GuShiio.com 
Robin Li, founder and CEO of Baidu (Photo source: video screenshot)
BaiduRobin Li, founder, chairman and CEO, was finally asked about the impact of DeepSeek in China.
On the afternoon of February 11, Beijing time, I learned thatOn the first day of the 2025 World Government Summit in Dubai, Robin Li held a round-table conversation with Omar Sultan Al Olama, AI Minister of the United Arab Emirates.
Oama first asked Robin Li directly: In your opinion, is the craze for DeepSeek expected? For example, do you really think this will happen? It will happen this way in the past. Or are you also trapped by DeepSeek? rdquo;
Li Yanhong said in this regard,When talking about big models today, the cost of big AI models is reduced by more than 90% every year, much faster than Moore’s Law of the computer revolution. In the American environment, they always build or develop the most powerful big language models, but in China, due to computing limitations, we must innovate to reduce costs.
“Basically, I don’t think innovation can be planned. You don’t know when and where innovation will come. All you can do is create an environment conducive to innovation. rdquo;In Li’s view, Baidu launched Wenxinyiyan (now called Wen Xiaoyan) three months after ChatGPT was released. This was the first listed company in China to launch a ChatGPT-like application.“This is a very exciting time, we see innovation everywhere, and we have to adapt to this fast-changing innovation. rdquo;
It is reported that as of November 2024, the average daily call volume of Baidu Wenxin’s model has exceeded 1.5 billion, an increase of about 30 times compared with the 50 million times first disclosed a year ago. Yanhong Li said that the growth rate exceeded expectations, indicating that AI is a real demand.
He lamented that this represents the explosion of application of large models in China in the past two years.
In January 2023, GuShiio.comApp reported that Robin Li bluntly stated within Baidu that short-term corporate losses are possible because rapid growth is possible, but in the long run, it is not possible. Many enterprise-level companies have no cash flow capacity and no cash flow. Eventually, the company will die if the funds break. Li Yanhong also criticized internal employees for having no business awareness, resulting in poor quality of new business income.
In November 2023, Robin Li said at an event in Shenzhen that China has too many big models. There are more than 200 large models in China that actually have no usage. He alluded to the fact that the Wenxin Large Model family has a higher call volume than the more than 200 large models. Li Yanhong also said that to develop a large model that is easy to use and usable, there are high technical and cost thresholds. Repeatedly building wheels is actually a great waste of social resources.
In December 2023, Li Yanhong emphasized at an event in Beijing that the Hundred Models War is a great waste of social resources, and more resources should be placed in super applications. He also bluntly said that large companies including Baidu have too slow internal reactions and backward productivity, so internally they have repeatedly emphasized the need to use literary thoughts and words to reconstruct applications, rather than using them as a tool. ldquo;First of all, in fact, large companies are very slow to respond. I even sometimes say that large companies represent backward productivity. You must not look at what big companies are doing.
In April 2024, Yanhong Li once again stated that with the same effect, the cost will be significantly lower, so the open source AI model will become more and more backward, attracting market attention.
In July 2024, Yanhong Li bluntly said that the open source model is an IQ tax. When you rationally think about what value a large model can bring and at what cost, you will find that you should always choose a closed-source model. Today, whether it is ChatGPT or Wenxinyiyan, closed-source models such as today, they must be more powerful and have lower reasoning costs than open source models.& rdquo;
In September 2024, Yanhong Li expressed his disagreement with the view that there were no barriers to large-scale model capabilities. Moreover, he believes that computing power is a key factor in determining the success or failure of large models, but open source models are too inefficient to be suitable for future development.In addition, Robin Li also predicted that in the future, the gap between China and OpenAI GPT’s AI models may become wider and wider.
Regarding the development trend of AI applications, Li Yanhong once said that agents are the most mainstream form of AI applications and are about to usher in a breakout point.
In February 2025, the craze for DeepSeek, a China open source AI model company, swept the world, and multiple projects of Baidu connected DeepSeek.
Among them, Baidu Intelligent Cloud Qianfan Model Platform officially launched DeepSeek-R1 and DeepSeek-V3 models. On the first day of the model launch, more than 15,000 customers have made model calls through Qianfan Platform; Baidu Baige released and deployed a full-blood version of DeepSeek R1+ online search service based on Kunlun Core P800; Baidu Intelligent Cloud Keyue, Xiling, Zhenzhi, and Yijian four major model application products have completed the adaptation of DeepSeek; Baidu Wenxiaoyan’s iOS version of the App connects to the DeepSeek-R1 model to optimize the photo problem solving function.
Robin Li also said at the World Government Summit that Robotaxi can greatly reduce traffic accident deaths. Judging from the actual record of radish fast running, the accident rate is only 1/14 of that of human drivers.& ldquo; Technology is advancing very rapidly, and autonomous driving is ten times safer than human drivers.” rdquo;
However, Li Yanhong also said that this (autonomous driving) is still anew industriesIn a new field, people have very low tolerance for accidents. If some serious accidents occur, they may drag down your (business) revenue, so you may need to stop operations for a while, find out what happened, and make all adjustments to restart operations. We’ve been very careful, so to today, even though we’ve been operating on a relatively large scale for two or three years, we haven’t had any accidents.& rdquo;
Yanhong Li focused on computing power and cloud infrastructure at the meeting. He believed that the revenue at the infrastructure level is much higher than that at the application level, and it is also an area that requires focused investment.
He pointed out that currently at the application level, AI big model companies have not made any money yet. When we talk about AI-native applications, the ARR or annual recurring revenue may be tens of millions or at most hundreds of millions of dollars, but at the GPU level, the talk is tens of billions of dollars. Although the technical cost of the AI big model is reduced by 90% every year, the performance is getting better and better. Therefore, the technology is still developing. You can’t stop investing, you have to invest, in order to ensure that you (Baidu) are at the forefront of this technological revolution.
“I think that in computer and cloud infrastructure, as well as investment in chips, is still very necessary to come up with the next basic model, we must use more computing to try AI models in various fields. Maybe at some point, you found a shortcut, say, spending $6 million to train a model, but you’ve spent billions of dollars to figure out which way is right. rdquo; Li Yanhong said.
Yanhong Li pointed out that it is unlikely that it is currently expected that super applications will appear in the AI application layer and be able to make money and generate value.
“I think we really need to worry about value creation at the application level. If you spend hundreds of billions of dollars on infrastructure layers, you can’t come up with 10 times more apps. This is unsustainable. So even if we have seen various use cases in various scenarios, it may be more like two B-sides. On the consumer side, we have not seen so-called super apps like the mobile Internet stage, apps with hundreds of millions of daily active users and people spending 2 hours a day. We have not seen such opportunities. But I believe that over time, someone will find out this, and you will be able to create a very sticky app.& rdquo; Robin Li said that there is still a long way to go from super applications, and the whole world is looking for such super application products.
Robin Li repeatedly emphasized that looking back on the past 100 years, most innovations have been related to cost reduction. If you can reduce costs by a certain amount and a certain percentage, it means that your productivity has also increased by that percentage.
“I think this is the nature of innovation, but today it’s just much faster than before. But for companies in China, we are subject to some type of restriction. rdquo; Li Yanhong said.
