XPeng Profits, Rolls Out Physical AI
💡XPeng's profitable physical AI stack eyes Robotaxi/humanoids; R&D rivals Big Tech
⚡ 30-Second TL;DR
What Changed
Q4 revenue 222.5B RMB (+38.2%), first profit 3.8B RMB; full-year 767.2B RMB (+87.7%)
Why It Matters
XPeng's profitability enables sustained physical AI investment, challenging rivals like Nio and Xiaomi in embodied intelligence amid maturing EV market. This could accelerate commercialization of Robotaxi and robots, boosting global AI auto share.
What To Do Next
Test XPeng VLA API on Ultra models for embodied AI benchmarking against MindVLA-o1.
Key Points
- •Q4 revenue 222.5B RMB (+38.2%), first profit 3.8B RMB; full-year 767.2B RMB (+87.7%)
- •Second-gen VLA rollout started March 19 on P7/G7 Ultra, expanding in April
- •R&D spend 94.9B RMB (+47%), cash reserves 476.6B RMB; overseas deliveries +96% to 45K units
- •Physical AI stack: chips, OS, hardware; plans L4 autonomy, Robotaxi, humanoid mass production
🧠 Deep Insight
Web-grounded analysis with 6 cited sources.
🔑 Enhanced Key Takeaways
- •The Q4 2025 profit was heavily bolstered by high-margin 'Technical R&D services' for Volkswagen, which yielded a 68.2% margin compared to the 12.8% margin on vehicle sales, effectively de-risking the core automotive business.
- •XPeng's in-house 'Turing' AI chip is the industry's first SoC designed for simultaneous application in cars, robots, and flying vehicles, featuring a 40-core processor capable of running 30-billion-parameter models locally without cloud dependency.
- •The 'IRON' humanoid robot utilizes a triple-Turing-chip configuration delivering 2,250 TOPS of compute and features 82 degrees of freedom (including 22 in each hand), allowing it to share the same VLA 2.0 neural backbone as XPeng's vehicles.
- •To address slowing pure-EV growth, XPeng pivoted to a 'dual-energy' strategy in early 2026, launching the G7 Super EREV (Extended Range Electric Vehicle) which achieved a record-breaking CLTC combined range of 1,704 km.
📊 Competitor Analysis▸ Show
| Feature | XPeng G7 Ultra | Tesla Model Y (Juniper) | Huawei AITO M9 |
|---|---|---|---|
| AI Architecture | VLA 2.0 (Vision-Language-Action) | FSD v14 (Pure Vision) | ADS 3.0 (Lidar-Fusion) |
| Compute Power | 2,250 TOPS (3x Turing Chips) | ~720 TOPS (HW 4.0/5.0) | ~500+ TOPS (Ascend) |
| Charging Tech | 800V / 5C (10-80% in 12m) | 400V / 250kW (V3/V4) | 800V / 4C (10-80% in 15m) |
| Starting Price | ~225,800 RMB | ~249,900 RMB | ~469,800 RMB |
| Unique Feature | Integrated Humanoid Control | Global Data Flywheel | HarmonyOS Ecosystem |
🛠️ Technical Deep Dive
Detailed technical specifications for XPeng's Physical AI stack:
- Turing AI SoC: 40-core processor with dual proprietary NPUs and dual ISPs; optimized for transformer-based end-to-end models with 100% utilization of raw compute power.
- VLA 2.0 Model: A unified Vision-Language-Action transformer that processes raw visual tokens directly into control signals, removing the traditional 'perception-planning-control' modular lag.
- Tai Chi AI Chassis: A predictive suspension system that adjusts damping in 300ms based on VLA road-surface reasoning, reducing vertical impact peaks by 49%.
- 857 Powertrain: Next-gen architecture featuring an 800V SiC platform, 5C-rate charging batteries, and a 93.5% CLTC efficiency coaxial motor.
🔮 Future ImplicationsAI analysis grounded in cited sources
⏳ Timeline
📎 Sources (6)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
📰 Event Coverage
Weekly AI Recap
Read this week's curated digest of top AI events →
👉Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: 虎嗅 ↗