CEO of Sinovation Ventures, Sinovation Ventures
Kai-Fu Lee: AI Superpowers – China and Silicon Valley
Kai-Fu Lee is the Chairman and CEO of Sinovation Ventures that manages a 2 billion dollar dual currency investment fund with a focus on developing the next generation of Chinese high-tech companies. He is the former President of Google China and the founder of what is now called Microsoft Research Asia, an institute that trained many of the AI leaders in China, including CTOs or AI execs at Baidu, Tencent, Alibaba, Lenovo, and Huawei. He was named one of the 100 most influential people in the world by TIME Magazine. He is the author of seven best-selling books in Chinese, and most recently the...
About Kai-fu Lee
Kai-fu Lee, CEO of Sinovation Ventures and founder of 01.AI, commented on the AI landscape at the 2026 World AI Conference in Shanghai. He described the Kimi K3 model as "excellent" and said his company would use it in its products. Lee stated that 01.AI is preparing for an initial public offering in 2027, saying its "financial numbers are ready right now, but it's a matter of going through the process." He argued that U.S. export controls on GPUs have "failed" to contain China, attributing any GPU shortage to business frugality or insufficient supply rather than the controls themselves. Lee predicted that in the AI model race, the U.S. will make more money while China will have larger market share due to open-source distribution and enterprise demand for on-premise deployment. He cited Chinese government support and high consumer optimism—noting a poll showing 84% of Chinese view AI positively versus under 50% in the U.S.—as strengths, while acknowledging challenges in getting enterprises to pay substantial fees for AI. In earlier remarks from 2018, Lee discussed AI's impact on employment, stating that routine white-collar jobs are more vulnerable to automation than blue-collar roles due to the difficulty of robotics. He described the Chinese approach to tech markets as "do whatever it takes to win" in a winner-take-all environment, with companies aiming for domination of increasingly broad product categories. Lee also noted that AI technologies have become more accessible through open-source tools, advising a shift from a laboratory mindset to a business-oriented approach, except for those with genuine technological breakthroughs.
Transcript (84 segments)
L
Lex Fridman0:00
The following is a conversation with Kai-fu Lee. He is the chairman and CEO of Sinovation Ventures, managing a $2 billion dual-currency investment fund with a focus on developing the next generation of Chinese high-tech companies. He's the former president of Google China and the founder of what is now called Microsoft Research Asia, an institute that trained many of the AI leaders in China, including CTOs and AI execs at Baidu, Tencent, Alibaba, Lenovo, and Huawei. He was named one of the 100 most influential people in the world by Time magazine. He's the author of seven bestselling books in Chinese and most recently the New York Times bestseller AI Superpowers: China, Silicon Valley, and the New World Order. He has unparalleled experience working across major tech companies and governments on AI applications, giving him a unique perspective on global innovation and the future of AI. This is the Artificial Intelligence Podcast. If you enjoy it, subscribe on YouTube and iTunes, support it on Patreon, or connect with me on Twitter at Lex Fridman. And now, here's my conversation with Kai-fu Lee.
I immigrated from Russia to the US when I was 13. You immigrated to the US at about the same age. The Russian people, the American people, the Chinese people each have a certain soul, a spirit that permeates throughout the generations.
K
Kai-fu Lee1:58
Mhm.
L
Lex Fridman1:59
So maybe it's a poetic question, but could you describe your sense of what defines the Chinese soul?
K
Kai-fu Lee2:08
I think the Chinese soul of people today is shaped by centuries of burden due to poverty the country has gone through, and suddenly shining with hope of prosperity in the past 40 years as China opened up and embraced a market economy. There are two sets of pressures: tradition, facing difficult situations, and the hope of wanting to be the first to become successful and wealthy. That creates a very strong hunger, desire, and work ethic that drives China forward.
L
Lex Fridman2:59
And is there something deeper, rooted in older generations, that's unique to China beyond the new economic developments?
