China, fuelled by recent successes from DeepSeek and Unitree, is making a play for pole position in humanoid robots. Can the US hold its ground? — SCMP
At the Consumer Electronics Show (CES) in Las Vegas on Jan 7 – the tech sector’s premier trade exhibition – Nvidia CEO Jensen Huang took the stage for his keynote address. During the speech, where he declared the field of robotics was nearing a watershed moment, he was flanked by 14 gleaming, waving humanoid models from industry leaders.
Weeks later, at the Spring Festival Gala – China’s annual televised Lunar New Year extravaganza, watched by over a billion people – robotics firm Unitree put on a demo of a different sort.
A group of H1 units, which the company touts as its “first universal humanoid robot”, performed a folk dance synchronised with the aid of advanced cloud computing and motion control systems powered by artificial intelligence (AI) software.
Both crowds marvelled at the spectacles. The high-profile events have set the tone for a competitive 2025, as numerous robot makers marshal their resources for mass production and global commercialisation.
Players from all over the world are eager for their share of the spoils. While no one country appears to be completely dominant in this 21st century “space race,” China and the United States are undoubtedly in the first tier.
Among the group of 14 robots onstage at CES – which included a Unitree H1 – six of their companies were based in China and four were from the United States.
While the two countries are neck and neck in the race for robotics supremacy, each is playing to its own strengths. China is leaning into its capacity for scale and speed, with an advantage in supply chain maturity, but analysts said the US has the edge when it comes to producing the next Tesla or OpenAI – game-changing giants that redefine the industry.
“In the US, the focus is on staying at the forefront of technology, exploring unknown technical challenges; in China, the focus is more on how to integrate existing technologies for practical application,” said Xu Xuecheng, a lead scientist with the Zhejiang Humanoid Robot Innovation Center.
He argued that Chinese companies are broadly more invested in building fully integrated humanoid robots for real-world use, while US firms focus on a more generalised intelligence that could include functions factories would consider unnecessary.
That calculus may have changed after recent events.
Shortly before Unitree’s New Year showcase, Chinese AI start-up DeepSeek stunned the tech world after debuting its R1 large language model, which matches OpenAI’s products for performance at a fraction of the use and development cost.
The company’s open-source product has challenged the prevailing assumption that substantial capital is necessary to train state-of-the-art AI systems, leading to an enormous sell-off in the stock of DeepSeek’s competitors and a re-evaluation of previous investment strategies.
“(DeepSeek’s) algorithmic innovations remind us that China and the US are neck and neck and that our technological edge isn’t guaranteed, pushing our industry to make AI more efficient,” said Eric Schmidt, former CEO of Google, in an interview with the Financial Times.
The US does maintain some significant advantages in robotics, however. Its universities have produced numerous “0-to-1” breakthroughs in research, and its companies have spearheaded progress in the “soft” technologies underpinning the field.
Each humanoid unit combines a highly advanced “brain” – powered by a suite of AI, machine learning, and sensors – to perceive its environment and act, with a sophisticated “body” built from lightweight materials employing flexible joints for lifelike motion.
A seamless integration system ensures the brain and body work in harmony, enabling the robots to mimic human behaviour like speech, gestures and adaptive movements in complex environments.
During his CES keynote, Huang unveiled Cosmos – the cutting-edge generative models the Nvidia executive said will power the humanoid robots of the future.
The hardware for humanoid robots is mostly mature – or close to it – according to Goldman Sachs Research, with components like cameras, motors, force sensors, transmission gears and batteries all available for commercial purposes.
“Progress in end-to-end AI could potentially enable much faster humanoid robot iterations as seen from the improvement of manipulation and interaction capabilities of various products in 2023,” the investment bank said in a report last year.
Many companies have said they will begin mass production of their models in 2025, widely seen as a critical milestone for operational stability and a means of reducing costs by building up economies of scale.
China – long known as the “world’s factory” – has extended its manufacturing dominance to the world of robotics. Tens of thousands of companies have a role in the supply chain, whether producing components, assembling units or designing prototypes.
According to the International Federation of Robotics, more than half of global robot installations took place in China in 2023, giving its firms an edge on cost control as they move to enter the home market with humanoid models.
“If companies can’t keep their prices under 200,000 yuan (RM121,640 or US$27,825) this year, it will be very difficult to push their products,” said He Liang, a professor at Northwestern Polytechnical University in Xian.
Firms are operating on thin margins, and the main strategy is to rapidly capture market share, said He, who also runs a humanoid robot firm with international presence.
“To some extent, we’re already seeing a price war (in China).”
Unitree, based in the eastern city of Hangzhou, has also revealed its plans for production at scale, with its basic G1 model priced at an astonishing 99,000 yuan (RM60,214) – a fraction of the typical 500,000 yuan (RM304,073) its competitors charge.
