Key Points
ใปAccording to Global Times reporting on August 26, 2026, a humanoid robot named Tiangong Ultra won the 100-metre final in 8.64 seconds at the second World Humanoid Robot Games in Beijing, faster on the clock than the 9.58-second human world record. In its 8.86-second semi-final run a day earlier, it could not decelerate at the finish, crashed into padding and caught fire.
ใปChina has redirected the component supply chains and mass-production capacity it built in smartphones, drones and electric vehicles into humanoid robots. According to a Fortune report in June 2026, Chinese vendors accounted for more than 97 percent of global humanoid shipments in the first half of 2026, and Unitree, which listed in Shanghai on August 19, has already surged and slumped.
ใปThe threshold for adoption is not the day a robot moves like a human. It is the day a somewhat clumsy robot works more cheaply than hiring one, and global capital is currently trying to price how far away that day is.
โผ Watch this article as a video
The Sprint That Beat Usain Bolt’s Time and Then Caught Fire
The second World Humanoid Robot Games held the final of the 100 metres in the large-robot class on August 26, 2026, at the National Speed Skating Oval in Beijing. Tiangong Ultra, developed by the Beijing Humanoid Robot Innovation Center, won in 8.64 seconds, according to Global Times reporting on the closing day. That is 0.94 seconds inside the 9.58-second human world record held by Usain Bolt. Timing methods and race conditions are not those of human athletics, so the two numbers are not a like-for-like comparison.
According to Global Times reporting on August 25, 2026, Tiangong Ultra could not slow down after crossing the line, hit the padding placed beyond the course, fell, and threw sparks from its body. Reuters described the barrier as a thick mat rather than a wall, so the exact obstacle differs by outlet.
The progression at this event was steep. According to Reuters and Global Times reporting in August 2026, the best 100-metre mark at the first games in August 2025 was 21.50 seconds, against 9.39 seconds on this year’s opening day, 8.86 seconds in the semi-finals and 8.64 seconds in the final.
| Date | Best 100m mark, large-robot class |
|---|---|
| August 2025 (first games) | 21.50 seconds |
| August 22, 2026 (opening day) | 9.39 seconds |
| August 23 to 24, 2026 (heats) | 9.32 seconds |
| August 25, 2026 (semi-final) | 8.86 seconds |
Separately, Unitree Robotics, a maker of quadruped and humanoid machines, listed on Shanghai’s STAR Market for high-technology companies on August 19, 2026. According to SCMP’s report on the debut, the shares closed the first day at 460 percent above the offer price. The stock then fell back. Reuters reported on August 25, 2026 that it had dropped roughly 45 percent from its peak, spreading concern about a robot bubble.
Related Articles
A State-Backed Sports Meet and a 97 Percent Share of Shipments
The World Humanoid Robot Games is a state-organised competition in which humanoid robots contest athletic events and practical work tasks, staged in Beijing by China Media Group and the Beijing municipal government. It is not a private trade show. According to Xinhua’s coverage of the 2026 edition, 666 teams and 2,056 machines from 16 countries across six continents competed in 51 events, including factory parts sorting and hotel cleaning alongside track and field.
The policy that set the target
Industrial policy sits underneath the spectacle. China’s Ministry of Industry and Information Technology issued guidance on developing humanoid robots in 2023, setting a goal of establishing mass production by 2025 and reaching world-leading capability by 2027.
The output followed quickly. According to a Fortune report in June 2026 citing industry analysts, Chinese manufacturers accounted for more than 97 percent of global humanoid robot shipments in the first half of 2026. The base is small and includes machines shipped for research and demonstration rather than production work, so the share describes an early-stage market rather than a mature one.
Japan’s entry was GMO Internet Group’s humanoid Hitomin, which according to the Japanese robotics outlet Robostart’s August 2026 report finished its 100-metre heat in 13th place out of 74 machines. The same company became Unitree’s authorised distributor in Japan in June 2026, which places Japanese participants closer to the buying and selling side than the building side.
Why components built for EVs fit straight into a robot
China’s humanoid surge is a convergence of existing industries rather than an invention from nothing. Smartphones supplied cameras and chips, drones supplied motors and attitude control, EVs supplied batteries and mass-production equipment, and generative AI and autonomous driving supplied perception and decision-making. A humanoid robot is largely those parts repackaged into one frame.
XPeng, the electric vehicle maker, is the clearest case. The company is targeting the start of mass production of its IRON humanoid by the end of 2026, and according to XPeng’s own announcement on August 24, 2026, its robotics unit raised more than 900 million dollars at a post-money valuation above 6.3 billion dollars. On the same day’s earnings call, chief executive He Xiaopeng said the hardware gross margin on IRON would be reliably higher than that of the car business.
