.png)
Humanoids have become extremely good at constrained, repeatable physical movement, but are clumsy at anything requiring precision. That distinction breaks the current euphoria.
Progress on mobility is nonetheless striking. The winning 100 metres at the games' first edition in August 2025 took 21.50 seconds. This year the same machine, Tiangong Ultra, opened at 9.39 seconds, then cut its own record to 8.86 in the semi-finals, well inside Usain Bolt's 9.58-second mark from 2009. The 400 metres fell from roughly 1:28 to 38.15 seconds, the long jump from 1.25 metres to more than 7 meters. At Beijing's half-marathon in April, over 300 robots ran alongside 12,000 humans and the fastest humanoid finished ahead of the human world record. A year earlier, only one of 21 entrants completed the course at all.
Markets read that trajectory and extrapolated from it. Unitree priced its IPO on 6 August 2026, at about $22.35 a share, valuing the company near $9bn , or roughly 38 times its 2025 revenue of about $252mn, which had quadrupled YOY on 60% gross margins and around 5,500 units shipped. On the day of the World Robot Conference, the stock surged as much as 629% intraday before closing up roughly 460%, valuing Unitree at approximately $47bn, closer to 200 times that same revenue figure. The same enthusiasm extended to other robotics businesses. XPeng's robotics unit, valued at $6.3bn, raised over $900mn in what it described as the largest single private round in China's embodied AI sector, guiding to hardware gross margin above 50% against 12.1% for its vehicle business recorded in Q2.
Sources: ShauryaMalwa & CoinDesk
The same event produced footage that never made the highlight clips, when the record-setting machine went down in a shower of sparks and caught fire, requiring officials to extinguish it. Another was carried off the 400-metre track on a stretcher. A high-jumper cleared the human world record but could not slow its own momentum and skidded to a halt, inches from the photographers. One robot went into convulsions mid-event while its handlers hunted for the remote.
A sprint is a generous test. The surface is flat, the distance fixed, the start known, and nothing unexpected enters the lane. Every variable has been removed in advance by the organisers. The harder questions are whether robots can recognise an unexpected obstacle and recover, which is not what the racetrack examined.
The games did assess those capabilities elsewhere, and the results travelled far less. Across 51 events involving 2,056 robots from 16 countries, roughly a third of total matches were scenario-based rather than choreographed, and more than 40% required full autonomy: hotel reception, emergency response, cable connection, material handling, folding laundry, pulling an espresso inside a 30-second window, repeatedly. The same machines that outran Usain Bolt struggled to lift beans with tweezers or open a jar. In fact, China's official press acknowledged that housekeeping and elderly-care robots as still in pilot and validation rather than ready to use. Speed makes a headline. Reliability however is what a factory or household would actually pay for.
Source: Reuters
Three factors will determine whether humanoid robotics becomes a real industry rather than an expensive demonstration: who controls the training data these machines need, who can fund years of pre-profit losses, and who retains access to which markets. All three came into sharper focus this summer.
The US moved first on national-security grounds rather than commercial ones. On 28 July 2026, the US Federal Communications Commission (FCC) added foreign-made humanoid and quadruped robots to its import Covered List, citing supply-chain, cybersecurity and data-collection risks. Items on this list are prohibited from importation, marketing, or sale in the US, including Huawei’s telecommunications equipment or foreign-produced drones. The new rule blocks new humanoid models from FCC authorisation and effectively requires future US-sold models to meet a threshold of more than 65% domestic components. China's Foreign Ministry called the move protectionism and vowed countermeasures. The timing added friction ahead of a Trump–Xi summit reportedly planned for September 2026, on top of existing chip-export disputes.
The ban comes as China already commands an estimated 85% share of the global humanoid robot market. Of an estimated 15,000 humanoid robots shipped worldwide in 2025, Unitree and AGIBOT each shipped more than 5,000, while Tesla and Figure AI each shipped only a few hundred units or fewer, according to Omdia.
Source: CNN
Valuations are running well ahead of disclosed revenue on both sides of the Pacific. Industry-wide disclosed humanoid revenue was under $1bn in 2025 against roughly 13,000–15,000 units shipped, yet bank forecasts diverge wildly on the industry’s eventual size. Goldman Sachs projects $38bn by 2035, Morgan Stanley projects a $5tn ecosystem by 2050, built on an estimated 13 million humanoids in service by 2035 and more than a billion by 2050, while Citigroup goes as high as $7tn by 2050. Morgan Stanley also expects unit prices to fall from roughly $200,000 today to around $50,000 in wealthier markets and as low as $15,000 where Chinese supply chains dominate. Against that backdrop, Unitree's roughly 200-times-revenue valuation, and XPeng's robotics arm with Alibaba and Tencent as strategic investors, reflect expectations of substantial long-term growth rather than near-term cash flow.
Robotics is emerging as one of the growth engines Beijing needs, even as the broader Chinese economy shows signs of slowing. China’s National Bureau of Statistics released July 2026 data on August 17. While electronics equipment output rose more than 19% and industrial robot manufacturing grew by over 30%, industrial production growth eased to 4.5% year on year, missing estimates and marking its first slowdown in three months.
The data points to an economy increasingly splitting into two very different halves: one struggling, the other thriving. Robotics sits firmly in the latter, giving Beijing every reason to keep supporting the sector’s expansion even before the underlying business case is fully proven.
The near-term effect looks like two separate markets developing in parallel. Chinese manufacturers, still able to undercut Western competitors sharply on price, will keep scaling at home and push harder into markets the ban doesn't touch—Europe in particular. That dynamic is already compressing prices and margins across the industry, not just for the Chinese firms driving it. US and allied manufacturers end up with a smaller but protected home market. What the ban doesn't do is close the gap in shipment volumes, scale, or price that already favours Chinese suppliers everywhere else, and it adds cost for any US integrator still relying on Chinese-made components.