Interview of March 2025: Chinese AI, AI for Cars, Chinese chips

Q1 What kind of transformation will artificial intelligence bring to global transportation? More and more automotive companies are integrating with AI; what are your thoughts on this?

The car will basically become like a train. Trains are already robots, but with the advantage that they run from a station to another station and only on railway tracks. Cars can go from anywhere to anywhere, and they have to deal with pedestrians and bicyclists. When AI becomes good enough to solve all the problems of the road, the car will be a robot like a train. Right now the impact of AI is simply that improves the safety of the driver, but it actually increases the cost because you have to pay extra for Tesla’s autopilot. But at some point the autopilot feature will become standard, just like the radio and the A/C became standard in every car. Then the fully autonomous self-driving car will become commonplace, and then it will change the entire economy because many of us will decide that we don’t need to buy a car anymore. You can just call the nearest self-driving car. No need to pay 30-40 thousand dollars for a car that you keep in your garage. Just call the self-driving car when you need it. When Uber and Lyft came out, they told us the same thing: why buy a car if you can just call Uber or Didi? But people still prefer to drive their own car because they don’t see a big difference between a Uber car driven by a person and their own car driven by themselves. It’s still a car driven by a person. A robotaxi is a fundamentally different experience. It’s a service on demand that doesn’t involve a human. It’s like when you pay to watch a video at home instead of going to the movie theater. This will happen slowly but it will transform not only transportation but the entire economy. And maybe it won’t be for everybody, just like there are still many people who prefer to watch a movie at the theater. I suspect that cost will decrease and safety will increase, and then it will be hard to justify why you really have to drive your own car instead of a robotaxi.

Q2 What is your ultimate vision for the integration of artificial intelligence and the automotive industry? Many people refer to artificial intelligence as the "Fourth Industrial Revolution" - what capabilities do you think companies must possess today to keep up with this technological transformation?

Companies obviously need to think of cars as software, not just hardware, and the winners will probably be those who integrate vertically: make the chips, make the AI software, make the cars, make the communication satellites, and so on. That’s the Tesla model, after all. And companies need to think of a future when people don’t buy cars but use them only when they need them. When Google created Waymo, it basically created this business plan: the car manufacturer becomes a taxi company.

In terms of the hardware, AI requires cameras, radars, lidars, sensors, communication and of course thousands of chips. We think of a car as a body and an engine, but clearly AI changes the definition of what a car is. It is a system of sensors and actuators driven by computers. The traditional car is a machine that controls itself, its interior, while the driver controls the environment outside: the traffic light, the child playing on the sidewalk, the ice on the road, and so on. The car of the future is a car that also has to control its environment.

A big transformation will take place in the infrastructure. At some point there will be a revolution in the way roads are organized. There will be places where only self-driving cars are allowed and these cars will communicate to each other their position and their speed. So the cars will be allowed to avoid each other and minimize the number of times that they stop. Therefore we won’t need traffic lights and traffic signs. These things were invented to avoid that cars hit each other and to make sure that everybody gets a chance at driving through a busy intersection, and safely. But traffic lights and traffic signs are not needed when the cars can talk to each other. Self-driving cars that communicate with other cars don’t even need to drive on one side of the road. They can drive to the right or to the left. They talk to each other, they see each other, they can avoid each other. The lanes may disappear. Of course this can happen only in the places where human drivers are not allowed. The moment you have one human driver, then you need a traffic light, and a stop sign, and a speed limit, and the police to catch those who disobey, and so on. The temptation to create zones where only self-driving cars are admitted will be very strong.

Q3 Globally, it is primarily China and the United States that are leading in AI and new energy vehicles. Can it be understood that in the future, the competition in the AI automotive sector will be between Chinese and American companies?

No, of course no. The German and Japanese car manufacturers have the technology and the money to compete. The first car with Level 3 automation was an Audi in 2019. Mercedes just introduced a Level 3 car in the USA, ahead of Tesla which is still at Level 2. Many innovations of AI originally came from Europe and Canada. Geoffrey Hinton is British and Yann LeCun is French. There are many countries that have top AI scientists.

In fact, I think that DeepSeek has opened opportunities for countries that normally don’t compete in car manufacturing. DeepSeek showed that there is a path to cheap AI. I can see a day when the intelligent platform will be cheap enough that any car manufacturer can license it, just like one can license Windows for a laptop or Android for a smartphone. Windows and Android have enabled many more companies to make laptops and smartphones. So there could actually be many more players in the field of self-driving cars. Not now, but in the near future.

Q4A: What about Chinese AI?
The last four months have been the season of the Chinese AI. Alibaba's Qwen has become the favorite open-source model both at Stanford and at Berkeley; and DeepSeek has shaken the confidence that Silicon Valley's AI was way ahead of China's AI. DeepSeek has integrated three technologies that i have been recommending for years: Chain of Thought, Reinforcement Learning, Model Distillation. All these Chinese models are an important steps towards sustainable AI, AI that doesn't require billions of dollars to train a model. DeepSeek was also the first company in the world to create a model that is all three things in one: a chatbot, a reasoning model, and a "deep research" system. And there are other Chinese startups that climbed the charts of best AI models: 01.AI with Yi (2024), Zhipu AI with GLM (2024), Stepfun, etc. I am also happy to see that the Chinese companies finally understand the importance of open source, another thing that I have been preaching for many years.
The restrictions on US exports of high-performance chips was probably good for Chinese companies because it forced them to focus on sustainable AI, and so now they have leapfrogged ahead of OpenAI, Anthropic, Google and Meta. The most sustainable AI is made in China, not in Silicon Valley. That's the big news of 2025. Silicon Valley is still stuck with the idea that you need to scale up in order to get more intelligence. It’s the Elon Musk model. China’s sustainable AI is closer to my vision of AI.