K
Kai-fu Lee3:07
Yeah. The Chinese tradition is about excellence, dedication, and results. Chinese education starts with memorizing 10,000 characters, then historic philosophers, literature, poetry. It's a strong rote learning mechanism that builds good memory but also suppresses breakthrough innovation while enhancing speed and execution. That characterizes the historic basis of China.
L
Lex Fridman4:04
That's interesting. There are echoes of that in Russian education as well with rote memorization of poetry.
K
Kai-fu Lee4:12
An emphasis on perfection in all forms.
L
Lex Fridman4:16
That's not conducive to creativity. Do you think that kind of education holds back the innovative spirit you see in the United States?
K
Kai-fu Lee4:28
It holds back the breakthrough innovative spirit seen in the US, but it does not hold back the valuable execution-oriented, result-oriented value-creating engines that make China very successful.
L
Lex Fridman4:45
So is there a difference between a Chinese AI engineer today and an American AI engineer, rooted in culture or education? What advice would you give to each?
K
Kai-fu Lee5:02
There's a lot that's similar. AI requires mastering sciences and tuning parameters. The American strength is trying new things and using technology to solve problems despite imperfect data. The Chinese approach enumerates through all possible ways, uses many machines, and spends resources cleaning data. The Chinese engineer relies on more data and cleansing, while the American thinks of new algorithms to overcome data issues.
L
Lex Fridman6:56
Where do you think the biggest impact in the next 10 years lies? In breakthrough algorithms or in rigorous data cleaning at scale?
K
Kai-fu Lee7:10
In a company delivering results with known techniques, enhancing data is more expedient, low risk, and likely to generate better results. That's why the Chinese approach has done well. But for challenging problems like autonomous vehicles and medical diagnosis, existing algorithms may not solve them, so the Chinese approach is more challenged, and breakthrough innovation has an edge.
L
Lex Fridman8:02
Let me talk to that a bit more. My intuition is that data can take us extremely far. You brought up autonomous vehicles and medical diagnosis. So your intuition is that huge amounts of data might not solve those problems. Breaking that down, in autonomous vehicles, huge amounts of data probably will solve trucks on highways, which China will lead. But full L5 autonomy likely requires new technology.
K
Kai-fu Lee8:59
I am.
L
Lex Fridman9:00
Where they are full steam ahead into L5 autonomy, trying to solve it purely with data. So the same thing you're saying for highway, a lot of people share your intuition. Do you think it's possible for them to achieve success with just huge amounts of training on edge cases in urban environments?
K
Kai-fu Lee9:21
Yeah.
L
Lex Fridman9:21
They're trying to solve with data. Just to linger on that further, do you think it's possible?
K
Kai-fu Lee9:27
I think it would be very hard. Tesla's approach is kind of a Chinese strength approach: gather all data and hope it overcomes problems. But in autonomous driving, many decisions aren't solved by aggregating data and feedback loops. There are things more akin to human thinking, and integrating expert systems with machine learning hasn't been demonstrated yet. Tesla also lacks lidar, which is a valuable source of input.
L
Lex Fridman10:04
And how would those elements be integrated? There hasn't been much success integrating human intelligence with machine learning. The question is how much you can push a purely machine learning approach. Tesla also has the constraint of not using all sensors.
K
Kai-fu Lee10:48
The advantage is capturing data no one has seen. In some cases, Chinese companies have accumulated data not seen in the West and delivered superior results. But speech and object recognition are suitable for deep learning and don't need the analytical planning elements.
L
Lex Fridman11:21
And on speech recognition, your intuition is that machine learning approaches won't take us to a conversational system that can pass the Turing test, which is akin to driving. It needs something more than simple language understanding and generation.
K
Kai-fu Lee11:44
Roughly right. Based on purely machine learning, it's hard to imagine leading to full conversational experience across arbitrary domains, which is akin to L5. I'm hesitant to use the word Turing test because the original definition was probably too easy. We probably already do that.
L
Lex Fridman12:09
The spirit of the Turing test.
K
Kai-fu Lee12:11
That's what I was referring to.
L
Lex Fridman12:12
Of course. So you've had major leadership research positions at Apple, Microsoft, Google. Continuing on the discussion of culture, what is the culture of Silicon Valley in contrast to China, and the unique culture of each of these three companies?
