AI Superpowers — China, Silicon Valley, and the New World Order

The China Thesis

Kai-Fu Lee’s central argument is one that was controversial in 2018 and remains contested: China is not merely catching up to the United States in AI — it is already competitive and in some significant respects ahead. The reason, Lee argues, is not that China has better researchers or better algorithms. It is that China has advantages in exactly what the current AI era requires most: data, implementation speed, and a willingness to deploy AI aggressively at scale.

Lee should know. He ran Google China, has led major AI research groups, and has spent decades investing in Chinese technology companies from his position at Sinovation Ventures. He is not making an argument from the outside; he is describing a world he has operated in for years.

The book’s foundational distinction is between what Lee calls “research AI” — the development of new algorithms, new architectures, new fundamental capabilities — and “implementation AI” — the deployment of existing AI capabilities in real products serving real users at scale. In research AI, the United States has clear advantages, rooted in its universities, its research culture, and its ability to attract global talent. In implementation AI, Lee argues, China has structural advantages that are decisive.

Why China’s Implementation Advantages Matter Now

The current era of AI development is an implementation era. The fundamental techniques — deep learning, transformer architectures, large-scale training — are broadly understood and widely available. What differentiates winners from losers in this environment is not who discovered the techniques but who deploys them most effectively.

China’s advantages in this environment are specific and real. The scale of its internet user base provides training data that Western companies cannot match. Its entrepreneurs have a willingness to iterate and deploy aggressively that Silicon Valley’s culture of perfectionism often limits. Its government is aligned with rather than sceptical of technology deployment, providing both regulatory tailwinds and direct investment.

Lee is careful to present this as an analysis, not a prediction. The future of the AI competition between the US and China is contingent on decisions yet to be made — about export controls, about regulatory approaches, about educational investment, about immigration. His argument is not that China’s AI dominance is inevitable but that its current trajectory, if maintained, points in that direction.

The Geopolitical Consequences

The second major thread of AI Superpowers concerns what US-China AI competition means for the global order. Lee argues that AI will contribute to a bipolar world in which the two AI superpowers — the US and China — extend their technological influence through different models: American AI spreading through market-based global technology companies, Chinese AI spreading through infrastructure investment and technology export agreements.

This analysis has proven prescient in outline, if more complicated in specific than Lee anticipated. The subsequent development of US export controls on advanced chips, China’s tech regulatory crackdowns, and the unexpected competitive capability demonstrated by Chinese AI companies like DeepSeek have all complicated the picture in ways that make the 2018 analysis require updating.

Jobs and the Human Cost

The third major theme is automation displacement: the removal of jobs by AI, and the economic and political consequences. Lee offers specific estimates of which jobs are most susceptible — routine cognitive tasks, customer service, manufacturing supervision — and which are less so — creative work, complex interpersonal interaction, caregiving. His estimate that 40-50 percent of current jobs face serious automation risk within the next fifteen years remains contested but impossible to dismiss.

The political dimension of this displacement is Lee’s most uncomfortable point: the democratic political systems of the US and Europe were not designed to manage the kind of rapid, concentrated job displacement that AI-driven automation may produce. The policy frameworks — education systems, social insurance programmes, labour market regulations — are calibrated to a pace of change that AI is already exceeding.

The Personal Epilogue

AI Superpowers contains a section that surprised many readers who came for the geopolitics and stayed for the argument: the account of Lee’s diagnosis with lymphoma in 2013 and what it taught him about the limits of the analytical framework he had spent his career developing.

Cancer forced Lee to reconsider what he had been optimising for. The work that seemed most important from within his professional framework — the largest impact, the biggest investments, the most significant technological bets — seemed different in the context of his own mortality. The question of what makes a human life valuable, and how AI relates to that question, became personal rather than theoretical.

The epilogue argues that the irreducibly human — love, creativity, compassion, the capacity for genuine interpersonal connection — is precisely what AI cannot replicate and what, therefore, a society navigating the AI transition needs most to protect and cultivate. It is a different register from the geopolitical analysis, and not all readers will find the integration satisfying. But it adds a dimension to the book that makes it more honest than most AI analysis manages to be.

Our rating: 4.3/5 — The essential geopolitical account of AI competition. The 2018 vintage requires updating, but the framework is durable.