On February 4, Kai-fu Lee, CEO of Zero One Thing and Chairman of Sinovation Ventures, talked about the AI ​​competition between China and the United States in an exclusive interview with the Financial Times. He believed thatChina will beat the U.S. in consumer AI.At the same time, businesses need to be more proactive in adopting this transformative technology.

Kai-Fu Lee

Over the past four decades, Kai-fu Lee has personally witnessed the rapid development of China's AI industry. He first participated in the establishment of Microsoft Research Asia and other institutions, and played a key role in cultivating and delivering AI talents that now support China's leading companies.

Kaifu Lee accepted an exclusive interview with Eleanor Olcott, China technology reporter of the Financial Times. The following is the full text of the interview:

Q: Can you tell us about your startup Zero One Wanwan?

Kai-Fu Lee: Zero One Wish is committed to providing tools for enterprises to develop AI agents. Based on the open source model, we will select the appropriate model for the specific application scenarios of the enterprise and customize it for each customer. We believe that in the early stages of AI agent application, the key is to provide high-standard "white glove" services and explain to enterprises how to implement the technology. We will work with enterprises to create the most valuable application solutions, which not only helps save costs, but also creates real business results.

Q: How prepared are these companies to adopt AI?

Kaifu Li: The clients we work with come from traditional industries such as banking, insurance, mining, and energy. Compared with technology companies, these enterprises are less prepared for AI applications. Some companies have not even completed the digital transformation required to apply AI. In this case, we will not cooperate with them because the transformation cycle is too long and the cost is too high. Another problem is that some companies' understanding of their own needs lags behind technology development. For example, they may ask for the development of a customer service agent, but this is not the most value-creating application area of ​​​​the current technology.

If an enterprise lacks AI professional capabilities, it must cooperate with an AI company to jointly formulate an AI strategy. This type of transformation is usually led by the CEO himself and is very difficult. Perhaps only one company in a hundred is ready for such profound change.

When we work with businesses, we get deeply involved. They have to be fully committed. We hope they hire a chief AI officer (CAIO) because chief information officers (CIO) are often not up to the task. CIOs tend to be very conservative, while CAIOs need to have a bold strategic vision and be able to rethink the company's organizational structure and strategic direction. They will work directly with the CEO to reshape the business. When clients are unable to provide a suitable candidate, we provide them with one.

Our business model is similar to Palantir in some aspects: we have a team of consultants to help companies formulate strategies, and an implementation team responsible for implementation. Our compensation is tied to the business results we create. During the strategy development phase, we charge a fixed fee to cover costs, but if no commercial results are ultimately produced, there are no subsequent fees.

Question: What are the reasons for the other 99 companies that are unwilling to do this?

Kai-Fu Lee: Sometimes it’s becausePeople treat AI as just another piece of software.Sometimes CEOs don’t have an intuitive understanding of AI itself. Sometimes people think of AI as just another kind of enterprise resource planning (ERP) software. Sometimes they leave it in the wrong hands. For example, it is often unwise to hand over AI strategic planning to the CIO, because the core of his responsibilities is to ensure the smooth operation of the company's computer systems and software, rather than to promote corporate transformation.

Q: Currently, the industry generally believes that China's large models lag behind the leading US models by about 6 to 12 months. What's the reason? Do you think this gap will persist?

Kai-Fu Lee: Currently, the United States occupies a dominant position in AI research breakthroughs. The United States has gathered most of the world's top researchers and has massive computing power, which has promoted the progress of large language models. But building on these breakthroughs, talented and engineering Chinese companies would quickly figure out how to build similar technology, often faster, giving rise to their own “aha moments.”

Putting a man on the moon was very difficult for the first country. But once someone does it, even if you don’t know exactly how, that very fact makes it much easier for a second country or company to do the same thing.