K
Kai-fu Lee12:40
In aggregate, Silicon Valley companies dream big, have visionary goals, believe technology will conquer all, and have a self-confidence that the world should use their products. That's exemplified by Steve Jobs' quote about not needing focus groups. But it also leads to a belief that companies shouldn't tread on each other's territory. The Chinese approach is to do whatever it takes to win, with a winner-take-all mentality. They are flexible, fast, and execution-oriented. Apple represents pleasing the user with design and brand. Microsoft represents a platform approach. Google is a value-oriented company with heart and soul, wanting to do great things for the world.
L
Lex Fridman14:53
But I think the Chinese approach is do whatever it takes to win. It's a winner-take-all market. The market leader extracts all the value, and the system gets broader and broader. That develops a practical, ultra-ambitious, gladiatorial mentality. If it takes copying, they'll do it without infringing laws. The flexibility and speed have helped the Chinese approach. The Silicon Valley approach is challenged if Chinese entrepreneurs learn from the whole world while American entrepreneurs only look internally and write off China as a copycat.
K
Kai-fu Lee14:53
Yes, and the second part about the three companies: Apple represents pleasing the user with design and brand. Microsoft represents a platform approach that builds giant products by efficiently delegating work. Google is a genuinely value-oriented company with a heart and soul that wants to do great things for the world.
L
Lex Fridman16:40
The unique elements of the three companies perhaps.
K
Kai-fu Lee16:45
Yeah, I think Apple represents pleasing the user, design, brand. Microsoft represents a platform approach. Google is value-oriented with heart and soul.
L
Lex Fridman17:58
And the whole process of doing that is kind of a differentiation that competitors can't easily repeat. Are there elements of the Chinese approach in the way Microsoft assembled those pieces and dominated the market?
K
Kai-fu Lee18:22
I think there are elements that are the same. The three American companies that have Chinese characteristics are Microsoft, Facebook, and Amazon. They tenaciously go after adjacent markets, build strong product offerings, and extract greater value from an ever-increasing sphere. Google is different; it's genuinely value-oriented.
L
Lex Fridman19:40
If we look at Google, you mentioned heart and soul. There seems to be an element of making the world better. They had the slogan 'Don't be evil.' Facebook has a more negative tint in perception of privacy. Do you have a sense of how these companies can achieve heart and soul and trust?
K
Kai-fu Lee20:02
It's really hard. The 'Don't be evil' mantra is dangerous because everyone's definition of evil is different. But watching Google's parent company Alphabet do things like healthcare and eradicating mosquitoes shows a heart and soul. But it's difficult to balance shareholder profit with doing good. You mentioned concern about too few companies controlling our data.
L
Lex Fridman21:59
You've mentioned concern about a future where too few companies like Google, Facebook, Amazon control too much of our digital lives. Can you elaborate? Do you have a better way forward?
K
Kai-fu Lee22:15
I think I'm hardly the most vocal complainer. Having a lot of data perpetuates their strength and limits competition. But AI is much broader than the internet space. Entrepreneurial opportunity still exists in using AI to empower finance, retail, manufacturing, education. I don't think it's full monopolistic dominance that stifles innovation. The best solution is to let the entrepreneurial VC ecosystem work well and find the next Google, next Facebook. China as an environment may be more interesting for emergence of mid-sized companies.
L
Lex Fridman24:39
You've mentioned the fascinating world of entrepreneurship in China, the fearless nature of entrepreneurs. Can you talk about what it takes to be an entrepreneur in China, the strategies, the dynamic of VC funding, and how the government helps?
K
Kai-fu Lee25:12
Many listeners would still brand Chinese entrepreneurs as copycats. 10 years ago, that wasn't inaccurate. An entrepreneur couldn't get funding without describing what product they were copying from the US. That was understandable because China had lower internet penetration and lacked indigenous experience. The lean startup methodology allowed copying as a starting point, then tweaking. Silicon Valley views that as not honorable, but it's not morally wrong. The Chinese approach has evolved: from copying to building better products, then innovating new products like mobile payment, TikTok, and Pinduoduo. The ecosystem is supported by VCs in a virtuous cycle. The government supports infrastructure like incubators and smart highways, and uses guiding funds that act as passive LPs, giving profit to GPS and other LPs.
