A robot that can thread a wiring loom faster than a tired line worker is no longer a lab curiosity on one Shanghai factory floor.
The company behind it, TARS Robotics (它石智航), just walked away with the biggest single check the Chinese embodied intelligence (具身智能) field has ever written.
Now the hard part begins: turning a record valuation into machines that ship by the thousand instead of by the prototype.
Why a near-pre-revenue lab commanded a $2B ticket
TARS closed a Pre-A round above $450 million — roughly RMB 3.07 billion (about HK$3.3 billion) — in April 2026, smashing the prior single-round record for Chinese embodied-AI (具身智能) startups. Hillhouse (高瓴创投) and Sequoia China (红杉中国) co-led, with Meituan's strategic arm (美团战投) and a long tail of state and industrial money following on.
This was not a first bite. The company had already banked about $242 million (RMB 1.65 billion / HK$1.79 billion) across two angel rounds in 2025, itself a record for the stage. People close to the deal, cited by Caijing (财经), put the post-money valuation of this latest round at roughly RMB 15 billion (about $2.1 billion / HK$16.3 billion), with the next round already structured above RMB 20 billion (about $2.8 billion / HK$21.7 billion).
The cap table reads like a who's-who of Chinese tech capital:
- State vehicles — Beijing Robotics Industry Fund (北京机器人产业发展投资基金) and Shanghai State-owned Assets (上海国投) making their first joint bet on an embodied-AI (具身智能) company.
- Strategic holders — Meituan (美团), which had already backed the angel+ round.
- Industrial names — TCL (TCL创投), Fortune Capital (孚腾资本), and SAIC-linked capital.
- Pure financials — CICC Capital (中金资本), BlueRun, and several returning early backers.
The bet: wiring harnesses, not wheelies
Most humanoid (人形机器人) headlines chase dancing robots. TARS aimed elsewhere from day one — the unglamorous work of snapping flexible wiring harnesses into cars and consumer electronics, the kind of fiddly assembly that automation has historically failed to crack and insiders have nicknamed the "Goldbach conjecture" of the factory floor.
In March 2026 its A1 unit set a Guinness record by completing over 100 sub-millimeter flexible-harness assemblies in a single hour, the first such record by a Chinese embodied-AI (具身智能) firm in precision industrial manipulation. The company says its general embodied model, AWE3.0, and its wearable data-collection system, SenseHub, are what let the machine hold that precision across unseen angles and long, uninterrupted runs.
That is the story investors bought: not a robot that performs, but a robot that wires.
The team investors are really pricing
TARS was founded in February 2025, which makes the valuation sprint doubly striking. The premium is mostly about people:
- Li Zhenyu (李震宇), former president of Baidu's intelligent-driving group, who built the Apollo (阿波罗) autonomous-driving platform and the "萝卜快跑" (Luobo Kuaipao) robotaxi service.
- Chen Yilun (陈亦伦), ex-Chief Technology Officer of Huawei's autonomous-driving unit and former machine-vision lead at DJI (大疆).
- Ding Wenchao (丁文超), a former Huawei "genius youth" (天才少年) who built the company's end-to-end decision network.
A founding team pulled from Baidu, Huawei, and DJI (大疆) is exactly the pedigree that lets a 14-month-old company command a ticket usually reserved for decade-old hardware firms.
Where the money meets the wall
The cash does not erase the industry's shared headache: building the robot is the easy part, shipping it is not.
TARS has told reporters it expects capacity of about 1,000 units this year, and that every unit is already spoken for — yet delivery had not completed at the time of the reporting. Industrial-scene robots in China typically sell for RMB 200,000–400,000 (about $28,000–$56,000 / HK$217,000–HK$434,000), before discounts, which implies revenue in the low hundreds of millions of yuan at best if fulfillment holds.
Caijing (财经) noted the broader awkward truth: even front-runners still hand-assemble robots because parts are not standardized, and a single missing component can stall an entire batch. For a company whose value rests on "real work, real volume," the gap between a Guinness clip and a thousand crates leaving the dock is the whole ballgame.
What readers can do now
- Watch delivery, not announcements. For embodied-AI (具身智能) names, track realized unit shipments and returned orders; a recorded demo proves little about a factory line.
- Read the cap table. When state industry funds and a strategic like Meituan (美团) co-invest, they often bring paid pilot sites — a better signal than headline valuation.
- Stress-test the "sub-mm" claim. Ask whether precision holds outside the curated harness task before treating a record as a moat.
Honest limitations
Figures here come from the company's own announcement plus Chinese financial reporting (Caijing / 财经, China Industry News / 中国工信新闻网, and tech press). No exchange filing or Western wire (Reuters/Bloomberg) covered this private round, so the dollar amounts rest on domestic outlets corroborating each other rather than an independent audit. The RMB-to-USD/HKD conversions use an approximate 7.2 RMB-per-dollar rate and may differ slightly from the companies' internal books. The valuation and "next round" numbers are attributed to a single exclusive and should be treated as indicative, not confirmed by a term sheet we can see. We have not independently verified order backlogs or production capacity.