China has solid engineering practice capabilities and scientific research strength. Therefore, Chinese companies have not only made achievements in independent research and development, but they can also explore the working principles of these American models even when American companies have not open sourced the code or published papers. Perhaps, the visible results alone are enough to bring inspiration; perhaps through clever reverse engineering; or through model distillation technology; perhaps by deducing from first principles; or perhaps, they found different first principles, but in the end reached the same goal by different routes.

Therefore, Chinese models can often catch up. When DeepSeek launched its model, that gap shrunk to about three months, but now it appears that Google Gemini has taken the lead, lengthening the gap to about 12 months.

This gap will sometimes narrow and sometimes widen.About six months is the median.All practitioners will learn from every innovative idea and model released publicly, because the AI ​​field attracts many top talents. This is true in both China and the United States, both sides are thirsty for knowledge. This learning does not flow in one direction. When the DeepSeek model was released, American companies also conducted in-depth research on it.

Q: When DeepSeek released its R1 model in January 2025, many parties, including OpenAI, accused it of taking shortcuts in building inference models through model distillation technology. OpenAI subsequently claimed that it had taken measures to prevent such situations. Putting aside hard-to-substantiate accusations of technology theft, is it becoming more difficult for Chinese companies to learn from U.S. companies now that the U.S. is taking more proactive measures to protect its technological secrets?

Kai-Fu Lee: OpenAI feels they must keep the model closed. After all, with so much money invested in training groundbreaking models, it would be easier for others to learn these techniques if they were open sourced. It's understandable that they feel they invested a lot of money into inventing this intellectual property and don't want to share it.

In addition, they believe that the future of general artificial intelligence (AGI, hypothetical AI with human-level cognitive abilities) will appear in the form of a giant leap. Once a company takes the lead in breaking through, it will crush all other companies in the world, whether American or Chinese. So, from this perspective, if you believe that the future of AGI is a "winner takes all" situation, then you must keep the methods for achieving this goal strictly confidential.

American businesses continue to raise hundreds of billions of dollars. They have to prove to investors that by building AGI they will dominate the world, and investing $50 billion today is cheap because the company will be valued at $50 trillion in the future. This makes the entire narrative true for OpenAI, a story that is understandable and has some credibility.

But I think the other possible story is that it's not just one talented team, or even four talented teams. In the United States, you have OpenAI, Anthropic, Google, and xAI, and each company thinks it is the genius who can beat everyone else and win the Nobel Prize by solving the ultimate problem of AGI.

But China's approach is different.The Chinese approach is more like a learning group: one company releases a model, and other companies study, try, and maybe even discuss with the company how the model was trained.All members of the study group develop open source technology and then share the results with each other. This study group is made up of very smart people funded by companies that still have to chase quarterly profits.

This is in sharp contrast to the situation in the United States. American businesses can invest without regard for short-term returns. In China, there are practical constraints on corporate spending capabilities: Alibaba cannot sustain a loss of US$10 billion next quarter, but OpenAI can. Therefore, all these factors jointly shape the behavior model of Chinese enterprises: under the circumstances of limited resources, through continuous learning, incremental improvement, and learning group-style collaboration,Rather than adopting the radical “winner-takes-all” strategy of the United States.

Q: There is a mainstream view that the United States’ leadership in AI is a strategic geopolitical advantage. But I think there is another possibility: in the future we may see that China's current relative lag may be turned into an advantage. Due to the time lag in technological development, China can observe the development of AI in the West, see the economic and social impacts brought by AI, and choose to adopt different strategies based on the mistakes and pitfalls it sees. What do you think of this point of view?

Kai-Fu Lee: It is almost certain that if there are negative consequences caused by AI in the future, their source is likely to come from American companies. Whether it is misused by criminals or a vulnerability occurs due to some negligence. This risk is rooted in their "winner takes all, move fast, and break the rules" operational mindset. This model will naturally reduce companies' vigilance to risks. At the same time, they are operating more dangerous "weapons" because their models and technologies are more cutting-edge.