Q4B: What about AI in the mobility sector?
China's electric cars are probably the best in the world right now, especially if considering the price. China has already surpassed the USA and Europe in sales of electric vehicles. The advantage that Tesla still has is in the software, not the car itself. Tesla is a software company before it is a car manufacturer. That's where Tesla still beats Chinese electric cars. However, the Chinese car manufacturers can now take advantage of their own native AI models.
For example, Geely is known in the West because it owns Volvo and Lotus, but it is actually also an AI company. It developed its own Xingrui model and it was the first car manufacturer to integrate its model with DeepSeek-R1.
Geely partnered with a new AI startup, Stepfun, to develop two open-source multimodal models: Step-Video-T2V for video generation and Step-Audio for voice interaction.

Most car manufacturers are still using Nvidia’s Orin chips. Thor is four times more powerful than Orin. The first car that used Thor was the Lync 900, and Lync is a subsidiary of Geely.
At the recent CES2025 in Las Vegas, Geely announced an AI platform based on Xingrui + R1 and now they announced that their platform Qianli Haohan is powered by dual Thor chips. This new platform comes in five levels of intelligent driving solutions: H1, H3, H5, H7, and H9. And H9 is Level 3 automation, the first platform ready for mass production.

I would take them very seriously. Just like DeepSeek was the first serious threat to OpenAI's dominance in LLMs, Geely could be the first serious threat to Tesla's dominance in AI for mobility.
As far as i know, Geely is also the only car manufacturer besides Tesla that boasts its own satellite technology. Geely's subsidiary Geespace (the equivalent of SpaceX) launched its first low-orbit satellites for autonomous cars already in 2022. These satellite constellation can be accurate to within a centimeter. SpaceX's Starlink constellation has a lot more satellites in orbit (almost seven thousand) but we have learned that the Chinese can catch up quickly. In 2022, Geely also acquired control of Meizu, a smartphone maker, so they will be able to develop cars with smartphone features before Tesla.
Incidentally, these leaders in the AI field are all based in Hangzhou, located one hour from Shanghai: Alibaba, DeepSeek and Geely. The same city has Zhejiang University, which has a record of innovation in AI and Robotics, and Hangzhou was also home to pioneers of bitcoin mining. I am following closely the developments in Hangzhou. A very interesting hub of high-tech.

Q4C: What about automotive chips?

Batteries were the foundation for the first phase of electric car development, giving BYD a big advantage, but now chips are the foundation for the second phase of electric car development.

In the USA that market is still dominated by chip companies like Nvidia and Qualcomm, and only Tesla (and partially General Motors) are making their own chips. China is way ahead in creating competition for car chips. In fact, China’s electric car race is also becoming a race about chips. Traditionally, Chinese car makers relied on US chips like Nvidia and Qualcomm, but there was a tidal wave of Chinese car chip makers, following the trade war with the USA. China now has more than 300 chipmakers specialized in cars. For example, SiEngine (originally formed in 2018 as a joint venture between ARM, the great British designer of chips, and ECARX, de facto a subsidiary of car manufacturer Geely), made China's first 7nm car chip in 2024, the "Longying No. 1". SemiDrive made the first car chip to be certified in Europe in 2018. And there are many other car chip makers in China.

The reference point is Tesla, that started making its own chips in 2019. Tesla’s top AI chips are: the Full Self-Driving chip and the Dojo D1 chip, which are both 7 nm technology.

But the Chinese are catching up.

A few months ago another Chinese electric car manufacturer, Nio, announced the first automotive chip based on 5-nanometer technology, and Xpeng’s Turing AI chip has been widely publicized (although they didn't disclose the nm generation) and soon SiEngine (de facto a subsidiary of Geely) will introduce its own 7nm chip, StarLight or Xingchen-1 or AD1000.

Q6 I'm curious how Silicon Valley views China's Deepseek and the AI technological innovations by Chinese companies that you just mentioned?

Silicon Valley was shocked when DeepSeek’s R1 came out. But there is general satisfaction that DeepSeek has achieved something important for everybody. It proved that sustainable AI is possible. We are also pleased that all these new great Chinese models are open source. So in the end it helps international collaboration. Of course, some companies, like OpenAI, are not happy because it hurts their business model, but the community in general sees this as a very positive development.

Q7 Regarding technological ethics and social values, you have repeatedly warned about "humans becoming dumber due to technology." How can we avoid the alienation of human nature in the process of intelligence development? Do you have any related warnings in the field of transportation?

It’s the story of any new technology. Humans lose some skills when a technology does the work for them. A thousand years ago, when very few people knew how to read and write, and there were very few books, it was normal to memorize a long poem. Now it’s difficult to memorize even ten lines. My generation was able to drive anywhere without a smartphone. The new generation needs a smartphone or doesn’t know where to go. In the future people won’t even need to know how to drive a car because the car will be self-driving. Some of these skills will probably never needed again, but personally I think it’s useful to keep them. So I personally go hiking on mountains where there is no signal and only my orientation skills can help me. I still calculate with paper and pen and often try to calculate without writing. It’s good mental exercise. You never know when these skills can become useful again.