Published on

XPENG Launches Mass Production of IRON Humanoid

Get your news fromHumanoids Daily

One tap, and our reporting appears more often in your Google Top Stories. It makes a real difference to a small newsroom.

Humanoids Daily
Written byHumanoids Daily
XPENG CEO He Xiaopeng gives a thumbs-up standing alongside the mass-produced IRON humanoid robot inside a brightly lit manufacturing facility. The robot features a textured white bodysuit, glossy black visor faceplate, and wears an XPENG employee ID badge around its neck.
XPENG CEO He Xiaopeng standing with the production-intent IRON humanoid robot inside the manufacturing facility, complete with an XPENG company ID badge as it prepares for deployment. Image: XPENG
  • XPENG has officially initiated mass production of its IRON humanoid robot, claiming a global first as production units assemble and walk off the line entirely under their own power without human assistance or teleoperation.
  • The production line features automotive-grade quality control and exceeds 80% automation across key assembly steps, leveraging an 85% supply chain overlap with XPENG’s EV manufacturing infrastructure.
  • The announcement sharpens an intensifying manufacturing contest with Tesla, which has missed its public Q1 reveal for Gen 3 but recently placed bulk orders for 5,000 component sets as it debugs its Fremont assembly line.
  • Powered by three in-house Turing AI chips delivering 2,250 TOPS, the production-grade IRON introduces multi-modal bilingual dialogue, dynamic obstacle avoidance, and autonomous charging dock navigation.
  • Initial IRON units are being deployed across XPENG showrooms and campuses for customer interaction and facility patrols, ahead of broader enterprise customer deliveries slated for 2027.

XPENG has officially initiated mass production of its flagship bipedal platform, IRON, marking what the Chinese automaker calls the world’s first high-level general-purpose humanoid robot to be manufactured on a commercial assembly line and walk off under its own power without human assistance or remote control.

The milestone follows XPENG Robotics securing an unprecedented $900 million private funding round at a $6.3 billion valuation. More crucially, it cements an aggressive schedule that directly challenges its chief automotive peer, Tesla, which has struggled to lock down a production-intent design ready for industrial-scale manufacturing while keeping its latest hardware behind closed doors.

Manufacturing Engineering Over Prototyping

While the wider humanoid robotics sector remains largely stalled in low-yield assembly and manual prototyping, XPENG is leveraging its automotive heritage to transform the discipline into scaled manufacturing engineering.

The production line—engineered to meet stringent vehicle-grade tolerances—achieves over 80% automation across key manufacturing stages, including dedicated automation equipment for the assembly of dexterous hands. According to the company, more than 85% of IRON’s physical supply chain overlaps directly with XPENG’s electric vehicle business, giving it immediate economies of scale in motors, structural alloys, wiring harnesses, and battery integration that pure-play robotics startups cannot match.

Stay ahead in humanoid robotics

One email a week: the launches, funding and research that mattered. Join the researchers, founders and technologists following the field.

"We want to be the Apple of integration, from chip to OS, from joint to hand," said XPENG CEO He Xiaopeng. "Only by going deep across every layer can we build something truly different".

The announcement validates the automaker's strategy after breaking ground on its 110,000-square-meter full-chain humanoid manufacturing hub in Guangzhou earlier this year. By treating humanoid assembly with the same operational rigor as EV production, XPENG has beaten its own late-2026 mass production timeline to market.

A Divergent Race Against Fremont

XPENG's visible move to mass manufacturing sharpens the competitive dynamic with Tesla. Both companies have championed the thesis that electric vehicle infrastructure, in-house silicon, and physical foundation models are the prerequisite ingredients to scale bipedal robotics. Yet their execution playbooks show starkly different approaches to publicity and validation.

While Tesla missed its promised Q1 2026 deadline to publicly unveil its production-intent Optimus Gen 3 prototype, recent supply chain signals indicate the automaker is quietly preparing for scale. Industrial reports suggest Tesla has placed initial purchase orders for roughly 5,000 component sets across core Chinese actuator and precision roller screw suppliers following summer line commissioning at its Fremont facility. Tesla’s internal targets reportedly aim to deliver between 10,000 and 20,000 humanoid units for in-house factory tasks across 2026, even as the company relies on closed-door demos and private viewings to generate momentum.

Side profile of the XPENG IRON humanoid robot walking autonomously off the mass production line. The bipedal robot is clad in a sleek white textured bodysuit and headpiece with a reflective dark faceplate, walking past overhead light panels in an automotive-grade assembly facility.
XPENG IRON walks off the assembly line entirely under its own power, marking the platform's transition from prototype trials to automated serial production. Image: XPENG

In contrast, XPENG’s autonomous line-exit demonstration puts its manufacturing reality front and center. Rather than keeping next-generation hardware behind closed doors while resolving the steep engineering hurdles of complex 25-actuator hands, XPENG is moving directly into commercial assembly line validation with its current design.

Upgraded Autonomy and On-Device Compute

The production iteration of IRON brings significant upgrades to its intelligence and mechatronic stack:

  • Triple-Turing Edge Compute: The robot carries three proprietary Turing AI chips delivering an aggregate 2,250 TOPS of compute, running physical foundation models locally to eliminate latency and cloud dependence.
  • Unified Data Flywheel: In Q2 2026, XPENG merged its robotics and autonomous-driving data teams. The platform reuses the Vision-Language-Action (VLA) pipeline honed across millions of highway kilometers to accelerate real-world spatial reasoning.
  • Multi-Modal Interaction: Equipped with an in-house multi-modal LLM, IRON supports persistent memory across 12 unique human faces and dynamic code-switching during Chinese-English bilingual dialogue.
  • Full-Process Obstacle Avoidance: Navigates dynamic, unstructured spaces autonomously, planning paths and reacting to moving obstacles in real time.
  • Autonomous Docking: Independently locates charging docks, manipulates power connectors to plug and unplug, and recovers from minor mechanical alignment faults without human intervention.

Commercial Deployment: Showrooms First, 2027 Deliveries

XPENG is maintaining a deliberate, phased commercial rollout. Rather than deploying immediately into high-risk heavy industrial tasks or unstructured private homes—where early trials previously revealed severe durability limitations in dexterous hands—the company is prioritizing structured commercial spaces.

Production units rolling off the line are being deployed immediately across XPENG-owned retail showrooms and corporate campuses for guided customer tours, product demonstrations, and perimeter patrols. Throughout Q4 2026 and early 2027, these deployments will serve as high-density data collection hubs, validating edge cases and feeding the "data-model-application" flywheel.

External customer deliveries are scheduled to begin in 2027, targeting enterprise clients across commercial retail, hospitality, and customer service. While Tesla works behind closed doors to ramp component pipelines and fine-tune its production lines, XPENG has taken the first public step in transitioning the humanoid race from pilot-line promises to documented factory throughput.

Share this article

Stay ahead in humanoid robotics

One email a week: the launches, funding and research that mattered. Join the researchers, founders and technologists following the field.