About an hour before it reported second-quarter results on Monday, XPeng dropped the bigger news: its robotics business has raised more than $900 million at a post-money valuation of over $6.3 billion, in what the company calls the largest single-round private capital raise in the history of China's embodied-AI industry. The money is meant to push IRON, XPeng's humanoid robot, from stage demo into a factory — with mass production promised by the end of this year.
The round is led by IDG Capital, with participation from Gaorong Ventures and strategic backing from Tencent and Alibaba. But the headline number deserves an asterisk before anything else: only about $600 million of it comes from outside investors. The rest is XPeng and its own executives writing checks to themselves.
Who actually put in the money
A filing with the Hong Kong exchange, first detailed by CnEVPost, gives the structure a name and a shape. The vehicle is called Dogotix, and the $900 million breaks down three ways: $600 million from external investors, $200 million from Xpeng Dogotix, a wholly owned XPeng subsidiary, and $100 million combined from companies controlled by chairman and CEO He Xiaopeng and co-president Brian Gu, subscribing for ordinary shares.
The pre-money valuation was $5 billion. The widely reported $6.3 billion post-money figure assumes the unit's equity incentive mandate is fully utilized, and excludes both a potential additional investment and a large tranche of warrants. CnEVPost changed its own headline to reflect that distinction after publishing — a small sign of how much interpretive room these numbers carry.
XPeng will hold roughly 81.97% of Dogotix, falling to about 68.41% if warrants letting He and Gu's vehicles add another $400 million and $100 million are exercised and the 15% incentive pool is fully used. Either way Dogotix stays consolidated on XPeng's books. Investors also negotiated a floor: if Dogotix fails to complete a qualified IPO within seven years, they can force a buyback at the higher of cost plus 8% compounded annually or 120% of cost. That is not the term sheet of investors who consider this a sure thing.
Two details the press release omits: Dogotix carried net liabilities of about 447 million yuan ($65.9 million) as of March 31 on unaudited management accounts, and none of the closing conditions had been satisfied or waived as of the announcement, meaning the transaction may not complete. The carve-out of assets, IP, personnel and systems is expected to take up to 18 months.
What IRON is, and what it is promised to do
IRON, unveiled at XPeng's AI Day in November 2025, has 76 degrees of freedom across the body and 21 in each hand, wrapped in a proprietary "fully enclosed flexible lattice structure" — synthetic skin. It runs three of XPeng's in-house Turing AI chips for a combined 2,250 TOPS, with a physical-AI foundation model running on-device. XPeng claims IRON can "autonomously perform complex tasks without remote operation," a pointed contrast with rival demos that turn out to be teleoperated off-camera. The robot looked human enough at launch that viewers accused XPeng of stuffing a person into a costume; He responded by cutting open its leg on stage to show the wiring.
He, who added the robotics CEO title to his job in June and reorganized the unit into nine second-tier departments, framed the raise in terms of the last decade of car-building:
"Over the past 12 years, XPENG has remained committed to full-stack in-house R&D, building a solid technological foundation for the physical AI era—across our physical world foundation model, Turing AI chips, and AI infrastructure. This has enabled us to pioneer a new phase of mass production and commercial deployment for advanced humanoid robots."
IDG Capital's statement, issued as a firm rather than through a named partner, made the thesis explicit: "The embodied AI industry stands at a pivotal juncture, transitioning from technical breakthroughs to scalable manufacturing and commercial deployment."
The targets are targets
Here is the ladder XPeng is asking people to climb. Mass production by end of 2026, initially deployed in its own stores and campuses. Monthly capacity of more than 1,000 units. Commercial launch and deliveries in China and overseas in 2027. And a floated goal of 1 million robots by 2030.
The last one should be read as ambition, not guidance. It appears in essentially one place — Electrek's coverage, which hedges it as "floated" — and shows up in neither the HKEX filing nor the press release. Going from a thousand units a month to a million cumulative units in three years requires roughly a 30x scale-up on top of a production line that does not yet exist.
The near-term target is more testable and still aggressive. Unitree, the world's leading humanoid shipper, moved more than 5,500 units in all of 2025 — roughly what XPeng says it wants to build in five months. Tesla, for its part, has missed every Optimus milestone since 2021, including a 2025 target of 5,000 units. "Automotive-grade mass production" is a real advantage on paper; XPeng delivered 103,295 vehicles in Q2, up 64.8% quarter-over-quarter. Whether an EV line's tooling and supplier discipline transfer to a machine with 118 degrees of freedom is an open engineering question, not a marketing one.
The valuation math is its own signal. At $6.3 billion, Dogotix is priced near Unitree's roughly $6.2 billion base and far below Figure AI's ~$39 billion or 1X's ~$10 billion — a reminder that Chinese embodied-AI valuations are being set against manufacturing plans while American ones are set against narrative.
What to watch
Three things. First, closing: the conditions are unsatisfied and the carve-out runs 18 months, so the deal is announced, not done. Second, the end-of-2026 line rate — a robot rolling off a line into XPeng's own showroom is a soft launch, and the real test is paying external customers in 2027. Third, autonomy. XPeng has staked its differentiation on on-device inference with no human in the loop. That claim is easy to make on a stage and hard to sustain in a store full of strangers. It gets checked when IRON is in customers' hands.
“Over the past 12 years, XPENG has remained committed to full-stack in-house R&D, building a solid technological foundation for the physical AI era—across our physical world foundation model, Turing AI chips, and AI infrastructure.”— He Xiaopeng, Chairman & CEO, XPENG