Alibaba Chair Joe Tsai: A Breakdown of the US vs. China AI Race

Economic Power: Kardashev Scale & Energy

• 💡 The Kardashev Scale, originally for alien civilizations based on energy harnessing, is proposed as a superior metric for Earth economies over GDP.
• 📊 China leads the US on this scale (0.606 vs. 0.57), generating 2.5 times more electricity annually.
• ⚡ China added 430 TWh of electricity last year (mostly solar), significantly more than the US's 110 TWh, and plans to bring nine times more energy online.

US Open Source AI Contributions

• ✅ Joe Tsai's assertion that the US lacks open-source AI is disputed, as OpenAI is not representative of the entire US market.
• 🧠 Companies like Google (Flan T5) and Meta (Llama family) are major contributors to the open-source AI ecosystem.
• 💡 These US tech giants are playing both sides, developing both proprietary and open-source models.

Value Creation Models in AI

• 🔄 The US AI market (e.g., Sam Altman, Jensen Huang) focuses on value creation at the bottom layer (hardware, proprietary AI).
• 🎯 China's approach inverts this pyramid, prioritizing consumer-level AI penetration and intense competition to drive down hardware margins.
• 🌱 The Chinese government's mandate for AI success is based on its widespread application and benefit to the people.

Circular Financing Risks in AI

• ⚠️ The relationship between Nvidia, Microsoft, and OpenAI involves substantial circular financing, where investments are recycled into service purchases.
• 🔍 While these deals are public and GAAP compliant (unlike Enron), a failure by OpenAI could severely impact companies like Nvidia and Oracle.
• 💰 Oracle faces significant risk, with $4.2 billion in credit default swaps placed against it due to its reliance on OpenAI's success.

Debunking Chinese LLM Cost Myths

• ❌ Reports of Chinese LLMs like Kimi K2 costing only $4.6 million to train are misleading, as this figure only covers the final training run.
• 🔬 The actual cost includes extensive research, experimentation, and failed runs, which are not accounted for in the reported figures.
• 📈 Chinese models are indeed more efficient due to US export controls, but the overall development costs are much higher than publicly stated.

The Global AI Talent & Language Shift

• 🌍 50% of the world's top AI researchers are graduates from Chinese universities.
• 🇺🇸 Major US tech companies, including Alphabet, have historically been significant employers of these Chinese AI graduates.
• 💬 Due to the high number of Chinese engineers, Mandarin is becoming the primary language for AI development, even within American companies.