On the evening of February 25, the Hong Kong Stock Exchange announced that Baidu and Huya Group reached an acquisition agreement, and Baidu acquired Huya Group's domestic video entertainment live broadcast business YY Live for US$2.1 billion. After the new agreement was reached, the US$1.6 billion (about RMB 11 billion) previously deposited into the co-custodial account has been fully refunded to Baidu, which will invest the money in the construction of cloud and AI infrastructure.

As the acquisition of YY is settled, Baidu not only completely ended this tug-of-war transaction for more than four years, but also added 10 billion yuan of investment chips in the current big model competition.

Considering the objective reality that YY Live is still making money, for Robin Li and Baidu, this is nothing more than finding a new cash bull for the company in addition to the original Baidu mobile ecosystem, and jointly shouldering the heavy responsibility of Baidu's blood transfusion in the new AIGC competition.

The greater strategic significance of ending the YY acquisition transaction is to make the outside world reevaluate Baidu's determination and perseverance to bet on AI. Previously, Tencent took the lead in taking action among the technology giants that were struggling with DeepSeek, and Baidu followed closely behind. But now it seems that Baidu is performing less inconsistent with AI development strategies than Tencent: it not only chooses to embrace DeepSeek, but also opens the capabilities of advanced large-scale models to all users for free, but also further chooses to open source its new generation flagship model of the future - Wenxin 4.5 series. After completing the acquisition of YY, it chooses to exceed 11 billion yuan to put AI into the safe.

At the beginning of 2025, the popularity of DeepSeek R1 has achieved a big model re-education for the entire industry, further improving the public's enthusiasm for understanding and experience of basic models.

More importantly, DeepSeek has shown signs of early arrival of the application of big models. David Luan, former vice president of OpenAI and now head of Amazon AGI lab, recently explained that the outside world misunderstanding about DeepSeek is that DeepSeek only verifies that more intelligence can be achieved at a lower cost, but "it does not mean that you will stop pursuing intelligence. On the contrary, you will use more intelligence."

The AI ​​technology narrative that has not been changed by DeepSeek has also been staged in a number of major technology manufacturers at home and abroad. OpenAI officially announced the upcoming news of GPT-5, and Baidu also reported that Wenxin 5.0 will be launched soon.

The pursuit of computing power demand among top players in big models is still in progress: after the new quarter's financial report was released, Google disclosed a capital expenditure plan for fiscal year 2025 of up to $75 billion to continue to expand its AI strategy; Microsoft announced that it would invest $80 billion to build an AI data center; Meta will also set aside $60 billion to $65 billion to build AI infrastructure and significantly expand its AI team.

CountryAlibaba, which has also recently announced a new AI plan, that is, to invest more than 380 billion yuan in the next three years to build cloud and AI hardware infrastructure.

Behind the high AI expenditure is the objective reality that AI is becoming a new engine to drive the business development of major technology companies. This is also clearly reflected in Baidu. In the latest release of the fourth quarter financial report for fiscal year 2024, Baidu's smart cloud business revenue reached 7.1 billion yuan, a year-on-year increase of 26%, ranking first among Baidu's many businesses.

From the new generation of AIGC launched by ChatGPT, Baidu has launched a war of 100 models in China, and then to the current rapid development of application waves driven by DeepSeek, technology manufacturers with cloud services have once again entered a new round of infrastructure investment period. No one wants to miss the AI ​​era express train.

The connection between Baidu and YY is all due to an acquisition more than four years ago. In November 2020, Baidu and Huya Group reached an agreement to acquire YY Live for US$3.6 billion.

Although the transaction delayed for four years, Baidu had already paid US$3.6 billion in advance at that time. According to the prospectus filed by Baidu in Hong Kong for the second listing in March 2021, after considering the adjustment of working capital of US$100 million, Baidu paid a total of US$1.9 billion to Huya and its designated joint custody accounts, and deposited a total of US$1.6 billion into multiple joint custody accounts.

