On August 19, Unitree listed on the STAR Market.
The issue price is 150.80 yuan per share, corresponding to a market value of about 61 billion yuan;
Opened at 11,000 yuan, with a market value that once reached 444.9 billion yuan;
Eleven trading days later, the share price fell below 55 yuan, and the market value dropped to about 22 billion yuan.
In three numbers, the first one is close to the company itself. The other two reflect market sentiment: first buy up the industrial space in 2030, then spit it back out.
The company isn’t overhyped—the scenarios are.
In 2025, revenue is about 1.7 billion yuan—more than triple year-on-year. Non-GAAP net profit is about 600 million yuan, gross margin is about 60%, and over 5,500 humanoid units have been shipped. Revenue exceeds four-legged models for the first time. They can build, can sell, and can make money—something rarely seen in the embodied AI race.
But the money still mainly comes from universities and demonstrations. The industrial and logistics parts that truly replace labor are still thin. In the first half of 2026, revenue was about 1.15 billion, up about +48% year over year. Growth has switched gears; expenses rise, and profit faces pressure. Wang Xingxing himself places the household scenario into a 2–10 year horizon; the inflection point is the completion rate of unfamiliar tasks at around 80%.
What got cut in half was the opening price, not the issue price.
150.8 yuan is the inquiry price. 1,100 yuan is the auction price at the 科创板 (STAR Market) before the first five days when there’s no daily trading limit on the price movement, with a float of about 7%. In the past two years, it’s common for new stocks to have “no drop on the first day, then start falling the very next day.” The more it spikes, the deeper the subsequent pullback.
Compared with the issue price, Unitree is still up by nearly three times; compared with the intraday high on the first day, it’s close to a near 50% drop from the peak. Both statements are true—because you didn’t buy the same thing in those two comparisons. The institutions’ target prices were far below the open from the very beginning.
Why doesn’t it rise like chips do
After the AI boom, what becomes the bill first is training and inference: GPUs, optical modules, and electricity. Those are the mandatory costs for cloud companies in the current period. What Unitree sells is a machine that still can’t reliably replace labor on its own.
Selling cards sells incremental demand; selling robots is replacing wages—you have to pass the hurdle of cost, reliability, and production-line upgrades. Labs will pay for backflips; factories only pay for ROI repeat purchases. China’s host-body/industrial-base manufacturers have already been churning out shipments and cutting prices—being first in shipments doesn’t mean being first in pricing power. The model can also be bolted on elsewhere, and isn’t monopolized by Unitree.
It didn’t miss the wave—it finished the wave early: a ten-year valuation reached and then issued at 6.1 billion, only for the first day to be marked up again to 40 billion. Computing-power stocks move the stock price along with profit upgrades; Unitree moves the story first, then multiplies it by killing it.
The sector already surged through a round earlier
A-share stocks don’t wait until Unitree is discovered to realize robots. In the round when Tesla’s actuators were priced, Sanyuan, Topa, Double-Ring, and Mingzhi were already priced in. In the round for harmonic drives, servos, and industrial base bodies, Green’s harmonic drives, Inovance, Estun, and Haozhi had also already risen. On the day Unitree listed, the index dropped and components gave back gains—this was good news being cashed out. When the full machine financial statements are laid out, the theme can’t rely only on named trades.
Unitree is both an anchor and a mirror. Anchor: the body/host manufacturer finally has publicly reported shipment and gross margin data to compare against. Mirror: the 400 billion shows how hot the story is; the 200 billion shows how long it still takes before the story turns into profits.
Look back at three things
Can industrial revenue clearly exceed lab income? Will new tasks require retraining again? When shipments move up, will the average selling price and gross margin also fall together?
The first two things improve—so when the high pulls back, it’s the pricing reverting. When all three things play out slowly, the market cap will still go find an even lower anchor. Market-cap management can’t change this clock; it only pulls expectations back from “entering ten thousand factories next year” to verifiable orders and model KPI metrics.
Unitree has already completed the hardest exam: hardware mass production. But the valuation anchor in the secondary market next isn’t determined by video demos—it depends on repeat purchase orders driven by ROI at the factory end.
