Robin Li is an important figure in China’s AI field, and he has led Baidu to continue to make efforts in the research and development and application of AI technology, from the rapid development of the Wenxin model to the extensive layout of autonomous driving, Baidu’s deep cultivation in the entire AI industry chain has provided practical samples for the industry. Despite the challenge of balancing technological ideals and commercial implementation, Robin Li continues to promote the progress of AI technology and help China’s AI move forward steadily in international competition.
In the past 20 years, too many companies have had ups and downs, but founders like Robin Li who are still on the front line are still rare.
Among the Internet tycoons, he is not only the first prophet to bet on AI (Baidu has laid out deep learning in 2010), but also the most persistent engineering school (firmly believing that “there is no shame in doing engineering”). When Musk imagined brain-computer interfaces and Zuckerberg was addicted to the metaverse, Robin Li’s AI narrative was always “useful”.
From the average daily call volume of the Wenxin model increased by 30 times, to the revenue of intelligent cloud tripled year-on-year, and then to the fully unmanned mileage of Radish Run exceeding 150 million kilometers, Baidu has verified a closed-loop path of “technology-application-ecology” in more than ten years.
However, the doubts from the outside world have never disappeared. Technological idealism can build competitive barriers, but commercial implementation is the only criterion for testing value. The public’s evaluation of Robin Li is polarized: supporters believe he is “single-handedly promoting the development of AI in China,” while critics question Baidu’s commercialization pace.
As the Internet company with the highest market capitalization in China, Baidu was the first company to systematically deploy AI, but the Wenxin model was complained of “lack of amazement”, the commercialization of autonomous driving was struggling, and the long-tail traffic of AI products was insufficient…… Today’s Baidu is the only AI company in China that realizes the vertical integration of “chip-framework-model-application”, but its market value is not as good as that of Alibaba and Tencent.
In the eyes of the industry, such a gap is the comprehensive result of the difference in AI technology investment cycle, commercial monetization efficiency and ecological strategy. The deeper contradiction lies in the fact that China’s AI needs both Robin Li’s long-termists and is trapped in the reality of “lack of AI super applications”.
Looking back at the node of 2025, China’s AI has entered “landing tackling” from “technological enlightenment”. When Baidu’s ten-year AI long-distance run became an industry specimen, the question of “whether we need more Robin Li” is essentially exploring three core propositions:
The value weight of technological idealism and business pragmatism, the balance logic between long-term investment and short-term efficiency realization, and most importantly, the path selection of China’s AI.
01 Robin Li’s relationship with AI is not the same as Baidu is to AI
There is no definite number of “who is the godfather of China’s AI”, but the industry voted for Robin Li the most.
Unlike Hinton, the “father of deep learning”, who contributed at the theoretical level, Robin Li played more of an AI technology practitioner and industry promoter. Through Baidu’s platform, AI technology has been deeply integrated into various businesses, and has been exploring an “AI+X” industrial landing path for more than ten years.
After 10 years of interaction design, why did I transfer to product manager?
After the real job transfer, I found that many jobs were still beyond my imagination. The work of a product manager is indeed more complicated. Theoretically, the work of a product manager includes all aspects of the product, from market research, user research, data analysis…
View details >
People know Baidu AII in AI, but they don’t necessarily understand the reasons behind it.
Robin Li mentioned in “Intelligent Revolution” that the evolution of search engines is essentially an upgrade of data processing capabilities: from early keyword matching, to personalized recommendations based on user behavior, to AI-driven intelligent decision-making, every step relies on the collaboration of data, computing power, and algorithms.
In other words, Baidu’s AI investment is deeply bound to the underlying logic of its search business, and as a tool for processing massive data, search engines naturally need AI optimization to improve user experience.
In 2010, Baidu established a natural language processing department, and after 2012, Robin Li personally recruited top talents such as Andrew Wu and Xu Wei to establish a deep learning research institute (IDL) to combine AI with search, advertising and other businesses. These actions seemed ahead of their time back then, but in fact they were inevitable for the iteration of the search business – the accumulation of user behavior data and the improvement of computing power made AI a key variable in optimizing the search experience.