For Chinese manufacturers worn down by the EV price war, robots look like the next high-margin category. A substantial part of the component network and production equipment carries over, which lowers the barrier to entry relative to other countries. Fine manipulation and the intelligence to complete a task do not carry over, and remain the open problems.
This is not only a Chinese race. Tesla is developing its Optimus humanoid in the United States and the startup Figure has attracted large funding rounds. The shape of the competition is Chinese lead in components and volume against American strength in the reasoning layer.
Japan dreamed of the humanoid first
Popular culture in Japan built the appetite for humanoid machines before anyone else. Astro Boy and Doraemon fixed the image of a human-shaped machine working alongside people, and that image became the background of Japanese robotics research.
The hardware followed. Waseda University built WABOT-1 in 1973, generally described as the world’s first full-scale humanoid intelligent robot, and Honda’s ASIMO demonstrated bipedal walking to a global audience in 2000. SoftBank’s Pepper became a familiar sight in Japanese shops for a period.
None of them found a route to volume. Honda announced the end of ASIMO development in 2018 and retired the robot from demonstrations in 2022. Pepper production was reported to have stopped in 2020.
Japan remains a leader in industrial robots, where Fanuc and Yaskawa sit among the largest firms in the world. In the humanoid form specifically, the country stopped just short of the stage where machines are built in quantity at a price people will pay. That is precisely the stage China is now attempting.
The Record Is Not the Number That Decides This
Why 8.64 seconds does not mean the robot can work
A sprint time and a shift on a factory floor measure different things. The machine that set the record could not stop and caught fire, which is the demonstration. Running fast is a question of peak output. Work requires eight hours of it, without breakdown, at repeatable precision.
Business interest has sharpened anyway, because the cost arithmetic has started to move. According to a JPMorgan estimate reported by MarketWatch in 2026, the operating cost of a humanoid robot in United States manufacturing and warehouse work could fall below 10 dollars an hour, against roughly 30 dollars an hour for a human warehouse worker. The bank has not published the assumptions or the cost breakdown behind the figure, so it is an indicative number rather than a verified one.
The comparison also carries a hidden condition. An hourly rate only means something if both sides do the same amount of work, and trials of current machines have reported throughput at around half the human rate. The real threshold is the product of capability, uptime and price. A robot at 70 percent of human capability that runs almost around the clock at half the annual cost will find buyers. A robot that runs 100 metres in 8 seconds but stops and breaks will not.
Why humanoid robots do not need to beat purpose-built machines
For a fixed task on a production line, a wheeled base with a robot arm is still far more rational. That is not the ground humanoids are competing for.
The claim is about the world humans already built. Climbing stairs, opening doors, taking items off a shelf, operating an existing machine tool. Buildings and tools are dimensioned around the human body, so a machine shaped like a human may let a facility avoid the cost of rebuilding itself.
For a buyer this is a capital expenditure question rather than a robotics question. A purpose-built machine performs better but requires the line to be redesigned around it. A humanoid performs worse per unit but may fit into the site as it stands. The logic of the human form lives on the installation-cost side of the ledger, not the performance side. Safety systems and supervision cost money either way.
Is the market buying this, or is the state?
Unitree’s share price swing is one measure of the distance between expectation and current reality in this industry. Its 460 percent first-day gain was roughly double the 226 percent average for Chinese listings over the previous three years, according to Reuters. Its prospectus, cited in the same reporting, shows adjusted net profit of about 40 million yuan in the most recent quarter, down 53 percent year on year. The story and the income statement point in opposite directions.
| Unitree metric | Figure |
|---|---|
| IPO price, August 6, 2026 | 150.80 yuan, valuing the company near 9 billion dollars |
| First-day close, August 19, 2026 | 845 yuan, up 460 percent |
| Average first-day gain, Chinese IPOs, prior three years | 226 percent |
| Price on August 26, 2026 | Around 591 yuan |
| Adjusted net profit, latest quarter | About 40 million yuan, down 53 percent year on year |
The composition of demand deserves the same scrutiny. According to a Morgan Stanley tally reported by Fortune in June 2026, orders from state-owned enterprises across China’s robotics industry exceeded 2 billion yuan in 2025. SCMP has reported more than 40 announced state-backed centres for collecting robot training data. A large share of early demand is being carried by policy.
Unitree’s own customer base is concentrated in research institutions. Founder and chief executive Wang Xingxing told CNBC in August 2026 that the point at which robots become genuinely useful is between two and ten years away. Before any argument about human wages, the record of machines being chosen on price in an open market is thin.
This may be the EV success repeating, or it may be the overinvestment repeating. In EVs the loop ran from state support to mass entry, mass production, price war, consolidation and export, and produced a winner in BYD. The expectation that the same circuit will turn again, and the fear that state demand is simply pulling future orders forward, both live inside the move from 845 yuan to 591 yuan.
