XPeng IRON humanoid robot production begins as its first unit walks out of the Guangzhou factory
The XPeng IRON humanoid robot has moved into series manufacturing. The first unit completed assembly at the company's newly commissioned Guangzhou plant and left the line on its own, without an operator guiding it. XPeng announced the milestone in early September and described it as the first instance of an advanced general-purpose humanoid being built on automated production equipment and then departing it independently, a step the company treats as the boundary between its research prototypes and planned volume output.
The factory applies manufacturing practices borrowed from XPeng's EV business. Its core production steps run at more than 80 percent automation, and its quality systems come from the company's automotive operations rather than from a purpose-built robot plant. XPeng has said it intends to feed this line from its existing EV supplier base instead of creating separate supply chains for robots.
The self-directed walk-off was the visible proof that the line was working. After final assembly, the unit moved out of the production area with no one controlling it, a distinction between factory-built machines and the hand-assembled demonstrators that most humanoid makers have shown so far, which typically require a team to relocate them.
Inside the XPeng IRON humanoid robot factory
IRON has 76 degrees of freedom, 21 in each hand, and performs all of its computation on board. Three in-house Turing AI chips supply up to 2,250 TOPS of effective compute and run a Physical AI foundation model locally, so the robot does not need a cloud connection to make sense of its surroundings. XPeng also designed a proprietary enclosed flexible lattice structure into the body, which the company presents as a safeguard for working around people.
Producing the silicon in-house matches XPeng's approach to its vehicles, where the company controls the computing stack instead of buying it from a supplier. Building the Turing chips and the Physical AI model together lets XPeng tailor the software to its own hardware, and it keeps the robot's intelligence layer out of the open market, where rivals could purchase the same components.
Running inference locally shapes the deployment plan. Avoiding server round trips shortens response time and keeps the robot operational in places with weak connectivity, both useful when a machine works in a store among customers and staff. On-device processing also limits what data leaves the robot, a consideration for commercial settings.
Mass production is set for late 2026, and the first units will remain inside XPeng. The company plans to place IRON at its own campuses and retail stores before it begins outside sales in China and overseas in 2027. That order of operations keeps early reliability problems inside controlled environments, generates operating data from real floors, and puts a working humanoid in front of store visitors months before it goes on sale. The retail network functions as a testing ground, and the operating data collected there will shape the commercial use cases XPeng offers external buyers in 2027.
| IRON production facts | Detail |
|---|---|
| Facility | Humanoid robot plant in Guangzhou, commissioned September 2026 |
| Line automation | More than 80% of core processes automated |
| Degrees of freedom | 76 across the body, 21 per hand |
| Compute | Three in-house Turing AI chips, up to 2,250 TOPS |
| AI model | On-device Physical AI foundation model |
| Mass production | Targeted for late 2026 |
| First sales | China and overseas, 2027 |
| Robotics financing | Over $900M raised in August 2026; post-money valuation above $6.3B |
IRON has appeared in public since its 2024 debut, including stage demonstrations at XPeng's AI Day in late 2025. The commissioning shifts the competitive picture. Walking demos are now routine in the humanoid sector; production lines that repeat the build are not. XPeng's end-of-2026 mass-production date also puts its schedule ahead of Tesla's Optimus program, which has not started series output of its own humanoid. The competitive gap is no longer about which robot can walk on stage but which company can reproduce the build at volume.
The margin bet and the funding behind it
XPeng describes robotics as its second growth curve, and the logic is straightforward: the company expects gross margin per unit on general-purpose humanoids to run well above the margins its new-energy vehicles earn, pointing to the high barriers to entry in the category. For an automaker competing in a price-sensitive domestic EV market, a second product line with better unit economics changes the shape of the business rather than merely extending it.
That thesis had capital behind it before the production line opened. In late August 2026, XPeng's robotics unit closed a private funding round that raised over $900 million and valued the business at more than $6.3 billion after the investment. IDG Capital led the round; Alibaba and Tencent participated. XPeng says it was the largest single financing in China's embodied-AI sector, and the company said the money was earmarked for series production, not further prototyping. The financing landed weeks before the Guangzhou commissioning, which is the first public evidence of where that capital went in physical form.
The manufacturing plan is built on reuse. Quality systems, automated processes, and supplier relationships developed for cars are being redirected at humanoids, which keeps capital intensity lower than it is for robotics startups that must fund their own tooling. The open question is execution. Humanoids operate in unstructured spaces around people, and their most serious failure modes are not the ones an automotive assembly line is designed to catch. XPeng's store and campus deployments are the first test of whether the transfer holds.
What serial production changes
The funding round also reset expectations across the sector. XPeng, the EV maker founded by He Xiaopeng and listed on the NYSE under XPEV and on the Hong Kong exchange, has put a public figure on the cost of bringing a humanoid to production, and that figure becomes a benchmark for later financings in China's embodied-AI market. The investor list adds weight: IDG Capital led, and Alibaba and Tencent joined, linking the robot push to the strategic capital of some of the country's largest technology groups. Startups that entered the race with prototypes are now raising against a company with an operating factory and a stated 2027 sales plan.
XPeng's timeline also shortens the run from first unit to full production. The Guangzhou line started running in early September, and the company targets mass production for late 2026, a ramp measured in months rather than years. Holding that pace depends on the automation rate holding up as throughput rises, the same scaling task XPeng has managed in its vehicle plants.
Two checkpoints will show whether the strategy works. The late-2026 ramp will reveal whether a line that runs at more than 80 percent automation can hold quality as output grows. The 2027 commercial launch will then put IRON's price in front of customers in China and overseas and give the company's margin claims their first market test.
Why this matters
XPeng has brought EV supply-chain economics into humanoid manufacturing, betting that a robot line can run on car-plant discipline while earning margins its vehicle business cannot match. If the XPeng IRON humanoid robot ramps on schedule, the sector's benchmark shifts from walking demonstrations to repeatable output, and China's embodied-AI valuations will be measured against a factory that actually ships. The internal deployments scheduled for late 2026 are the first evidence; the 2027 commercial launch is the verdict.
Related Articles
- XPeng Robotics Funding Round: Humanoid Unit Valued at $6.3B in Record Raise
- China pushes 10,000 humanoid robots by 2026 as factories struggle to absorb them
- Xiaomi Tieda humanoid robot hits 98% factory accuracy after auto-plant trial
✔Human Verified
Researched and cross-referenced against primary sources by the Bytevyte editorial team. This article was generated with the assistance of artificial intelligence and reviewed by the Bytevyte editorial team.