In China, people generally do not believe that AGI will be monopolized by one company and crush all competitors. I think the general public is more inclined to think that this is going to be a linear evolutionary trajectory, with winners changing over time.

Q: Of course Chinese companies also want to be winners, right?

Kai-Fu Lee: They do want to win, but they are unwilling to pay an exorbitant price, raise US$500 billion, and bear the risk of failure or bankruptcy. Chinese companies pay more attention to business results and are committed to developing products that can be profitable through models. Whether it is Tencent’s WeChat, Alibaba’s Taobao, or ByteDance’s Douyin, these companies hope to build a competitive model that is closely integrated with their own products and can create revenue.

Q: What kind of development trend do you think China’s AI industry will show this year?

Kai-Fu Lee: I think China will lag behind the United States in terms of enterprise applications because Chinese companies are unwilling to pay high subscription fees. In contrast, China will lead the United States in consumer applications. There are a large number of startup companies in both countries dedicated to the development of consumer-grade applications. The time is now ripe and the model capabilities are sufficient.

But I think the Chinese technology giants will far surpass the American giants in building excellent applications because they have always been tenacious, driven and have the awareness to dominate the market. These companies regard technology as a means to achieve application innovation, so they will focus more on application implementation. They will extend the functions of existing applications through AI and use AI to build new applications. This trend has already emerged.

I believe that Chinese Internet companies will become the main source of such application innovation, and their innovation momentum will surpass their American counterparts. If you look at typical American apps, whether it's Instagram, YouTube, or Snapchat, they're becoming pretty boring. I don’t think American Internet companies have the same drive, nor are they willing to constantly innovate themselves like Chinese consumer application companies.

In contrast, China’s consumer app companies, such as Bytedance, Tencent, Alibaba, Meituan, Pinduoduo, and Xiaohongshu, have a tenacious spirit and desire to win, and are committed to creating new innovative products. Many of these companies are already building excellent AI technology, agents, and models, and are investing heavily in this area, far beyond what is typical for U.S. companies.

Secondly, 2026 will be the first year of “AI native devices”. This year, we will see, touch and buy truly AI-native devices for the first time. It may not be the final form, nor may it be the ultimate victory paradigm. It could be a "Nokia moment," it could be a "BlackBerry moment," or it could be an "iPhone moment." We don’t know yet which one, but all three moments are important in the history of mobile devices.

The necessity of AI-native devices lies in the fact that humans are always eager to give instructions to devices through voice and natural language. This means that we only need to tell the device the expected results, rather than guiding it step by step to complete the task, and the agent will complete all subsequent steps autonomously.

Intelligent technology is already developing in this direction, but it needs to be equipped with a voice-driven interactive interface, and smartphones are not an ideal carrier. The limitation of mobile phones is that they cannot be constantly online or listen at all times. Therefore, we need a device that can operate around the clock, continuously listening and recording information. It will store what you see and hear and make inferences based on it. Although this answer is a bit lengthy, I think the key lies in this "ambient AI": it is always online, constantly listening, has infinite memory, but is invisible.

Q: Looking back on your career in China’s AI industry, if you look back to the beginning, what would you be surprised about today’s industry, and what aspects have remained basically the same?

Kai-Fu Lee: I think the optimism about the belief that AI will change the world has never changed. What surprises me is how quickly AI has developed in the past three years. I thought this would be a slow process lasting 10 or 20 years, but it happened faster and matured more rapidly. Of course, we still have a long way to go.

When I first entered the industry in the 1980s, AI was always viewed as a collection of impractical technologies. Whenever one of these technologies happened to work (which was rare), it was immediately productized and no longer called "AI." At that time, people used to laugh at us, thinking that we were a bunch of lunatics who imagined that AI could think like humans. Therefore, the industry has developed from "only dreamers and visionaries are engaged in AI" to "now everyone wants to participate in AI."