L
Lex Fridman33:08
Where let me put it this way: there's no such government-driven large-scale support of entrepreneurship in Russia, and probably the same is true in the US. How did the Chinese government arrive to be so supportive? How can we copy it?
K
Kai-fu Lee33:53
Yes. These techniques are the result of trial and error. The guiding funds idea came from Singapore and Israel, and China made tweaks. Local governments compete with each other to make their cities successful, so they experiment. The central government made it a competition. The best approach was the guiding funds. For autonomous vehicles, the Chinese government builds smart highways and cities. They view infrastructure as their responsibility, unlike the West. It's a different way of thinking that may be hard to inject elsewhere.
L
Lex Fridman35:52
Uh the autonomous vehicle and the massive spending in highways and smart cities, that's a Chinese way. It's about building infrastructure to facilitate.
K
Kai-fu Lee35:52
Yes, it's a clear division of government responsibility from the market. The market can't afford infrastructure, so the government appropriates large amounts. This happened with 3G and 4G too. It's a government-driven approach. The US highway system was built during Eisenhower, so it's not impossible for Western countries.
L
Lex Fridman37:33
What's your sense? Do you think it's possible to solve full autonomy without significant investment in infrastructure?
K
Kai-fu Lee37:50
Well, it's hard to speculate. It's not a yes/no question but how long it takes. With infrastructure augmentation, it will accelerate L5. In the US, people probably wouldn't build a new city the size of Chicago, but smaller ones are being built. Infrastructure spend is not impossible for the US.
L
Lex Fridman39:07
If I may comment, building infrastructure creates jobs. In your book, you talk about jobs that can and cannot be automated. Is building infrastructure a job that would not be easily automated? There might be a role for government to create jobs that can't be automated.
K
Kai-fu Lee39:48
Yes, that's a possibility. In the last financial crisis, China put a lot of money into infrastructure to boost the economy and deal with employment. It's a legitimate government approach as long as the infrastructure is truly needed.
L
Lex Fridman40:22
Taking a step back, you've been a leader and researcher in AI for several decades, at least 30 years. How has AI changed in the West and the East as you've observed over these decades?
K
Kai-fu Lee40:42
AI began as the pursuit of understanding human intelligence, but it drifted into machine intelligence, which uses pattern recognition to do incredibly well on limited domains with large data but simple planning. It's not creative. We didn't build human intelligence; we built a machine that's better than us on some problems but nowhere close on others. People still misunderstand AI, thinking it's about replicating human intelligence, but the products are closer to the internet or spreadsheet.
L
Lex Fridman41:52
And speaking to the near-term fears about AI, you're commenting on the general intelligence from sci-fi movies versus the narrow AI that automates particular jobs, which you talk about in the book.
K
Kai-fu Lee42:16
The routine white-collar workers are easiest to replace because you just need software. Blue-collar workers require robotics and dexterity, which is very difficult. Back-office jobs like copying data, new employee orientation, lawsuit searches, and reference checks are most in danger.
L
Lex Fridman44:05
So basic searching and management of data is most in danger of being lost.
K
Kai-fu Lee44:10
In addition, simple interaction jobs like telesales, customer service, and physical jobs like fruit picking and assembly line are at risk. Blue-collar will be smaller initially, but over 15-20 years, dexterity and autonomous vehicles will displace many. Modest numbers in 5 years, increasing rapidly after.
L
Lex Fridman45:15
On the worry of jobs in danger, are you familiar with Andrew Yang?
K
Kai-fu Lee45:23
Yes, I am. I think his thinking is generally in the right direction, but his approach may be ahead of the time. Displacements will happen, but unemployment is low. I agree on displacement but disagree with simple UBI; retraining is key because AI replaces all routine jobs. The goal is not just money but meaning.
