XPENG Iron Humanoid Walks Off Line as China Scales Robot Output

On September 8, 2026, a XPENG IRON humanoid robot stepped off a production line under its own power — the company says the world's first "high-tier general-purpose humanoid" completed automated final assembly and walked away unassisted. XPENG frames the moment as crossing from hand-built prototypes into factory manufacturing: the line, designed and developed entirely in-house, is billed as the first automated production line for advanced humanoids anywhere, with core-process automation above 80 percent.

The launch of the line follows, by about two weeks, the biggest fundraising event China's embodied-AI industry has seen. On August 24, XPENG said its robotics business had signed agreements to raise over US$900 million at a post-money valuation of above US$6.3 billion — the largest single-round private capital raise ever recorded in China's embodied-AI sector — with IDG Capital leading, Gaorong Ventures participating, and Tencent and Alibaba joining as strategic investors. XPENG retains controlling ownership and will keep consolidating the unit into its financial statements.

The financing and the factory answer to the same deadline. XPENG targets entry into scale mass production by the end of 2026 — Chairman and CEO He Xiaopeng calls it a bid to be "the world's first to scale an advanced high-tier humanoid robot" — with initial deployment inside XPENG's own stores and campuses, then a formal launch and deliveries in China and overseas in 2027. Retail-facing uses such as store guiding, tour-guiding and service are expected to begin in early 2027. The plan calls for monthly capacity of more than 1,000 units by end-2026, ramping to several thousand units a month — and potentially higher — through 2027 as quality and the supply chain mature.

On hardware, IRON carries 76 degrees of freedom across the body and 21 in each hand. The dexterous hands use direct-drive actuation rather than the tendon-driven designs several rivals favor, sized to an adult hand and built with a biomimetic skin under development. Intelligence is handled by three in-house Turing AI chips delivering up to 2,250 TOPS of effective on-device compute — enough for XPENG's Physical AI foundation model to run on the robot itself, enabling low-latency, autonomous performance of complex tasks without remote teleoperation. The stack includes a second-generation VLA model and a fourth-generation motion-control system.

The cost argument is central to the bet. XPENG says more than 85 percent of IRON's supply chain overlaps with its electric-vehicle supply chain, giving it a manufacturing head start most robotics startups lack. He notes the industry generally prices robots at roughly 2.5 to 3 times materials cost, and expects IRON's hardware gross margin to run clearly above the car business — the premise behind The Wall Street Journal's framing that XPENG bets robots will be more profitable than its cars. He also cautions that an advanced humanoid is, in his phrasing, "more than twenty times as complex" as a smart car, which is why XPENG develops chips, actuators and control in-house.

Manufacturing sits at the Guangtang Sci-Tech City embodied-AI industrial park in Guangzhou's Tianhe district — a full-chain base of roughly 110,000 square meters whose construction began in the first quarter of 2026. XPENG has worked on robotics for around eight years and, since June 2026, He Xiaopeng has also served as CEO of the robot business. At group level, IRON sits alongside smart EVs and robotaxis as one pillar of XPENG's "Physical AI" strategy, built on a shared stack of Turing chips, a physical-world foundation model and AI infrastructure.

China's crowded 2026–27 robot race

The ramp drops XPENG into a crowded field. Unitree's prospectus shows roughly 5,500 humanoid units shipped in 2025; UBTech reported first-half 2026 full-size humanoid revenue of ¥590 million (up 1,445 percent) on 921 units sold (up 1,947 percent) at a 66.8 percent gross margin — evidence that Chinese firms already dominate the upper tier by volume. Tesla has meanwhile indicated Optimus production begins toward the second half of 2026 and into 2027. What distinguishes XPENG is less the ambition than the beachhead.

Original insight: the storefront, not the factory

The following is analysis, not established fact. XPENG is deliberately staging its entry inside its own retail and service network rather than chasing the factory or home segments most rivals target. The logic is a controlled real-world data loop: stores are places XPENG commands from end to end, where IRON can generate real operational data — and, by XPENG's accounting, monetize — while still learning. It is a bet that a vertically integrated, single-company model, carried by a decade of automotive supply chain, can scale an advanced humanoid faster than the more horizontal, hardware-agnostic approach startups favor. The risk is not ambition but execution: if general-purpose manipulation and data generalization do not mature by 2027, an attention-grabbing ramp can become "mass production before capability" — in the Chinese industry's blunt phrase, 量产即落后 (shipping volume that arrives before the robot is truly capable). This line proves XPENG can build; whether it can sell at scale will depend on whether those storefronts create real value rather than demos.

The bottom line

The immediate takeaway is narrower than the hype: XPENG now has a funded, full-chain factory, a self-styled world-first automated humanoid line, and a hard date. The December 2026 ramp and the 2027 deliveries are now verifiable milestones — which is exactly why they will be scrutinized. If IRON walks off the line in volume and works in real stores, XPENG will have moved China's embodied-AI story from demonstration to industrialization. If it stalls, the $6.3 billion valuation will read as a claim on a future that has not yet arrived.