Company founder Wang Xingxing has said costs will drop even further as production scales, with annual shipments projected to reach hundreds of thousands of units. With the current state of the industry’s development, Unitree said, more clear-cut commercial uses for humanoid robots should become manifest within three to five years.
Tesla CEO Elon Musk, meanwhile, has said his goal for the company’s general purpose robot Optimus in 2025 is to manufacture thousands of units before ramping up production the following year.
In the firm’s most recent earnings call, Musk predicted Optimus output will eventually exceed 1 million units per year. He expressed confidence the robot’s production cost will fall below US$20,000 (RM88,559), but added its final price will depend on market demand. Musk had previously projected the price would fall somewhere between US$20,000 and US$30,000 (RM88,559 and RM132,858).
Agility Robotics, another industry leader in the US, has received US$150mil (RM664.29mil) in investment from Amazon and already built a factory for its Digit robot. Its facility, RoboFab, has a mandate to produce tens of thousands of units each year from the state of Oregon.
Above all, staying ahead in development and deployment is the criterion for success in the cutthroat, crowded sector.
The focus is no longer on overcoming isolated technical challenges, said He Liang of Northwestern Polytechnical, but excelling in systems engineering and building an optimal system. Being ahead in some major metrics – not necessarily all – can secure a competitive edge.
“To put it bluntly, if you’re behind in 30 out of 100 technical areas, it’s fine as long as you’re ahead in 70,” He said. “At this point, I don’t believe there’s any challenge we can’t overcome. The real issue now is ‘the fast fish eats the slow fish.’”
The pace of deployment also matters, however, and disparities are evident between US and Chinese companies in production scale, funding reserves and talent.
Tesla, for example, can move rapidly through the research and development stage and penetrate the market quickly, thanks to a powerful computing platform and substantial financial resources. While Tesla could end up selling millions of units annually, individual Chinese companies are likely to produce at much lower rates.
But the sheer volume of companies in China creates a cumulative output that could rival or even exceed the US firm’s potential yield.
“Here’s the catch,” He Liang said. “The more units you sell, the more data you gather, especially in real-world client scenarios. With a robust computing platform, that data helps make the robots smarter.”
Over time, this cycle of iteration could widen the gap between companies that move quickly and those that lag behind, especially as the industry reaches maturity.
Wang Lei, chairman of Shanghai Qingbao Engine Robot – which sold dozens of units in 2024 – said Chinese companies should not attempt to follow Tesla’s path.
“Once Tesla has explored a particular area, we should avoid replicating it,” he said. “It’s like flying in a plane while they’re riding a rocket – the gap keeps widening. So, we must steer clear.”
Unlike Musk’s firm, which has already fully established itself in the field of electric vehicles, Chinese companies are shifting focus to untapped markets.
“If we keep trying to follow Tesla’s path, we’ll always be playing catch-up,” Wang noted, emphasising the importance of carving out distinct niches to stay competitive.
As the tech race between China and the US heats up, and the latter tightens its hold on critical technologies – especially advanced chips destined for China – pressure is mounting on Beijing’s hi-tech ambitions, raising questions in turn about how this competition will shape the future of humanoid robotics.
Former US president Joe Biden announced new regulations last month designed to further restrict the export of artificial intelligence chips, primarily to China, Russia, Iran and North Korea.
These restrictions, released one week before Biden stepped down on January 20, would cap AI-related technology exports to most countries, completely block these exports to the four geopolitical adversaries and allow nearly unlimited access to the components for Washington’s closest allies.
Lu Hancheng, director of the Shenzhen-based Gaogong Robot Industry Research Institute, said the chokehold on advanced chips has not inflicted great pain on Chinese humanoid robot companies – not yet, at least, as basic chips still suffice for basic training tasks – and demand for computing power has not reached its full potential.
“However, when data volumes grow exponentially, the demand for computing power increases significantly, and high-end chips will subsequently be needed to meet those demands,” Lu said. “When the industry’s focus shifts from data to computing power, that’s when the growing pains will truly begin.”
Despite the fierce competition between countries and companies in tech – humanoid robotics included – academic collaboration has largely managed to continue unimpeded, a rare bright spot in a bilateral relationship now mostly characterised by rancor.
“In the exploratory stage, you often see (China-US) efforts to push the boundaries of robotic intelligence, rather than competing in isolation,” said Xu Xuecheng of the Zhejiang robotics centre.
He called attention to the frequency of joint projects between institutions from the two countries, particularly in the field of embodied intelligence and brain design. Top robotics conferences also frequently feature papers co-authored by figures from leading institutions in both countries. – South China Morning Post
Unlock 30% Savings on Ad-Free Access Now!