What Japan still holds inside the machine
Japan stopped at the finished humanoid, but it did not stop at the parts. The critical joint component is the precision reduction gear, which converts a motor’s fast rotation into slow, high-torque movement. Nabtesco states on its own site that it holds about 60 percent of the world market for the medium and large industrial robot segment.
In small strain wave gearing, Harmonic Drive Systems has been reported by Toyo Keizai to hold 80 to 90 percent of the global market.
A humanoid amplifies this. Arms and legs mean many more joints, and industry estimates circulating in 2026 put the number of reduction gears per humanoid at six to ten times that of a conventional industrial robot. If Japanese parts continue to be selected, Chinese volume converts directly into Japanese component demand. In an era of physical AI, meaning AI systems that act on the physical world rather than only producing text and images, whether Japan keeps hold of the joint is the competition that matters on its side. Japan’s own workplaces, where labour shortage is more acute than anywhere else, are also a candidate site for trials and training data.
The advantage is not permanent. China is pursuing domestic production of reduction gears as policy, and Japan has already lived through the version of this story where a components-based strategy was eroded by localisation in the EV supply chain. The opening is real, and it is not the kind that survives being left alone.
Japanese Reactions to China’s Robot Sprint Record
Coverage in Japan travelled mostly through news accounts rather than commentary, and the dominant register was plain surprise rather than alarm. What makes the Japanese response distinctive is that Japan competed, and the Japanese team published its own gap. These are posts on X and do not represent Japanese public opinion as a whole.
1500m่ตฐใใฒใจใฟใใฏ็กไบๅฎ่ตฐใงใใพใใ๏ผ
— GMO AI๏ผใญใใใฃใฏในๅไบ (@GMO_AIR) August 24, 2026
็ดๅใฎ่ชฟๆดใฎ็ฒๆใใฃใฆใ็ด4ๅจใๅฎๅฎใใฆ่ตฐใๅใใพใใ๐
่จ้ฒใฏ7ๅ43็ง89ใงใ35ไฝไธญ 7ไฝ ใงใใ๏ผ
ๆฌกใฎๅบ่ตฐใฏ8/25(็ซ)ใฎ100mใงใ๏ผ
ๅผใ็ถใๅฟๆดใใใใใ้กใใใพใ๐ฃ#ไธ็ไบบๅใญใใใ้ๅไผ pic.twitter.com/LSFNjUVlwg
Our humanoid Hitomin finished the 1500 metres in 7 minutes 43.89 seconds, placing 7th out of 35 machines. Thank you for the support.
This post from GMO AI & Robotics, the Japanese entrant, drew more engagement than any news account’s coverage of the record itself, with over 8,000 likes. The most widely shared Japanese message about the games was a Japanese team reporting that it had finished.
100mไบ้ธใใฒใจใฟใๅฎ่ตฐใใพใใ๐
— GMO AI๏ผใญใใใฃใฏในๅไบ (@GMO_AIR) August 25, 2026
่จ้ฒใฏ24.3็งใงใ74ไฝไธญ 13ไฝ ใงใใใ
ๆใใใไบ้ธ็ช็ ดใฏใงใใพใใใงใใใใๅๅ ใใชใใจใใใใชใใใจใๅญฆในใพใใใ
ๅคงไผๆ้ไธญใใใใใใฎๅฟๆดใ้ ใใใใใจใใใใใพใใ๏ผ๏ผ
ใฒใจใฟใใฎๆฌกใฎๆๆฆใใๆฅฝใใฟใซ๐ค#ไธ็ไบบๅใญใใใ้ๅไผ pic.twitter.com/fsLHPhErmE
Hitomin ran the 100-metre heat in 24.3 seconds, 13th out of 74 machines. We did not advance, but we learned things you can only learn by taking part.
The same account published the number that sits next to Tiangong’s winning time. Disclosing a 24.3-second heat without softening it is a particular kind of composure, and it gave Japanese readers the size of the gap in a form no news report supplied.
ใไบบๅใญใ100mใใซใ่ถ ใ ็ดๅพ็ไธใhttps://t.co/rMgpRZy8lG
— Yahoo!ใใฅใผใน (@YahooNewsTopics) August 22, 2026
Humanoid robot beats Bolt over 100m, catches fire immediately afterwards
Yahoo News Japan put the record and the fire in a single headline, and this framing is where much of the Japanese audience first encountered the story. The laughter in the replies to this kind of post carried some relief in it.
