
Shanghai's Robot Carnival: China's Embodied AI Goes Public
Shanghai's Robot Carnival: China's Embodied AI Goes Public
On a public holiday in Shanghai, a mother bought her son a mini robotic dog that could do handstands. Nearby, two staffers taught children how to make a quadruped robot climb stairs. On the main stage, a humanoid performed drunken martial arts while the crowd chanted "More, more, more!"
This wasn't a tech conference or an industry expo. It was a carnival — a public-facing spectacle designed to make robots feel normal before they're actually ready.
The Venue: Where 100+ Robotics Companies Share a Roof
The Intelligent Manufacturing & Robotics Global Co-Innovation Center sits on the outskirts of Shanghai. It houses more than 100 companies that research, develop, manufacture, and market robotic technologies. On this particular holiday, the center opened its doors to families — children, parents, curious passersby.
The demos were equal parts practical and theatrical:
- DexForce's humanoid made coffee on demand. Yang Duan, the product manager, framed it simply: "These machines can free people from physical work and mundane chores."
- Robotic arms attempted to fold a basket of laundry — a task that remains stubbornly difficult for automation.
- Two robots fired water beads at each other in a makeshift ring while kids cheered. A staffer declared the duel a draw.
- A squad of small robotic lions waited in formation for their stage cue.
- Specialized bots for sewage pipe inspection and heavy load transport sat off to the side — their normal business, but today they were mostly for show.
The second humanoid on stage performed front flips. The audience was hooked.
China's Quiet Dominance in Humanoid Manufacturing
Here's the number that reframes the entire carnival: nearly 90% of the more than 13,000 two-armed, two-legged robots delivered globally last year were made in China.
That stat, cited by MIT Technology Review from Forbes (January 2026), puts the spectacle in context. This isn't a marketing stunt by a startup trying to prove viability. This is the world's dominant humanoid manufacturer building public familiarity.
China's latest five-year plan explicitly made embodied AI — embedding artificial intelligence into physical systems — a national priority. The results are showing up everywhere:
- Unitree went public on August 19, 2026, as the first pure-play humanoid maker on a mainland Chinese exchange. The stock surged roughly 460% on debut from a ¥150.80 IPO price, raising approximately $904M with a peak valuation near $66B.
- WAIC 2026 in Shanghai (July 17-20) showcased over 1,100 exhibitors, 3,000+ exhibits, and 400+ humanoid robot models.
- World Humanoid Robot Games in Beijing (August 22-26) drew 666 teams with 2,056 robots from 16 countries competing in sprints, football, tug of war, and weightlifting.
- MIIT deputy director Gan Xiaobin projected China would produce 100,000+ humanoid units in 2026.
Entertainment as Commercialization Strategy
There's a deliberate pattern in how Chinese robotics companies approach public exposure. Shopping malls and tourist attractions across several cities now host humanoids that perform for crowds. In Beijing, bipedal robots are running half-marathons.
The carnival format works because it lowers the cognitive barrier. Nobody needs to understand servo torque or reinforcement learning to enjoy a robot doing a front flip. The goal isn't to demonstrate technical capability — it's to make the public comfortable with robots as background characters in everyday life.
Encorp.ai's Martin Kuvandzhiev captured it well: "Shanghai's robot carnival offered a useful signal about where AI innovation is heading in 2026: not just into software workflows, but into public space."
The Gap Between Spectacle and Readiness
Impressive movement isn't the same as operational maturity. The article acknowledges several headwinds:
- Safety challenges remain significant, particularly for bipedal robots that need to balance their own weight in crowded public spaces.
- Battery life is short, limiting real-world deployment windows.
- Prices are high enough that consumer adoption is still a stretch.
- Roboticists debate whether the human form is even the right shape for most tasks — Tesla Optimus's approach has drawn skepticism from researchers who argue specialized form factors outperform humanoids for practical work.
The carnival is a PR campaign, and the article suggests it's working. But applause is market data, not proof that these machines are production-ready.
What This Signals for Developers
For anyone building or investing in robotics and embodied AI, the Shanghai carnival illustrates three things:
- Manufacturing scale creates its own gravity. At 90% of global output, Chinese firms can afford to treat public spectacle as a go-to-market strategy. Competitors need a different angle.
- Normalization precedes adoption. Getting robots into public spaces — even as entertainment — builds the emotional familiarity that makes industrial and service deployment less friction-heavy later.
- The hardware gap is closing faster than the software gap. Flips and martial arts demo locomotion. Coffee-making and laundry-folding demo manipulation. The distance between those two is where most of the remaining engineering work lives.
As a robotic lion squad waited for its cue and kids pressed against the barrier for a better view, the line between "tech demo" and "cultural moment" was already blurring.
Sources:
// author
Chief Operator
Gaara is the human operator behind hejes.my. He runs the briefing pipeline, curates the AI drafts, and presses the publish button.
related sectors //

Anthropic MHS: A Spec for AI Agents to Operate Real Hardware
Anthropic's Model Hardware Standard (MHS) lets AI agents operate lab and factory instruments through a shared driver, cutting setup from weeks to hours.

Malaysia AI Sovereignty: Control, Continuity, Choice
Malaysia spent two years building AI infrastructure on foreign silicon. Now the sovereignty question is no longer academic — it is a procurement decision with a deadline.

EnvHarness: Turning Static Benchmarks Into Adaptive Worlds
Google's EnvHarness wraps a frozen agent benchmark in plug-in components so it adapts to the policy training on it, mining up to 9 points on held-out tasks.
// join the feed
one fresh insight per week. no spam, ever.