AI Notebook] Kai Fu Lee On Global AI Race

EJ's note:

I recently listened to Mishal Husain’s conversation with Kai-Fu Lee, and what stayed with me most was how he ended on something deeply human: kindness and heart. In a conversation largely about AI, work, competition, and the future, that felt like the most important reminder.

I’ve been following Kai-Fu since ChatGPT took off in 2023, and I find his perspective particularly worth listening to because he has a rare vantage point across both the US and China. He has spent decades working inside the technology ecosystems of both countries and continues to build and invest in AI businesses, giving him a perspective that is less theoretical and more grounded in how AI is actually developing and being adopted on both sides.

I don’t necessarily agree with every prediction he makes, but his views on the US–China AI race, the future of work, leadership, and what will remain uniquely human are fascinating to think about.

Below is a summary of the conversation, put together with the help of AI, highlighting the ideas I found most interesting and worth revisiting.

This is a detailed summary of Mishal Husain's interview (Bloomberg's The Mishal Husain Show) with Kai-Fu Lee — AI pioneer, former Apple/Microsoft/Google executive, venture capitalist, and founder of the Chinese AI firm 01.AI. Lee, born in Taiwan and sent to the US at age 11, now lives in Beijing and has a new book out, AI Native: The Mandate to Transform Your Company.

1. What's still underappreciated about AI: Lee names two things. First, the speed of improvement and collapse in cost: AI can now complete tasks roughly ten times longer than it could a year ago (a task that would take a human 40 minutes versus 4 minutes previously), which viewed skeptically means it can potentially displace ten times more human work. He argues adoption will outpace the steam engine, the internet, and Moore's Law. Second, the organizational consequences: CEOs are, in his view, among the least aware of how transformative this is. Within five years, successful companies' org charts and the qualifications of their key people will look very different, because AI workers are getting better, cheaper, and faster — and they don't need hierarchies built to manage humans.

2. The skills and people who will matter: The key future skill is learning to solve hard problems, formulate new ones, and "command armies of AI" to solve them in parallel — which, he stresses, does not require coding or an engineering background; a humanities student can do it. More important are soft traits, embodied in what Steve Jobs and Jack Dorsey called the DRI (Directly Responsible Individual): singular accountability (the buck stops with you, since AI can't be accountable), tenacity and ownership, and the ability to work with people and processes.

3. Job displacement and its social stakes: When Husain worries her podcast team could shrink to just her plus AI colleagues, Lee clarifies he isn't predicting zero humans — perhaps a unit of 20 people and 100 AIs, scaling people up or down with the business. But he cites a real one-person company that uses AI to edit and promote videos for handymen and gardeners — work that once would have needed about 20 people. Husain connects this to recent protests in India over vanishing entry-level jobs (e.g., call centers). Lee argues society must rethink its dependence on jobs: AI will generate enormous wealth, and wealthier countries should redistribute it so people can work fewer hours and be paid for activities not previously valued economically — while conceding this is cold comfort to someone who just lost a job. He's been warning of displacement since his 2018 book and says urgency has recently spiked.

4. The US–China AI race: iPhone vs. Android: Chinese labs like DeepSeek and Moonshot (maker of Kimi) pose a significant challenge to OpenAI and Anthropic. The American leaders stay first or second on quality but sell closed models at high prices; Chinese models are largely open source and trail by roughly six months (historically three to fifteen). His central analogy: OpenAI and Anthropic built the iPhone; Chinese companies are Android — nearly as good, sold near-free to win market share. Asked who profits, he's unambiguous: the American companies will make more money, thanks to the US enterprise software market, while Chinese models gain footprint but often get paid nothing (users sometimes copy the model and pay only for hosting). He thinks Chinese models becoming "AI for the developing world" is a likely outcome, with two caveats: a Silicon Valley breakthrough could stretch the gap to two or three years, and there's currently no strong "carrier" building localized apps and interfaces for developing-country consumers, since there's little money in it.