ไธญๅฝใง่กใใใฆใใใญใใใๅคง้ๅไผใใใฎ400m่ตฐใงใญใใ่ฆใใใใฎๅฅๅฆใชใใฉใผใ ใฏใๆฎ้ใซไบบ้ใจๅใใใใซๅทฆๅณใฎ่ ใๆฏใฃใฆ่ตฐใใจ่ฉ้ข็ฏใฎใขใผใฟใผใ้็ฑใใฆใดใผใซใพใงใใใชใใใจใใใ่งฃๆฑบ็ญใจใใฆAIใซใใ็นฐใ่ฟใใฎใทใใฅใฌใผใทใงใณใซใใ็ใฟๅบใใใใใฎใ ใใใ ใ่ ใไธใใฆไธไฝใพใใใจโฆ
— MMGP (@motto_motogp) August 25, 2026
Swinging the arms like a human overheats the shoulder motors before the finish. The strange form you are seeing is what repeated AI simulation arrived at as the optimum.
A motorsport account supplied the most-shared technical explanation of the robots’ unnatural running posture. The reading that stuck in Japanese timelines was that the machines are not imitating humans badly, they are solving a different problem correctly.
Read together, these posts describe a country watching from the stands at an event it once would have hosted, without much bitterness about it. Threat framing was largely absent from the widely shared posts. The Unitree listing, by contrast, circulated mainly among Japanese investment accounts and did not reach the general audience.
Performance Curves and Profit Curves Do Not Move Together
Performance records for a new technology improve startlingly fast. The point at which the technology pays for itself moves on a separate curve. Autonomous driving, solar power and electric vehicles all covered long ground between a successful demonstration and a viable business. A 100-metre time falling from 21.50 seconds to 8.64 seconds in a year and the arrival of a robot that works more cheaply than a person are not the same statement.
China is spending money and manufacturing capacity to close that distance. The number worth watching next is not a race result. It is the first repeat order placed by a company that concluded, without subsidy, that the machine was cheaper than the hire.
The country that first imagined the humanoid now supplies the parts turning inside the joints of the country mass-producing it. Whether that is the residue of an early lead or the entrance to the next one is the question sitting behind 8.64 seconds.
Frequently Asked Questions
How fast did the Chinese robot run the 100 metres?
Tiangong Ultra won the large-robot 100-metre final in 8.64 seconds on August 26, 2026, at the second World Humanoid Robot Games in Beijing, according to Global Times reporting. That is faster on the clock than Usain Bolt’s 9.58-second human world record, though timing methods and race conditions differ from human athletics. In its 8.86-second semi-final run a day earlier, the robot could not decelerate at the line, hit the padding beyond the course and caught fire.
Are humanoid robots cheaper than human workers yet?
No. According to a JPMorgan estimate reported by MarketWatch in 2026, humanoid operating costs could eventually fall below 10 dollars an hour against roughly 30 dollars for a US warehouse worker, but the bank has not published its assumptions. Trials of current machines report throughput at around half the human rate, and Unitree chief executive Wang Xingxing told CNBC in August 2026 that genuine usefulness is two to ten years away.
Who is actually buying humanoid robots in China?
Largely the state and research institutions rather than commercial operators buying on price. According to a Morgan Stanley tally reported by Fortune in June 2026, orders from state-owned enterprises across China’s robotics industry exceeded 2 billion yuan in 2025, and SCMP has reported more than 40 announced state-backed robot training data centres. Unitree’s own customer base is concentrated in research bodies.
How did Japanese social media react to the robot games?
Mostly with plain surprise rather than threat framing. The single most-shared Japanese post about the games came from GMO AI & Robotics, the Japanese entrant, reporting that its humanoid Hitomin finished the 1500 metres in 7 minutes 43.89 seconds, and the same account disclosed its 24.3-second, 13th-of-74 result in the 100-metre heat. A Yahoo News Japan headline pairing the record with the robot catching fire also circulated widely.
Sekahan on YouTube
We publish video summaries of articles like this one, along with short clips built around Japanese reactions.
Reference Links
- Chinese robot Tiangong clocks sub-9-second 100 metres in Beijing๏ฝAl Jazeera (Reuters)
- Chinese humanoid robot Tiangong breaks 100m record again with 8.86-second sprint๏ฝGlobal Times
- Tiangong Ultra resets 100m mark at 8.64s as humanoid robot games close in Beijing๏ฝGlobal Times
- China robot maker Unitree’s post-listing slump sparks bubble fears๏ฝReuters (Investing.com)
- Chinese humanoid robot maker Unitree prices IPO at $9 billion valuation๏ฝCNBC
- Unitree Robotics surges 629% to US$66 billion valuation in Shanghai share debut๏ฝSCMP
- China builds 85% of the world’s humanoid robots๏ฝFortune
- XPENG Robotics Raises US$900M for Humanoid Robots๏ฝXPeng
- JPMorgan predicts booming humanoid robot demand๏ฝMarketWatch (Yahoo Finance)
- 2026 World Humanoid Robot Games coverage๏ฝXinhua
- About Nabtesco precision reduction gears๏ฝNabtesco
