The magic survival number for Chinese EVs + why NIO/XPeng/Li Auto are dumping NVIDIA. Two stories, one theme: Chinese EVs leveling up.

THE 500K SAFETY LINE:Not 2M. Not 1M. At 500K annual units, Chinese EV makers gain:

• Financial viability (100-150B yuan revenue = profitability)
• Strategic breathing room (absorb 20% drops like Li Auto)
• Operational efficiency (economies of scale kick in)
• Community critical mass (self-sustaining ecosystems)
• Supply chain power (justify vertical integration)

WHO'S WHERE:

• Li Auto: 400K (safe, crossed line 2024)
• Leapmotor: 600K (just crossed, suddenly legitimate)
• XPeng: 430K (+126%), but 41% from budget model—not safe yet
• NIO: 326K, but cost structure SO high needs 600-700K
• HarmonyOS: 590K, Aito 423K, Huawei backing = different game

THE NVIDIA EXODUS:ALL going in-house:

XPeng: G6/G7/G9/P7+ use Turing chip. He Xiaopeng: "Best AI companies develop own chips."

Li Auto: M100 chip = 2 NVIDIA Thor-U (LLM tasks), 3 Thor-U (vision). Delayed products waiting for Thor, learned lesson.

NIO: Spent $300M+ on Orin X in 2024. Shenji NX9031: 2 chips = 8 Orin X, saves 10K yuan per car. Rolling to full lineup.

WHY:

1. Thor disaster: 2,000 TOPS promised, 700 delivered, multiple delays
2. Cost: 10K yuan savings x 500K units = billions
3. Independence: core tech can't be foreign dependency

NVIDIA'S PROBLEM:

• China share: 39% → 25% despite having Orin + Thor
• 1.16M units (NIO+XPeng+Li Auto) moving away
• +HarmonyOS +Leapmotor = 2M+ annual systematically shut out
• Short-term: only 1% revenue
• Long-term: most advanced market building own silicon

THE CONNECTION:Same story. 500K scale enables strategic independence. NVIDIA exodus IS exercising that independence.

At 100K units: must use NVIDIA.At 500K units: $200-300M chip development pays back in 2-3 years.

Autonomous driving = core differentiator. Why outsource it?