XPeng's Production Capacity Problem: A Systemic Analysis

Today I'll focus on XPeng's production capacity. In the just-released August sales structure, the L03 numbers missed as expected.

I hold 5% XPeng (was 10%). These two months, watching GX and L03 orders explode but cars not getting delivered, hasn't felt good.

GX launched May 20: 12-hour deposits 24,863, Ultra flagship over 80% of mix. The pure-electric flagship's wait went from 15–17 weeks at launch to 31–35 weeks, and by late June approached 40 weeks — nearly 8 months. L03 is even more extreme: Munich launch July 16, 1-hour deposits 46,859, pure-electric 12–17 weeks, EREV 7–12 weeks.

He Xiaopeng again livestreamed from the factory and set up a "special task force." I say "again" because I've seen this before. Four times in four years, same script.

1. This is the fourth time, not the first.

2023, G6: 40,000+ orders in a month, Max 70% of mix — but XPeng expected only 10–15% LiDAR fitment and stocked low trims. RoboSense LiDAR couldn't keep up; Max delivery dragged to 12 weeks (sport package 17). Some users cancelled and switched to NIO ET5. Stacked orders + slow delivery = inevitable loss.

2024, MONA M03: called "the best production preparation in XPeng history," yet the top trim only delivered after Spring Festival 2025 — actually dragging to May. "If you can't produce, don't sell," owners said.

2025, 2025 G6: capacity constrained again. XPeng's Q4 was its first profitable quarter (+380M), the price being aggressive localization. X9 air springs switched from German Vibracoustic to domestic Baolong (~18,000 RMB saved per car) — but in July 2026 the X9 recalled 33,473 units for heat-related breakdown. Industry analysis points to cost-cutting compressing validation cycles.

2026, GX + L03 both blow up at once. GX is stuck on ternary-battery and Bosch Huayu steer-by-wire ramp-up; L03 says global AI demand squeezed high-end foundry capacity for its self-developed Turing chip.

Every time: orders beat expectations → key parts shortage → delivery stretches → order loss. That's a systemic problem, not luck.

2. The root cause isn't "not enough factories."

XPeng's three domestic bases are all 200k/year now (Zhaoqing, Guangzhou Huangpu, Wuhan) — 600k/year combined. The problem is deeper. Three of them:

Demand forecasting is systematically off. G6 underestimated Max mix (expected 10–15%, actual 70%); GX underestimated Ultra mix (80%+). Same mistake twice, three years apart. RoboSense later said suppliers follow XPeng's conservative forecasts conservatively — when orders flood, supply breaks. A two-way trust problem.

Supply-chain vertical integration is too low. Chips: old models use NVIDIA Orin-X, 2026 models switch to the self-developed Turing chip — but still fabbed at a foundry. Batteries: mainly CALB (lead), EVE (supplement), BYD FinDreams (MONA). Steer-by-wire: Bosch Huayu. Core 3-electric + chips are almost entirely outside its hands — unlike BYD, where battery, SiC, motor, and controller are all in-house.

A key turning point: the CATL breakup. Early on ~70% of XPeng batteries were CATL. A 2021 capacity-quota dispute broke trust, and 2022–2025 CATL share fell to near zero. Late Dec 2025, CATL's Robin Zeng led a team to Guangzhou to meet He Xiaopeng — their first formal interaction in four years. Whether they reconcile, I'm watching.

Organizationally it's constant firefighting, no systemic capability. Every crisis is solved the same way: CEO personally supervises, special teams, new molds, double shifts. Sina's May field report: Huangpu South's June GX target was 7,000 units, "a death order — if not done, work overtime." Temporary measures, not systemic capability.

Two more I won't expand: factory/product-timing mismatch (high-end cars all crammed into one plant), and supplier trust crisis (a vicious cycle of bad forecast → distrust → no stock → shortage → loss → more distrust).

Of the five reasons, demand-forecast failure is the trigger, low supply-chain integration is the structural disease, and the rest are complications.

---

Source: 姬永锋 (Weibo) — translated & adapted for XPENGTech.