Xpeng Raised $900 Million to Build Humanoid Robots on a Car Assembly Line

Xpeng's robotics business just raised more than $900 million at a valuation above $6.3 billion, in what the company says is the largest single private round China's embodied AI sector has seen. IDG Capital led it, Gaorong Ventures joined, and Tencent and Alibaba came in as strategic investors.

The money goes toward one machine: a humanoid robot called IRON, which Xpeng expects to enter mass production by the end of 2026.

What IRON is

IRON has 76 degrees of freedom across its body and 21 in each hand, which is what lets a robot do things like turn a handle or pick up a glass without knocking it over. Hand dexterity is where most humanoids fall apart, so putting 21 joints in each one is a real commitment rather than a spec-sheet flourish.

The body uses a fully enclosed flexible lattice structure, which keeps the moving parts covered. Three of Xpeng's own Turing AI chips run onboard, delivering a combined 2,250 TOPS. That compute lives on the robot, so Xpeng's physical AI model runs locally instead of phoning home, which cuts latency and keeps the data on the device.

Why a car company is building this

Electric vehicles and humanoid robots need a lot of the same things: batteries, motors, sensors, embedded compute, computer vision, and a factory that can build the whole assembly to a consistent standard. Xpeng has spent twelve years assembling that stack and says it's applying automotive-grade quality standards and production capacity to robotics.

CEO He Xiaopeng framed the company's advantage as full-stack in-house development across its world model, its Turing chips, and its AI infrastructure. Xpeng also builds robotaxis and flying vehicles, so the robot is a third use for the same underlying pile of engineering.

The part to watch skeptically

Read the deployment plan carefully. IRON's initial commercial deployment is at Xpeng's own stores and campuses. Deliveries to actual customers in China and overseas start in 2027.

Putting robots in your own showrooms is a sensible way to gather behavioral data and catch failures before strangers are involved. It's also the step where a lot of humanoid programs have stalled, because a robot that impresses visitors in a controlled retail space is a long way from one that earns its cost doing repetitive work in somebody else's warehouse. Tesla's Optimus timeline has slipped repeatedly on exactly this gap.

Xpeng hasn't published a monthly production target, a per-unit price, or what specific job IRON will do first. Those three numbers are what would tell you whether this is a product or a very well-funded demonstration.

The bigger money story

Chinese EV makers are turning into robotics companies while their car businesses face brutal margins. Investors are backing the pivot at scale, and this round sets a record for the sector. Whether that capital produces working machines or a very expensive lesson in how hard hands are will be clear sometime in 2027, when the first units land with people who paid for them.