“When I tried searching, I found that the feeling of applying AI in any field was different from the past. Massive data, stronger and stronger computing power, and lower and lower computing costs have come together in search, paving the way for the return of artificial intelligence. ”
In Robin Li’s view, domestic data is the largest in the world from the perspective of any single market, and artificial intelligence is essentially the intelligence of data.
As he said, the Internet has changed the information infrastructure, the mobile Internet has changed the way of resource allocation, and under the interaction, it not only produces massive data that scientists dream of, but also gives birth to cloud computing methods, summarizing the computing power of tens of millions of servers, and the most direct result brought about by the improvement of computing power is the so-called “thousands of people and thousands of faces”.
The ultimate form of the Internet depicted by Robin Li is an “agent operating system” with AI as the core driving force and a deep reconstruction of the way social resources are allocated. This vision transcends the boundaries of traditional search engines and builds Baidu into an infrastructure connecting the physical and digital worlds by real-time bite of “data-AI-human needs”.
Any new technology full of unknowns needs an evangelist, and to some extent, this is the role Robin Li has played in the past decade.
According to media reports, as a member of the National Committee of the Chinese People’s Political Consultative Conference, Robin Li has put forward 13 AI-related proposals in the “Two Sessions” for 8 consecutive years, and 40 public speeches (a total of 140,000 words) in 3 years. In the AI boom in 2016, Robin Li publicly mentioned “AI” more than 500 times. Anyone who is interested in AI, whether it is from students to technicians, from enterprises to local leaders, does not miss any opportunity to popularize AI.
At the 2015 Boao Forum, Robin Li, Musk, and Zuckerberg had a conversation, and at that time, their AI projects were just starting, and most of their views on AI were in imagination and silence.
Less than a year later, OpenAI was founded, and in the same year, Robin Li submitted a proposal for the “China Brain Project”, hoping to build AI basic resources and public service platforms at the national level and seize the commanding heights of a new round of scientific and technological revolution.
In contrast, Alibaba and Tencent’s strategic focus has always revolved around user traffic and business monetization.
Alibaba builds a “business operating system” with e-commerce as the core, connecting transactions, payments, and logistics into a closed loop through Alipay, Cainiao Network and other infrastructures; Tencent takes social networking as the entrance, covers communications, entertainment, finance and other scenarios through the WeChat ecosystem, and converts social traffic into local life service income by investing in Meituan, Didi and other enterprises, and financial technology and enterprise services have become the second growth curve.
Compared with Baidu, the business layout of these two focuses more on the in-depth mining of user stickiness and the rapid transformation of commercial value, and the difference is directly reflected in the financial structure: over the years, Baidu’s search business revenue still accounts for the majority, while the core businesses of Alibaba and Tencent can significantly increase revenue in e-commerce, games, financial technology and other sectors through diversification.
It is precisely because of this that in the case of a single source of income and bottlenecks in the growth of the main business, the outside world has found the most reasonable explanation for Baidu’s efforts to develop AI.
But it is clear that compared with the continuous investment in AI for many years and the dismal ROE, Robin Li’s sustenance for AI is not only to activate the existing business, but in his vision, the impact of AI will be more far-reaching, and “rebuilding Baidu” is only one of them.
02 About the commercialization of Baidu AI
Compared with the significant growth in revenue of Internet companies such as Alibaba and Tencent, Baidu’s high annual R&D expenses are somewhat out of place: since 2015, Baidu’s annual R&D investment has exceeded 10 billion yuan, and the proportion of investment has remained at about 15% of revenue.
But in reality, as Robin Li said, “I learned AI well, but after doing some research, I found that there are no application opportunities and cannot solve practical problems.” ”
Throughout these years, Robin Li’s strategic thinking on AI has shown a trend of iteration with the evolution of the industrial cycle.
With 2020 as a watershed, before that, Robin Li advocated that “full-stack self-research can control core competitiveness”, and gradually built full-stack technical capabilities from chips, frameworks to models and applications. He believes that full-stack self-development can achieve a closed loop of technology and avoid being subject to external technology ecology.