L
Lex Fridman48:55
So what kind of jobs can't be automated?
K
Kai-fu Lee49:10
AI cannot reason, plan, or think creatively. It needs humans to pose problems. Compassionate jobs like healthcare, elderly care, teaching, and bartending require human touch. AI creates $16 trillion in value, leading to more service jobs, but there will be a shortage due to low pay. Retraining, especially for healthcare, is essential.
L
Lex Fridman53:09
Which will be replaced a lot sooner. Right. Very true. So overall are you optimistic?
K
Kai-fu Lee55:37
I am optimistic if we take action in the next 5 years: retraining, vocational schools, government subsidies. This revolution is similar to electricity or the internet.
L
Lex Fridman56:38
Do you think there's a role for policy?
K
Kai-fu Lee56:44
Absolutely. Revamp vocational schools, incent training for elderly care given the aging population.
L
Lex Fridman57:46
Do you have concerns about large entities controlling AI development?
K
Kai-fu Lee58:12
No easy answer. Companies and governments can provide better services but also lead to monopoly or corruption. Checks and balances are needed.
L
Lex Fridman58:47
So again I come from Russia. Do you see international conflict over AI?
K
Kai-fu Lee59:36
I believe in greater engagement. Cold war mentality is dangerous. Protocols for interaction are needed. The level of engagement is decreasing, distrust increasing, especially from the US toward China and Russia.
L
Lex Fridman1:01:50
Is there a way to make that better?
K
Kai-fu Lee1:02:18
We look foolish having the opportunity to create $16 trillion while not solving poverty. Idealistically, a benevolent world government could use it to deal with disease and inequality, but competition and wealth disparity are real issues. Poor countries will suffer as outsource jobs disappear.
L
Lex Fridman1:05:07
Different countries have different value systems. Does censorship affect innovation in China?
K
Kai-fu Lee1:06:12
Empirically, the correlation between free speech and innovation is not perfect. China has been successful. Fundamental values like privacy are there, but the timing of opening up is unclear. China's privacy laws are similar to others.
L
Lex Fridman1:07:53
On the point of privacy, how do we balance data use and user experience?
K
Kai-fu Lee1:08:42
People misunderstand: taking back all data breaks services. There should be severe punishment for misuse. Technology like homomorphic encryption and federated learning can help. User choice with transparency is important, but not granular like GDPR. A slider with AI customization could work.
L
Lex Fridman1:12:42
And I think getting that right requires balancing heart and soul versus profit-driven decisions.
K
Kai-fu Lee1:13:19
Yes, absolutely. Long-term projects that do the right thing. Utopian ideas like a trusted data platform are exciting but hard to fund.
L
Lex Fridman1:14:52
So how do you do that? You've recently had a fight with cancer. What did it feel like to face your mortality?
K
Kai-fu Lee1:15:11
I was a workaholic, working 9am to 9pm six days a week, not paying attention to family. Facing death made me realize that accomplishments meant nothing. I've resolved to live a balanced life, prioritizing love and family. I still work hard but drop everything when family needs me.
L
Lex Fridman1:19:02
That's profound. But the ability to become obsessed and passionate is a beautiful part of human nature. Would you take back some of your hard work?
K
Kai-fu Lee1:20:32
It's about percentages. When younger, you give a smaller percentage to family but still high priority. When family has a need, be fully present. That priority doesn't drastically reduce accomplishments; it may even reduce conflict. I would take that reduction for greater happiness.
L
Lex Fridman1:22:27
Given your success, what advice would you give to young people launching an AI startup?
K
Kai-fu Lee1:22:52
Understand technology waves. AI is moving from rocket science to mainstream. Focus on business value, customer pain, and valuation expectations. Use open-source tools like TensorFlow and PyTorch. Only if you have a genuine breakthrough does the old lab-to-company model still work.
L
Lex Fridman1:25:54
We're far from AGI, but if you could ask it one question, what would it be?
K
Kai-fu Lee1:26:14
What is it that differentiates you and me?
L
Lex Fridman1:26:19
Beautifully put. Thank you so much for your time today.