Xpeng Robotics Raises $900 Million to Mass-Produce IRON Humanoid

A record round for a robot that isn't on sale yet

Xpeng, the Guangzhou electric-vehicle maker that now calls itself a Physical AI company, says its robotics business raised more than US$900 million in its first outside funding round. The deal values the unit at more than US$6.3 billion after the money came in.21 IDG Capital led the round. Gaorong Ventures also invested, and Tencent and Alibaba joined as strategic investors.21 Xpeng calls it the largest single private round so far in China's embodied AI industry.223

The money is going to several places at once: software and hardware R&D for the robotics unit, training and updating its Physical AI models, producing high-quality training data, building end-to-end mass-production facilities, and expanding sales abroad.21 The product behind all of it is IRON, Xpeng's next-generation general-purpose humanoid. The company says IRON will enter mass production by the end of 2026, work first in Xpeng's own stores and campuses, and go on sale in China and overseas in 2027.3

The news is reaching a wider audience now, in October, but the deal itself is not new. Xpeng announced it with its second-quarter results on August 24, and coverage at the time called it a Series A.292 Xpeng is putting the round back in front of investors as it takes IRON to the Paris Motor Show, where its stand includes a section devoted to humanoid robots.2227

Read the cap table carefully

The $900 million figure is real, but not all of it is outside money. Electrek's breakdown puts roughly $600 million from external investors, about $200 million from an Xpeng subsidiary and about $100 million from company leadership.2 So about a third of the round comes from Xpeng itself and its own executives. Most reports repeat the headline number without mentioning this. It doesn't make the round less meaningful, since IDG, Tencent and Alibaba still put serious money into a separately financed robotics unit. But investors buying the $6.3 billion valuation should keep it in mind. Humanoids Daily describes the structure as a carve-out: robotics became its own subsidiary with its own funding.5

The sources mostly agree on why this matters. Funding the unit separately lets Xpeng pay for an expensive, long-term bet without loading it all onto a car business that is still losing money. Xpeng posted a $200 million net loss in the second quarter, on revenue of $2.91 billion.24 Another report says the stock had fallen about 50% this year as demand for EVs in China weakened.28 One analysis noted that investors had given the robotics arm a multibillion-dollar price before it had earned a single dollar of commercial revenue.29

From stage demo to production line

What separates Xpeng from most humanoid hopefuls is that it has started building a factory. On September 8 it switched on what it calls the world's first automated production line for advanced humanoid robots in Guangzhou. The first IRON walked off the line under its own power.1 Xpeng says more than 80% of the line's core processes are automated, and it applies the same automotive quality systems it uses for its cars.16

The published specs are ambitious. IRON has 76 degrees of freedom, including 21 in each hand. It runs on three of Xpeng's own Turing AI chips, which the company says deliver up to 2,250 TOPS. That is enough to run its foundation model on the robot itself, without a human operating it remotely.1 Xpeng says it aims to produce 1,000 units a month starting at the end of the year.26 On the strategy side, the company says per-unit gross margin on IRON should be well above what it earns on EVs.8 CEO He Xiaopeng has argued that a single robot could earn more over its lifetime than a car.24

The sources differ on how much the factory proves. Several outlets treat the launch as a lead over Tesla, whose Optimus program they describe as stalled.14 More skeptical coverage points out that one robot walking off a line is not mass production. Xpeng has not published yield rates, cycle times or per-unit costs.6 Nor has it explained which assembly steps are done by robots and which by people. "More than 80% automated" is not the same as "robots building robots."8 Durability is a known concern too: earlier factory trials reportedly found the robot's hands wore out within about a month of repetitive work.5

Our reading: the factory matters more than the funding, but neither proves demand. A production line shows Xpeng can build humanoids. Only paying customers can show the humanoids are worth buying.

The timeline is already drifting

The most telling difference between sources is about timing. Xpeng's official line remains mass production by the end of 2026 and customer deliveries in 2027.3 But President Brian Gu has more recently said IRON production should ramp at the start of 2027, first for Xpeng's own stores and offices, with external sales in the second half of 2027. The 1,000-a-month figure is a capacity target, not output Xpeng has achieved.

This is not a scandal. He Xiaopeng himself called the factory launch a small step.1 Still, the gap between end-2026 and mid-to-late 2027 is worth watching. Humanoid timelines across the industry have repeatedly slipped, and Xpeng's has slipped before.1

Why automakers think they have an edge

The strongest argument for Xpeng is its supply chain. More than 85% of IRON's parts suppliers already supply Xpeng's cars, according to BigGo Finance.26 A Hanyang University robotics professor quoted there says EV makers already own much of the supply chain a humanoid needs. He also says designing your own AI chips, instead of buying them, makes the robots cheaper to build.26 The same report places Xpeng alongside Hyundai and Tesla, which are also reusing car factories and suppliers to build humanoids. Hyundai is preparing to put robots to work at its Metaplant America facility in the U.S.26

That makes this more than a startup funding story. A company with three car plants and roughly 430,000 vehicles sold last year is bringing the same manufacturing discipline that drove EV prices down.26 If that works, humanoid prices could fall the way EV prices did.

Where this sits in the humanoid funding race

Xpeng's round is large, but it isn't an outlier. Figure AI is valued at about $39 billion after its $1 billion-plus Series C.19 NEURA Robotics announced a Series C of up to $1.4 billion.12 Skild AI raised close to $1.4 billion at a valuation above $14 billion.20 Unitree went public on Shanghai's STAR Market in August. One tracker says its shares closed the first day up 487%, valuing the company at roughly $53 billion.15 Sources disagree on how much Unitree's IPO raised: one puts it near $905 million, against an earlier target of about $610 million.1213

Next to those figures, $6.3 billion puts Xpeng's robotics unit in the industry's second tier, roughly level with Apptronik, NEURA and AgiBot.19 China's real strength is volume. AgiBot and Unitree each shipped around 5,000 humanoids in 2025.14 One forecast expects the two to account for nearly 80% of global humanoid shipments in 2026.16 Xpeng is entering a home market where specialist robot makers are already producing at scale.26

The warehouse question

For anyone following warehouse automation, IRON is not yet a logistics robot. Xpeng's first planned uses are greeting and guiding customers in stores, reception, tours and inspection, mostly at its own sites.10 Earlier messaging talked about industrial work and helping in factories.7 The commercial launch puts shop-floor customer service first.26

Humanoids already doing paid warehouse work look different. Agility's Digit has moved more than 100,000 totes at GXO, and GXO is testing several humanoid suppliers instead of committing to one.20 Agility reports more than 65,000 operating hours and over $300 million in multi-year orders, though those orders depend on hitting milestones.19 Yet logistics-focused humanoid companies are valued well below general-purpose ones: Agility's reported valuation is about $2.1 billion.20 Investors are paying more for broad, general-purpose ambition than for narrow tasks that already work.

Xpeng's bet fits that pattern. Its backers describe IRON as a gateway to real-world data, feeding a production-to-data-to-models cycle.3 If that cycle works, warehouses and factories are the obvious next market, since controlled settings and measurable tasks make humanoids easiest to justify there.12 But that is Xpeng's long-term pitch, not its current plan.

The bottom line

Xpeng has done two things most humanoid companies have not: it has raised funding at scale and built a working production line. The open questions are commercial. Can IRON do useful work in Xpeng's own stores, when do outside customers start paying, and can the robotics business grow while the car business loses money? The real test comes in 2027, and the company's own executives have already pushed parts of that schedule back.