At this stage, Baidu launched Apollo, focusing on L4 autonomous driving; Open source PaddlePaddle, which supports internal search, recommendation and other services, and provides AI development tools to the outside world. The release of the original Wenxin model also gives priority to serving internal businesses such as search and information flow, and feeds back technology iteration through scenarios.
Baidu’s early technical route was not completely closed-source, but more reflected in the relatively clear and effective cognition of large models in commercialization scenarios. Several iterations of the Wenxin model in the past have been closed-source, charging enterprises through API interfaces and privatized deployment.
Robin Li once mentioned in an internal speech that the closed-source model can gather computing power and talents to achieve technological leadership through large-scale applications. At the same time, Baidu is also cultivating a developer ecosystem through open source and quickly establishing industry standards. It is understood that its open source efforts during this period far exceed that of OpenAI in the same period, which did not open source GPT-2 until 2018.
After 2020, Baidu’s closed-source tendency has gradually strengthened.
After the release of Wen Xin Yiyan, Baidu turned to fully closed source and launched paid services to maintain the competitiveness of the enterprise service market. At the Baidu World Conference, Robin Li mentioned that closed-source large models are usually better than open-source large models in terms of capabilities, especially in commercialization and practical application scenarios, which can more efficiently meet the customized needs of enterprises, reduce usage costs and protect enterprise data security.
However, under the impact of Deepseek’s open source, it was announced that the Wenxin Model 4.5 series embraces open source and launches zero-code development tools such as AgentBuilder and AppBuilder to focus on commercial applications. C-end applications, including Wen Xin Yiyan, are completely free.
This shift shows that Robin Li’s closed-source proposition is more of a phased strategy for the technology accumulation period rather than a long-term unchanging principle.
“Technical strength is not the same as commercialization ability.” As early as the beginning, he realized that the route choice of AI largely depends on the ability to commercialize.
Commercialization is not only Baidu’s problem, but also the confusion of the industry. The high investment and high sensitivity of technology make there is a natural contradiction between AI and the efficiency demands of the business world.
Looking around at large enterprises at home and abroad, there are not a few AI companies trapped in commercialization, and the cycle mismatch between technology value and commercial monetization is a common problem in the industry. From OpenAI, Google, and DeepMind in the United States to SenseTime and Cambrian in China, large enterprises are generally facing the dilemma of “leading technology but difficult to make profits”.
Today, the commercialization of Baidu AI presents a pattern of parallel technology open source and ecological expansion, in-depth implementation in vertical fields and continuous breakthroughs in general scenarios.
For example, through low-price strategies and full-stack technology openness, developers are attracted to participate in pre-research, and cooperation with international flagship products such as Samsung and Apple has been reached. The “open source drainage + cloud service monetization” model gives Baidu an advantage in the MaaS market, and at the same time connects third-party models such as DeepSeek through the Qianfan large model platform to form a “hybrid ecology” advantage.
In addition to these, on the eve of the commercialization of autonomous driving entering large-scale, Apollo launched large-scale commercialization, targeting Beijing, Shanghai and other first-tier cities; In terms of embodied intelligence, Baidu cooperates with Zhiyuan Robot to develop solutions that are effective in education, safety production and other fields.
Compared with major manufacturers such as Byte and Alibaba, Byte relies on Doubao and Instant Dream AI to expand rapidly on the consumer side, and Alibaba reconstructs the search experience with Quark’s “AI Super Box”, and Baidu’s advantage lies in enterprise-level solutions.
In the future, whether Baidu can find a balance between “technology open source” and “commercial closed loop” will determine its final position in the global AI competition, and will also directly provide a route reference for the commercialization of AI in China.
03 Robin Li and his “fans”
There was an interesting thing two days ago.
Robin Li mentioned at Baidu’s internal awards ceremony that on the day of the Create conference, he took the “Radish Run” to the venue, and suddenly a passerby shouted “Baidu Niu X” at him with a throat.
