Is China catching up with the US in AI? With Kyle Chan

Episode Summary

Executive Summary: The episode argues that China is rapidly closing the gap with the US in frontier AI, but its strategy is less about AGI and more about practical adoption, industrial productivity, and global diffusion. Kyle Chan says China is strong in model performance, energy buildout, and commercialization, while still constrained by chips, data, and parts of the AI supply chain. The US and China may talk cooperation, but competition will remain intense.

Main Topics: China’s AI strategy: adoption over AGI (Priority: 5/5): Beijing is portrayed as highly enthusiastic about AI, but focused on using it to improve manufacturing, robotics, and economic productivity rather than pursuing an AGI-driven transformation. Rapid narrowing of the US-China model gap (Priority: 5/5): The discussion highlights how Chinese frontier models have moved from being far behind US rivals to being only months behind, with Chinese labs now seen as serious global contenders. Private-sector-led ecosystem with strong local government support (Priority: 4/5): Chinese AI progress is driven mainly by private firms such as Alibaba, Tencent, ByteDance, DeepSeek, and startups like Moonshot, while cities like Beijing, Shanghai, Hangzhou, and Shenzhen compete to become AI hubs. Open source as strategy and commercial pressure (Priority: 4/5): Chinese companies have leaned heavily on open-source model releases to gain adoption and global visibility, but may shift toward more closed systems as they seek revenue and face policy/security concerns. China’s strengths and bottlenecks in the AI stack (Priority: 5/5): China is said to have a major advantage in energy and grid expansion, but still faces constraints in top-end chips, high-quality data, memory, and other infrastructure layers. US reaction: market share, cyber risk, and strategic leadership (Priority: 4/5): Washington is worried about Chinese models undercutting US AI revenues, weakening American tech leadership, and enabling cyber capabilities through widely available open-source models. Limited but possible US-China AI dialogue (Priority: 3/5): The speakers see upcoming talks as useful mainly for confidence-building and narrow cooperation on non-state threats, not as a path to a major AI arms-control agreement.

Key Arguments: China is not primarily chasing AGI; it is trying to make AI useful in the real economy, especially in manufacturing, robotics, and industrial automation. Chinese AI models are now strong enough that the global race is effectively China versus the US, with little room for other players. Local governments in China are deeply involved, competing to attract talent, labs, and companies, which accelerates regional AI development. The private sector, not state-owned enterprises, is the main force behind Chinese AI innovation and commercialization. Open source has been a major Chinese advantage for adoption and international reach, but commercial incentives and security concerns are pushing toward more restrictions. China’s energy buildout gives it an advantage in powering data centers, a major bottleneck in the US. China still lags in critical inputs like advanced NVIDIA chips, high-quality labeled data, memory chips, and parts of the broader AI infrastructure supply chain. US policymakers are concerned that cheaper Chinese models could take market share from US labs and firms and expand cyber risk. There is real US demand for Chinese models because they can be cheaper, more controllable, and customizable. A September US-China meeting may produce limited discussion on shared concerns like malicious non-state actors, but not a sweeping AI treaty.

Data Points: China model gap vs US: as little as 4 months - FT reporting cited in the opening on how quickly Chinese models are catching up to leading US models China model gap vs US: 6 to 10 months behind - UK AI Security Institute estimate for Chinese models in 2025 before further narrowing China frontier AI score: 9/10 - Kyle Chan’s rating of China’s current position in frontier AI development DeepSeek moment improvement: from 1-2 to about 6 - Chan described the jump in China’s AI standing after DeepSeek’s breakthrough more than a year ago US data center investment: on the order of $1 trillion - Referenced as the scale of projected US investment in AI infrastructure China power grid expansion: about 40% of the entire US power grid - Amount China reportedly built out in a single year, highlighting its energy scaling capacity Renewable share of new energy capacity in China: 80% - Last year’s new energy capacity in China was said to be mostly renewable, especially solar and wind Local governments mentioned as AI hubs: 4 cities - Beijing, Shanghai, Hangzhou, and Shenzhen were identified as competing AI centers

Pivotal Quotes: "Beijing is super AI-pilled. They really see AI as this transformative, powerful technology, but they're not AGI-pilled like folks in the U.S. are." — Kyle Chan: Explaining the difference between China’s practical AI posture and the US obsession with AGI "I think they're at a 9. Not quite at the level of US frontier models, but there's basically no one else in the game." — Kyle Chan: Assessing China’s current standing in frontier AI "What would really make the biggest difference in the US is trying to integrate AI more directly into people's lives in a way that's actually productive and beneficial to them." — Kyle Chan: Arguing that the US should emulate China’s emphasis on adoption and diffusion

Implications: China is becoming a credible AI power with real commercial and geopolitical influence. Expect more competition on models, more pressure on open source, and greater focus on energy, data, and deployment. US firms and policymakers may need to prioritize adoption and cost control, not just frontier breakthroughs.