 AI model (released June 2025) from other, unrelated hardware products that might use similar names. The BAAI model is designed to be the "brain" of robots rather than a specific physical camera or hardware unit.

 

RoboBrain 2.0, an embodied vision-language model by BAAI, requires robust hardware to run its 7B or 32B parameter models, specifically high-end NVIDIA GPUs for inference and substantial memory. It is built on modern AI frameworks like LLaVA-NeXT and vllm for processing vision-language tasks, requiring a Linux environment with CUDA support.

Hardware Requirements

 

    GPU: High-performance NVIDIA GPU capable of handling large-scale vision-language models (e.g., A100 or H100 recommended for 32B variant).

    Memory (RAM/VRAM): Significant GPU VRAM is necessary for the 32B model, with 16GB or more required for efficient processing.

    Processor: High-speed multi-core CPU (e.g., Intel Core i5/i7 or equivalent) to support data preprocessing.

    Storage: Sufficient SSD space for storing large models and datasets.

 

Software & Frameworks

 

    Operating System: Linux (typically Ubuntu) optimized for AI workloads.

    Frameworks: PyTorch, vllm (for efficient inference), and LLaVA-NeXT.

    Model Weights: Access to the pre-trained RoboBrain 2.0 (7B or 32B) model weights.

    Libraries: CUDA drivers, Transformers, and related AI packages for running visual encoder and LLM decoders.

 

For integration into physical robots, the system interfaces with vision sensors (cameras) and motor controls, often running through ROS (Robot Operating System).