He sighed about this and cited it as a footnote to explain the multi-dimensional presentation of Baidu’s technical heritage – from the chip layer to the application layer, from model research and development to scene landing, Baidu has no vacancies. At the end, he also mentioned a line in the show: “Do you really think you can change the world? His answer was “I want to try”.
Compared with Robin a decade ago, Robin Li now seems to be full of technological ideals, and for a techno-fundamentalist entrepreneur, the biggest recognition is the public’s perception of technology.
In fact, Robin Li’s fans are not a niche group.
A report more than ten years ago wrote that Baidu 4,000 employees moved to the new building, and Robin Li spent his 41st birthday in Baidu Building with fans from all over the country that day. In the afternoon, he didn’t even forget to interact with 230,000 fans in his post bar: This day is very busy, everyone is very excited, I am very excited, thank you to my friends! With just a few words, it quickly got hundreds of replies from fans.
At that time, Robin Li was a veritable darling of the times, he represented the earliest batch of returnee elites in China, archaeological Robin Li Tieba, fans called themselves Hong fans, most of this group of people, most of them were students, and there were many professional and technical talents, they were the purest group of people in the technical circle that year.
When Robin Li laid out AI, he could not see the future of this technology, but for this research, he promised that there was no upper limit on budget investment.
More and more artificial intelligence scientists have jumped from laboratories in well-known universities to Baidu. On the one hand, universities cannot provide the massive data needed to develop artificial intelligence, nor can they bear the huge cost of computing hardware clusters, on the other hand, Robin Li has almost given the best AI talent treatment in China.
As a typical example, Robin Li offered a sky-high acquisition fee of $12 million to DNN Research, a “three-no company”, in order to attract talent, Professor Jeffrey Hinton and his two students.
The following year, Baidu Deep Learning Research Institute was established, which is also “the first research institute in the global business community to be named after deep learning”. At present, Baidu Research Institute has input a large number of professional and technical personnel to Baidu and the AI industry, contributing 27,000 global AI patents, including the world’s first deep learning patent, the first large model patent in China, and the world’s first in the number of autonomous driving patent families, benefiting dozens of industries and more than 100 application scenarios.
With the full-chain integration of “chip-framework-model-application”, this technology penetration capability provides talents with a complete practice chain from theoretical research to scenario landing, Baidu is known as the Whampoa Military Academy in the field of AI, Robin Li said, “In the next 5 years, another 10 million AI talents will be cultivated for the society”.
From an internal point of view, Baidu had more than 40,000 employees at its peak, and the number of technicians once reached 70%; In the industry, early members of IDL, such as Yu Kai (founder of Horizon), Ni Kai (co-founder of Pony.ai), Tao Ji (CTO of Wenyuan Zhixing), etc., have become leaders in the field of autonomous driving, and nearly 30% of the core members of the Baidu Apollo team have led or participated in the research and development of L4 autonomous driving systems, and their technical path choices directly affect the development direction of the industry.
In addition, Baidu sends thousands of AI engineers to the industry every year, most of whom enter major manufacturers such as Huawei and ByteDance, as well as AI unicorns such as SenseTime and Megvii. These people not only bring Baidu’s technical methodology, but also inject “full-stack” problem-solving capabilities into new enterprises.
The AI company founded by former Baidu employees has also formed a unique “Baidu system” technology school. For example, the “Flying Paddle Biocomputing Platform” led by Dai Zonghong of the Zero One Ten Thousand Internet of Things applies deep learning to new drug research and development; The “Zhipu Qingyan” model developed by Zhang Peng, the founder of Zhipu Huazhang, led the team, surpasses some international competitors in the field of code generation, and its technical route is in line with Baidu Wenxin.
Robin Li once wrote in his blog: “In China, what is silent is not management, but technology, too few people really care about technological progress, and too many people are obsessed with treating management as a war.” “When the Wenxin model supports the innovation of tens of millions of developers, Apollo’s engineers redefine urban travel, Baidu is not only exporting talents, but also reshaping the technical standards and talent paradigm of the AI era.
This positive cycle of “technology-talent-industry” is the core competitiveness that distinguishes it from other enterprises, and also provides a replicable path sample for the sustainable development of China’s AI industry.
