On the afternoon of August 17, a 30-second video exploded across Chinese tech circles.

A headless, armless bipedal robot launched vertically from the ground, clearing a 2-meter height marker with room to spare, then landed solid as a rock. Cut to the same robot sprinting down a track, the speedometer hitting 12.66 meters per second — roughly 45.6 km/h.

The publisher: Unitree Robotics. Two days later, on August 19, the Hangzhou-based robotics company formally lists on Shanghai's STAR Market under ticker 688836. The offer price is 150.80 RMB per share, giving the company a market capitalization of approximately 60.99 billion RMB ($8.5 billion).

A capital feast on one side, a muscle-flexing tech demo on the other. Two mega-events packed into the same 48-hour window. It's hard not to ask: coincidence, or masterfully orchestrated IPO marketing?

The answer is probably both. But if all you see is 'marketing,' you're missing what actually matters — the physical capabilities of humanoid robots are approaching and surpassing human limits far faster than anyone expected.

What Does a 2m Jump and 12.66 m/s Actually Mean?

Let's unpack the numbers properly.

A 2-meter standing high jump. For context, the Guinness World Record for human standing high jump is roughly 1.70 meters, set by Christopher Spell in 2021. Even using a looser 1.8-meter benchmark, Unitree's 'Superhuman' robot clears the best human performance by at least 20 centimeters.

The more revealing reference frame is leg length. Superhuman's legs are only 0.85 meters long — significantly shorter than an average adult's. Jumping 2 meters on legs that short means the leg joints have to output absolutely terrifying torque at takeoff. To put it in perspective: imagine a 5'3" person dunking on a standard basketball hoop, and landing on their feet.

Now speed. A top running speed of 12.66 m/s exceeds Usain Bolt's peak instantaneous velocity of roughly 12.42 m/s from his 2009 Berlin world record. That's right — a humanoid robot's sprint peak has already crossed the threshold of the fastest human who ever lived.

At that pace, a theoretical 100-meter dash would take about 7.9 seconds. To be clear, that's a peak-speed extrapolation, not an actual 100-meter result. But the number itself is staggering regardless.

What's really worth watching is the pace of iteration. In August 2025, Unitree's H1 humanoid ran at just 3.3 m/s — roughly a casual jog. By April 2026, H1 broke 10 m/s. Four months later, Superhuman pushed it to 12.66 m/s.

Speed has nearly quadrupled in one year.

This kind of iteration speed in hardware is the extreme end of 'Shenzhen velocity.' It's not just algorithm tuning — motor power density, reducer transmission efficiency, battery discharge rate, structural component strength — every single link in the chain has to level up simultaneously for overall performance to jump like this.

This Isn't Just Marketing

Let's be honest — the timing is suspicious. IPO pricing results announced August 14. 'Beyond human limits' video drops August 17. Stock starts trading August 19. The World Robot Conference opens right after.

But dismissing the whole thing as 'IPO theater' is too easy.

First, Superhuman isn't a complete humanoid robot. It has no head, no hands, the upper body is basically empty. It's fundamentally a leg locomotion testbed. Unitree has been clear about this: it's a technology verification prototype, built from scratch in just over three months, with plenty of room for improvement. No production plans, no pricing announced.

That distinction matters. If a company showed you a half-built robot and claimed it 'fully surpasses humans,' that would be a scam. But Unitree's positioning is precise — this is a prototype built solely to test the limits of running and jumping, to find the ceiling of their powertrain and control algorithms.

Second, the numbers are technically plausible. Unitree's approach is high cadence + short legs, trading stride length for step frequency. This route demands extreme real-time control, but when it works, you get better dynamic response and agility. The 0.85m leg length means lower rotational inertia and faster joint response — that's not cheating, that's engineering.

Third, and most importantly: the technology from extreme prototypes trickles down.

The joint motors and balance algorithms validated in Superhuman will show up in production models tomorrow. Unitree shipped over 5,500 humanoid robots in 2025, more than any other company globally. When a company shipping thousands of units a year can push极限 performance up a whole tier in three months, it means they've built a reusable engineering methodology at the foundation level.

The Bull and Bear Case for a $6.1B Valuation

Back to the capital markets side.

Unitree's IPO set several records: ~9.78 million online subscriber accounts — the most in STAR Market history. A final lottery win rate of just 0.018% — the lowest ever. One lot (500 shares) costs 75,400 RMB, which is on the high side for this year's new listings.

The offer P/E ratio is 219x. The industry average for general equipment manufacturing is 38.56x. At nearly 6x the sector multiple — is it worth it?

The bull case is straightforward: humanoid robots are the next trillion-dollar赛道, and Unitree has already taken the global lead in shipment volume. 2025 revenue of 1.7B RMB, up 335% YoY. Adjusted net profit of 600M RMB, up 674%. Growth like that arguably justifies a premium multiple.

The bear case is equally valid: today's humanoid robots are more like 'feature phone era' devices — they can walk, run, carry things, but they're still far from genuinely 'intelligent.' Without foundation model integration, a robot is essentially just advanced automation equipment. How much of the 61B RMB market cap is already priced in?

The strategic investor list is interesting though. DeepSeek took 933,400 shares worth about 141M RMB, with a 36-month lockup — the longest of any strategic investor. Tencent is also in the lineup through its Qishan Investment vehicle. State-owned enterprises like PetroChina Kunlun Capital, China Southern Grid, and Tianyi Capital round out a heavyweight industrial roster.

DeepSeek's long-term bet is particularly telling. If large models are the robot's brain, then robots are the large model's body. The marriage of brain and body is the ultimate form of embodied intelligence.

What's the Real Inflection Point?

After all these tech specs and valuation debates, the real question is: when does embodied AI actually hit its inflection point?

We used to think the bar was 'can robots enter the home?' That might be too distant a horizon. Entering the home means solving safety, cost, utility, privacy — each one a massive problem on its own.

The more realistic inflection point is simpler: robots start generating real economic value in one vertical scenario after another.

And that拐点 is already happening.

At Beijing's Embodied Intelligence Innovation Park, Wujie Dynamics pulled in nearly $100M in global orders in the first half of this year. Unitree's quadruped robots are already widely deployed in industrial inspection and park security. Not to mention the millions of robotic arms in factories worldwide — those aren't humanoid, but they're still commercial embodiments of embodied intelligence.

Superhuman's significance isn't that it can outrun Bolt. It's that it proves something: the ceiling for humanoid robot locomotion is far higher than we thought. And it's rising much faster than we expected.

As robots' bodies get stronger, the moment the 'brain' (foundation models) truly catches up — the chemical reaction could be exponential. Right now there's still a gulf between the two: foundation models can't yet control robot motion in real time, and robots don't have genuine autonomous decision-making.

But that gulf is narrowing by the day.

Today we're amazed by a robot that outruns Bolt. Five years from now, we might look back and see this as just a footnote in the embodied intelligence era. The real show hasn't even started.

See you tomorrow.