XPeng IRON Walks Off the Line; China EV Pivots to Humanoid Hardware
The first IRON robot just shipped itself—and XPeng says margins could beat selling cars.
XPeng flipped the switch on humanoid production this week. The IRON robot—bipedal, 5'7", roughly 130 pounds—walked off the assembly line under its own power on September 7. Not a prototype. Not a stage demo. Off the production line, through a door, into a shipping queue. The Chinese EV maker says first customer deliveries start in 2027, and the company believes robotics margins could exceed automotive margins within three years.
The production line is live in Guangzhou. XPeng hasn't disclosed unit capacity, but the company confirmed serial production has begun and that IRON is being built alongside internal tooling for its own vehicle plants. The bot is designed for repetitive tasks in manufacturing, logistics, and—eventually—retail. The company is positioning it as a labor arbitrage play in markets where wage inflation is structural, not cyclical.
Samsung SDI, LG, and SK On are now racing to supply batteries for humanoid platforms. All three confirmed active development programs this week, targeting energy density above 200 Wh/kg and cycle life beyond 3,000 charges. The prize: becoming the default power supplier for a category that doesn't yet have volume orders but has every EV and electronics incumbent scrambling to avoid getting Blockbustered by a hardware shift they didn't price in.
XPeng's move follows a familiar Chinese playbook: subsidize a loss-leader consumer product (EVs), use the scale to build manufacturing competency, then pivot that competency into B2B hardware with better unit economics. The company isn't alone. Multiple Chinese EV makers are expanding into robotics, treating humanoids as the next vertical integration opportunity rather than a science project.
Meanwhile, Tesla's Optimus remains in limited pilot production. No public timeline for volume manufacturing. No third-party sales announced. XPeng is shipping a product while Tesla is still posting Instagram Reels of Optimus folding laundry. The gap isn't capability—it's commercialization speed. One company is treating humanoids like a product line. The other is treating them like a keynote slide.
Elsewhere in AI capex: a new report pegs shadow financing behind the AI buildout at $3.5 trillion, spanning vendor credit, sovereign wealth commitments, and off-balance-sheet construction deals. Separately, analysts estimate AI and energy infrastructure projects could generate $200 billion in new commercial insurance premiums over the next 24 months. The buildout is so large it's creating new financial instruments just to move the money around.
Data center investment in Australia hit a decade high in Q3, driven entirely by hyperscaler expansions and AI inference capacity. Microsoft, speaking at a rural infrastructure forum, emphasized that new facilities must be "good neighbors"—code for: don't blow out the local grid, don't piss off the zoning board, and don't assume the water supply is infinite. The AI boom is now large enough that it has a public relations problem in towns that have never heard of a transformer (the electrical kind or the neural kind).