The joint escrow account is jointly supervised by both parties. Once the delivery conditions established in advance are triggered, the corresponding amount will be automatically transferred to the buyer or seller. Now, as the price of acquiring YY dropped from US$3.6 billion to US$2.1 billion, the total $1.6 billion in the account has been returned to Baidu.

In addition to recovering tens of billions of funds for AI investment, Baidu, which completely acquires YY, is expected to continue to make profits from YY.

Although the acquisition price has dropped, YY Live is still a continuously profitable business and is expected to become the new cash bull under Baidu. Previous data showed that YY's revenue in 2018, 2019 and 2020 was 10.1 billion yuan, 10.7 billion yuan and 9.7 billion yuan respectively, and the annual profit was more than 3 billion yuan. Although no specific revenue data has been disclosed in recent years, according to people familiar with the matter, YY's annual revenue is still above 10 billion yuan and has maintained a good profit.

In addition, YY, which has more than 40 million monthly active users, can also become a new channel for Baidu to focus on short dramas.

In early 2024, Baidu APP opened a secondary entrance for short dramas, and the company began to restart short drama projects. According to Photon Planet, in the fourth quarter of last year, Baidu short drama users had more than 40 million daily active users, and the quarterly GMV increased by 51.2% month-on-month.

Now, with the addition of YY, Baidu short dramas have greater operation space, whether in channel distribution or content ecosystem construction.

More importantly, as a live broadcast application, YY can also be provided to Baidu's richer AI technology application scenarios.

After ChatGPT triggered the wave of AIGC, Baidu executives mentioned at a financial report meeting that using artificial intelligence to generate short video content is a typical application of AIGC. This can not only create opportunities for the platform to attract new users, increase user time spend and user stickiness, but also hope to help Baidu increase online marketing revenue and promote the growth of cloud revenue.

When Baidu searches fully connect to DeepSeek, YY has recently announced the access to DeepSeek almost at the same time and launched YYDS. Earlier, YY also officially released the AI ​​companion digital person "Ling'er", becoming the first live broadcast service-oriented digital person it launched. Official data shows that the digital person "Ling'er" has currently covered more than 6,000 live broadcast rooms, serving more than 1 million users per day on average.

Faced with the big model track that is entering a new round of investment period, Baidu is currently in need of more funds to ensure the smooth development of AI infrastructure. The 10 billion yuan of funds released from YY's acquisition undoubtedly played a role in adding icing on the cake.

The current access to the open source model DeepSeek has gradually evolved into a resource competition among large technology companies. The official service of DeepSeek often prompts busyness, but the full access to the full version of DeepSeek's full-blooded Baidu search has basically achieved a smooth experience on the user side. Behind this is that Baidu invests more funds to call for more capital expenditures of computing power clusters.

In addition to DeepSeek, Baidu also officially announced on February 13 that it will open Wen Xin Yiyan to all users for free and that Wen Xin Yiyan will launch a new in-depth search function.

As the free use cycle begins on April 1, Baidu will inevitably attract a wave of user traffic peaks. To ensure a good interactive experience between new influx users and Wen Xinyiyan, Baidu also requires more investment to expand its computing power service cluster.

In addition to meeting the needs of more C-end users, Baidu Cloud is also increasing its service to the B-end developer group at the same time. "We continue to enhance the MaaS platform to improve the model and application development experience for customers and partners. Our platform adopts the Wenxin series of models and provides access to a range of high-quality models around the world, and DeepSeek is one of our latest members," said Baidu executives at the new quarterly earnings conference.

In addition to deploying DeepSeek on the cloud platform, Baidu has also made the idea of ​​open source, announcing that the Wenxin Mockup 4.5 series, which will be launched in the next few months, will be officially open source from June 30.

"One of the things we learned from DeepSeek is that the best open source model can greatly promote users' adoption and recognition of the model," Li Yanhong said when explaining Wen Xin's 4.5 open source strategy.