5. How China advanced despite chip export controls: US export controls underestimated Chinese engineers' tenacity. With perhaps 1–3% of the GPU power of top American rivals, they achieved near-comparable results through unglamorous "dirty work" — maybe ten times the engineering effort. He estimates the US lead shrank from three to four years at ChatGPT's launch to about six months today.

6. Why Chinese companies open-source: Because they don't believe they'd win a closed-model race. His metaphor: Silicon Valley labs are like geniuses who feel destined for a Nobel Prize; Chinese companies are strong students who form a "study group" — publishing papers, releasing open models, citing each other (DeepSeek references Kimi and vice versa), with talent circulating between firms — all "in a fishbowl for the world to see," while American labs have stopped publishing.

7. Personal history and the two education systems: Sent to Tennessee at 11 on his brother's advice, Lee lived with his brother and sister-in-law, scientists at a US national lab. He contrasts Asian schooling (training everyone to be a "good student," instilling discipline and tenacity) with American schooling (letting people become what they independently want to be, giving him free thinking and confidence) — and credits his success to the combination.

8. Cancer, family, and rethinking priorities: Around his 2018 book AI Superpowers, Lee faced his own cancer and his mother's dementia. A lifelong workaholic, he was struck that his family gave him unconditional care despite his minimal time with them, and a Buddhist monk in Taiwan warned him that wanting to "change the world" can become a self-serving excuse. After remission he rebalanced toward family — and realized many people, like his former self, define themselves entirely by their jobs, which he calls extremely dangerous as AI upends work.

9. Advice to CEOs: Don't transform too fast: rapid upheaval destroys institutional memory AI can't yet replace. Companies have a responsibility to train people early for emerging roles — even if they end up elsewhere (he references Meta's layoffs). And pragmatically: if every firm ruthlessly cuts people, who's left to be the consumer? An economy needs both suppliers and buyers.

10. Ethics, surveillance, and facial recognition: His firm doesn't invest in surveillance companies, only civilian applications. He notes new Chinese regulations requiring consent for facial recognition (citing his own bank visit — though Husain points out he couldn't realistically refuse; he concedes he could only change banks). On street cameras, he says they've driven crime down dramatically with near-zero murders, footage isn't sold or disclosed, and Chinese people accept that trade-off — while acknowledging it may not hold elsewhere.

11. How Lee uses AI personally: He runs an "AI for the CEO" that surfaces early risk signals, catches product problems, and identifies unsung heroes. Most strikingly, it serves as an always-present management coach that attends every meeting and analyzes six months of longitudinal data, backing critiques with evidence chains he can inspect. Two examples: it told him that stress had pushed him from his usual gentle style into harshly criticizing people publicly — problematic in a face-conscious Chinese company — and that 93.6% of his meeting comments were detail-seeking questions for his own benefit rather than the strategy his team actually needed to hear. He trusts it because it's objective, data-rich, and has no incentive to flatter or bill hours.

12. AI-first hardware: He predicts small, speech-driven, always-on, always-listening devices (he shows a lapel-pin-sized recorder — currently only on during meetings) will become hugely popular, since phones are too slow for voice-first interaction. Transcription happens on your phone rather than leaking to company servers, and the hardware will be "invisible" within two years. He believes Asian cultures will accept these devices for their benefits, but Western societies probably won't.

13. Closing: AI's thinking plus human love: Revisiting his 2018 formulation, Lee distinguishes two meanings of "love." First, human connection — he predicts the largest-growing job sector will involve human-to-human touch (foster care, home help, entertaining guides), because even if AI mimics connection, people don't want it from a machine. Second, passionate love of an idea: early AI scientists persevered when the field was a joke; Steve Jobs created the iPhone with no data saying people wanted it; he cites Coco Chanel and Elon Musk similarly. The partnership he envisions: humans supply the data-free creative conviction they'd "bet their lives on," take the credit and the consequences, and AI does the execution.