How China Wins The AI War

SUMMARY

Ed Elson interviews Alice Han, Director at Greenmantle and co-host of the China Decode Podcast, to analyze the evolving AI competition between China and the United States. The discussion covers DeepSeek’s $50 billion valuation, China’s semiconductor ambitions, and the broader implications for global security and society.

MAIN POINTS

• DeepSeek, a leading Chinese AI startup, targets a $50 billion valuation with significant state-backed investment, aiming to build a full-stack AI ecosystem to rival the U.S. [1:00]
• Huawei’s new Asend 950 PR chip outperforms Nvidia’s China-approved H20 on inference tasks, but still lags behind Nvidia’s top-tier chips, highlighting China’s focus on domestic chip development. [6:00]
• China holds upstream advantages in rare earths and electricity, while the U.S. leads in data center infrastructure and AI models, with both countries competing intensely at the application layer. [13:00]
• AI is increasingly viewed through a national security lens, with parallels drawn to nuclear arms and calls for strategic dialogue between Beijing and Washington on AI arms control. [17:01]
• Recent studies suggest AI and digital technologies may be contributing to declining cognitive skills among younger generations, raising concerns about the societal impact of widespread AI adoption. [24:00]

DETAILED ANALYSIS

The conversation opens with a look at DeepSeek, a prominent Chinese AI startup seeking a $50 billion valuation in its first fundraising round, a figure that appears modest compared to U.S. counterparts like Anthropic and OpenAI. Alice Han explains that this lower valuation reflects the structure of Chinese capital markets, which traditionally assign lower forward price-to-earnings ratios and rely more on state-led investment than the U.S. model of private fundraising. DeepSeek has garnered attention for its R1 and V4 models, the latter boasting a one million token context window, placing it on par with leading U.S.

AI models in certain benchmarks.

A central theme is the technological and geopolitical contest over semiconductors. Han details how Huawei’s Asend 950 PR chip, now powering DeepSeek V4, has become a focal point for major Chinese tech firms. This chip reportedly delivers nearly three times the inference performance of Nvidia’s H20, the export-restricted chip designed for the Chinese market.

However, Huawei’s chips still trail Nvidia’s most advanced offerings, such as the H200 and Rubin, which remain inaccessible to China due to export controls. The Chinese government’s priority is to foster a domestic ecosystem for inference chips, aiming to reduce reliance on U.S. hardware. Despite these efforts, China’s domestic chip production currently meets only a fraction of demand, with projections indicating it may reach just 2% of Nvidia and TSMC’s compute output by 2027.

Political imperatives, rather than market efficiency, are driving Beijing’s push for self-sufficiency.

The discussion broadens to the multifaceted nature of the AI race. Han introduces a five-layer framework: rare earths, energy, data center infrastructure, AI models, and applications. China dominates upstream in rare earths and electricity, both critical for semiconductor manufacturing and data center operations.

The U.S., meanwhile, leads in data center infrastructure and the development of advanced AI models, though Chinese models are rapidly improving through techniques such as model distillation. At the application layer, both countries are innovating, with China producing notable agentic AI applications that have attracted international attention.

Security concerns loom large, with the AI race increasingly likened to the nuclear arms competition of the Cold War. Han references Henry Kissinger’s advocacy for a strategic dialogue on AI arms control, noting that both the U.S. and China recognize the potential for AI technologies to disrupt national security. Recent developments, such as Anthropic’s Mythos AI and its purported cybersecurity capabilities, have heightened anxieties in Beijing.

The conversation underscores the need for high-level diplomatic engagement and expert involvement to manage risks and avoid escalation, echoing Kissinger’s warnings about the dangers of insufficient dialogue.

In the latter part of the episode, attention shifts to the societal impact of AI. Citing research by Dr. Jared Horvath and analysis by Mia Sario, the hosts highlight a decline in cognitive skills among younger generations, with AI and digital technologies implicated as contributing factors.

Data shows falling scores in math, science, and literacy across high-income countries, and studies suggest that reliance on AI tools can significantly suppress brain activity, with effects that may persist even after discontinuing use. While AI promises productivity gains, concerns are mounting about its broader implications for education and cognitive development.

LINKS

• Prof G Markets Tour tickets and dates information
• Prof G Markets YouTube channel
• Prof G Markets newsletter subscription
• "Notes On Being A Man" book order page
• No Mercy / No Malice YouTube subscription
• Prof G Markets Instagram
• Scott Galloway Instagram
• Ed Elson Instagram
• Ed Elson X (Twitter)
• Ed Elson Substack
• Mia Sario's Substack, including the post 'Is AI making us dumb?'
• LinkedIn ads promotional page for Prof G Markets

KEYWORDS