At that time, on Baidu Cloud Qianfan big model service platform,B-end users can not only call third-party models such as DeepSeek, but also directly replicate the training secrets of Wenxin’s big model. In order to meet the needs of more users to call different models, Baidu needs to further increase its capital investment to expand its AI infrastructure.

Especially under the stability of the existing mobile ecosystem advantages, one of Baidu's current top priorities also includes further improving user experience and improving product functions. This tests Baidu's ability to continuously develop and iterate on big models.

In order to provide users with an amazing experience that is comparable to DeepSeek, early funding preparation and investment are even more indispensable.

In the long run, Baidu's investment in AI is far more than 10 billion.

In the past decade, Baidu has invested more than 180 billion yuan in R&D, and its R&D investment has remained above 20% of its revenue for many years. In addition, YY's recovered funds, it is estimated that in the next three years, Baidu will invest more than 100 billion yuan in AI R&D.

With the pressure-based R&D investment, Baidu has achieved certain results in AI technology and applications. According to the fourth quarter and full-year financial reports of last year, Baidu Smart Cloud grew strongly in the fourth quarter, with revenue increasing by 26% year-on-year. In 2024, smart cloud AI-related revenue increased by nearly 3 times year-on-year. At the same time, as of December 2024, Baidu Wenku's AI function's monthly active users reached 94 million, a month-on-month increase of nearly double, ranking first in China and second in the world, second only to Microsoft Copilot. Baidu Wenku has thus become the only Chinese AI application to achieve commercial scale effect.

The R&D investment of over 100 billion yuan has also brought Baidu's full-stack layout in the artificial intelligence chip layer, framework layer, model layer and application layer, and promoted Baidu to become one of the few artificial intelligence companies in the world that have full-stack layout.

Recently, Baidu Smart Cloud successfully illuminates the third-generation Wanka cluster of Kunlun Core, which is also the first self-developed Wanka cluster officially illuminated in China. In the future, Baidu Smart Cloud will also light up the 30,000-karat cluster.

"From hardware to software, the technical challenges of the domestic Kunlun core Wanka cluster and the future 30,000 kcal cluster are all-round." According to Baidu internal personnel, Baidu Baige AI heterogeneous computing platform 4.0, which was upgraded in September 2024, has played a crucial role in the construction of Wanka cluster, and it currently has the mature deployment and management capabilities of 100,000 kcal clusters.

The completion of self-developed chips and Wanka clusters not only brings strong computing power support to Baidu's AI battle, but also effectively improves the overall utilization rate of Baidu and its service customers, and produces a series of chemical reactions in reducing the cost of large model training.

Among the current cloud platforms deploying DeepSeek, the above technological achievements have also become Baidu Cloud's core advantage in attracting users. “Based on our full-stack AI capabilities and end-to-end optimization, we canTo ensure that any model hosted on the platform has the best performance and stability while maintaining a high level of competitiveness. "Baidu internal personnel introduced.

The change in the daily average API call volume intuitively illustrates the growth status of customers who choose Baidu AI services. In the past year, the daily average API call volume processed by Baidu Wenxin series models has soared from 50 million to 1.65 billion, an increase of 33 times.

As the cost of large-scale model calls decreases, the above-mentioned daily average API call volume is expected to explode further.

"Today, the innovation speed is much faster than before. In Moore's Law, every 18 months, performance doubles and price halved; today, the cost of large-scale model inference can be reduced by more than 90% per year. "In a recent interview with Li Yanhong said that looking back on the past few hundred years, the essence of innovation is cost reduction and productivity improvement. In AI, IT and even more fields, most innovations are related to cost reduction. For example, if the cost is reduced by a certain proportion, productivity can also be increased in the same proportion.

The investment of hundreds of billions of funds and the cost reduction brought by investment will jointly constitute Baidu's confidence and confidence in chasing AGI.