They don’t need a pre-race carb-loading meal or a playlist to settle pre-start jitters — just a full charge. Of the runners set to line up for the Beijing Yizhuang Half Marathon on April 19, 2026, roughly 300 will be robots, about five times the number that entered the race a year ago.
In a contest that trades sweat for circuitry, these humanoid competitors will pound the pavement alongside human runners, turning the course into a proving ground for the future of machine mobility. At the inaugural event in 2025, the robots lagged conspicuously behind, their movements tentative and inefficient over the grueling distance. This year, engineers say, the gap is narrowing — and fast — as machines inch closer to something resembling competitive form.
In the 2025 race, the winning human runner finished the 21-kilometer race in 1:02, whereas the winning humanoid, Tiangong Ultra, finished in 2:40. It fell down once, and had to receive three battery changes. In a post-race interview with Tiangong Ultra, made possible through ChatGPT, it stated that victory tasted like lithium.
The entrants represent more than 100 company and university teams, and like their human counterparts, they vary widely in form and function. Some have simple clamps for hands; others feature heads that appear designed more for utility than familiarity. Most move with a halting, mechanical gait, though a handful display strides that are unexpectedly fluid. The race, as much as anything, is a public snapshot of how far robotics has advanced in just a year.
That progress was already on display earlier this year, when kung fu–capable humanoids appeared at Beijing’s 2026 Spring Festival Gala. Where last year’s machines struggled with even basic coordination, newer models from Unitree executed flips, kicks and acrobatic sequences with a precision that hinted at rapid underlying improvements in balance, control and responsiveness.
The rules for the race are straightforward: the robots must operate as standalone systems. Competitors are divided into two categories — autonomous and remote-controlled — with autonomous machines making up about 40 percent of the field. A practice session held April 11 gave teams an opportunity to test their systems under real-world conditions and troubleshoot performance issues before race day.
“On this unfamiliar road section, including the weather, the environment, and the visibility, these factors do have a certain impact on us,” said Unitree Robotics engineer Chu Yang. “Based on the results of this test race, we will make some corresponding adjustments after going back.”
Liang Liang, deputy secretary-general of the Chinese Institute of Electronics, said the results of the half marathon will serve as a broader validation of the country’s sustained investment in robotics — and of the reliability of the machines themselves.
“In terms of performance, the format and goal settings of this year’s marathon will drive many teams to make significant innovations in software development, algorithm development, and gait research,” Liang said.
As in last year’s race, performance varies widely. Some humanoids veer off course or topple into barriers; others simply power down mid-run. But at the other end of the spectrum, a few demonstrate steady pacing, smoother strides and coordinated arm movement. Sizes range from toddler-scale machines to those approaching the dimensions of an adult human.
If the race offers a glimpse of the present, forecasts from analysts suggest a far more expansive future. The market for humanoid robots — still in its infancy — could grow into a multi-trillion-dollar industry over the next few decades, reshaping labor, manufacturing and even domestic life.
Morgan Stanley Research estimates that the humanoids market could reach $5 trillion by 2050, including the broader ecosystem of supply chains, maintenance and support services. That growth would place it among the largest industrial sectors in the world, potentially rivaling or exceeding the scale of the global automotive industry.
“Adoption should be relatively slow until the mid-2030s, accelerating in the late 2030s and 2040s,” said Adam Jonas, Morgan Stanley’s head of global autos and shared mobility research.
By midcentury, the firm projects that more than 1 billion humanoid robots could be in use worldwide. The overwhelming majority — roughly 90 percent — would be deployed in industrial and commercial settings, handling repetitive, structured tasks in factories, warehouses and service environments. Household adoption, by contrast, is expected to lag significantly.
“The forecast for household usage is much more conservative, with only 80 million humanoids in homes by 2050,” Jonas said. “We are not going to see a robot in every home overnight.”
Technological and economic barriers remain substantial. Building a general-purpose humanoid capable of performing a wide array of household tasks requires advances not only in hardware, but also in artificial intelligence models that can interpret and respond to complex, unstructured environments. Analysts estimate that meaningful progress on that front could take another decade.
“Once we get to that stage, humanoid volume and penetration should pick up quickly,” Jonas said.
Cost is another limiting factor. Morgan Stanley estimates that a single humanoid robot cost about $200,000 in 2024 in high-income markets. Prices are expected to decline as manufacturing scales and technology improves, potentially falling to $150,000 by 2028 and to around $50,000 by 2050. In lower-income markets, where manufacturers may benefit from more cost-efficient supply chains, prices could drop as low as $15,000.
Even then, adoption is likely to be uneven. In the United States, Morgan Stanley projects that about 10 percent of households could own a humanoid robot by 2050 — roughly 15 million units — with ownership concentrated among higher-income households.
“We recognize that, hypothetically, the average household could have more than one unit, creating a fleet of humanoid butlers,” Jonas said. “However, in our forecasts, we assume one humanoid per household at this time.”
Much of the momentum in the sector is currently centered in China, where strong government support and a robust manufacturing ecosystem have helped accelerate development. Analysts point to coordinated national investment in what is often described as “embodied AI” — systems that combine physical robotics with advanced artificial intelligence — as a key driver of the country’s lead.
“It is becoming apparent that national support for ’embodied AI’ may be far greater in China than in any other nation, driving continued innovation and capital formation,” said Sheng Zhong, Morgan Stanley’s head of industrials research. “In our opinion, China’s lead in AI-robotics may need to widen before rivals, including the U.S., pay closer attention.”
China’s advantage is also rooted in its supply chain. Many of the core components required for humanoid robots — from motors and reducers to batteries and precision fasteners — are produced at scale within the country or elsewhere in Asia. That dominance allows Chinese firms to iterate more quickly and at lower cost, particularly as the technology moves toward mass production.
