laitimes

Has Baidu's AI medicine worked?

Has Baidu's AI medicine worked?

City Boundary

2024-05-18 07:00Posted on the official account of Beijing Municipality

Recently, Baidu's life has been full of internal and external troubles.

First of all, his own public relations No. 1 "exposed" the wolf culture, creating a public opinion crisis for his company. Although Cui Shanshan, Baidu's senior vice president, reaffirmed Baidu's "simple and reliable" values in an internal meeting a few days later, Baidu's public image has undoubtedly been hurt.

When the "internal troubles" are not solved, competitors have "new": OpenAI, Google, Byte, Huawei and other large model competitors have gathered to iterate on new products, and AI applications have accelerated into the deep water area.

More than ever, Robin Li needs AI to deliver a satisfactory answer sheet and inject a shot of strength into Baidu.

Has Baidu's AI medicine worked?

Before the U.S. stock market on May 16, Baidu announced its unaudited financial results for the first quarter of 2024. AI, as expected, has become the keyword that Baidu struggles to export in its financial report.

At the evening earnings conference, Baidu founder Robin Li disclosed several positive figures: 11% of Baidu's search results have been generated by AI; Generative AI contributes 6.9% of revenue to Baidu's intelligent cloud; AI has helped Baidu monetize its ads, bringing hundreds of millions of yuan in incremental revenue every quarter since the second half of last year.

However, the market still shows a wait-and-see attitude towards Baidu's AI medicine.

After the earnings report, Baidu's U.S. stock fell briefly after the open, falling as low as -4.9%, and then gradually recovered, and the stock price rose 1.66% as of the close.

1. Profits are inseparable from reducing costs and increasing efficiency

At present, maintaining profits and cash flow is the top priority of Baidu's strategy.

After all, Baidu's current core search product, DAU, has weak user stickiness; Sluggish advertising growth amid economic pressure; AI is still in the early stages of commercialization and requires long-term investment.

According to the financial report, in the first quarter of this year, Baidu achieved revenue of 31.513 billion yuan. net profit attributable to Baidu (NON-GAAP) was 7.011 billion yuan, a year-on-year increase of 22%; NET PROFIT ATTRIBUTABLE TO BAIDU'S CORE (NON-GAAP) WAS 6.628 BILLION YUAN, A YEAR-ON-YEAR INCREASE OF 26%. However, under GAAP caliber, Baidu's net profit in the first quarter became 5.446 billion yuan, down 6% year-on-year - the market believes that the profit change is mainly affected by investment behavior.

In order to ensure that profits are not lost, while actively transfusing blood for AI, Baidu has not stopped "reducing costs and increasing efficiency".

According to the financial report, Baidu's capital expenditure in the first quarter increased by 56% year-on-year, and a large part of it should be invested in AI hardware construction. At the same time, Baidu's core R&D expenses fell by 1%; Option incentives fell 24% over the same period. It is reasonable to guess that Baidu's personnel costs in the last quarter are likely to shrink.

Has Baidu's AI medicine worked?

Of course, the biggest pressure on Baidu at this time still comes from the lack of "hematopoietic" engine.

According to the financial report, the "online marketing revenue" reflecting Baidu's advertising revenue was 17 billion yuan, an increase of only 2.7% year-on-year. In contrast, Tencent's "online advertising", which announced its financial results in the same period, benefited from the video account, and its revenue increased by 26% year-on-year.

Advertising growth has hit a bottleneck, and Baidu's explanation for this is that advertisers in real estate and vertical sectors have declined in their willingness to invest in the first quarter. However, as this year's launch gate, the stimulation of AI advertising to Baidu is not enough to fill this gap.

Although Robin Li said at the earnings briefing: "At present, 11% of the search results on Baidu search have been generated by AI. However, according to overseas reports, AI has been able to directly contribute about 10% of the company's advertising revenue to major advertisers Meta and Google.

According to market insiders, according to their observation in the first quarter of this year, the Kimi model released by the dark side of the moon was widely distributed on various platforms: "For a period of time, whether it was Sowen Xin Yiyan or Sotongyi Qianwen and Zhipu AI, the first one to appear on Baidu was Kimi. ”

Baidu, which is competing to enter the big model, can tolerate competing products to override its own products, which also confirms Baidu's advertising anxiety.

2. AI needs big customers

In addition to keeping profits "cash cow", the core business that can truly reflect Baidu's AI strength is still on the "Baidu Intelligent Cloud". According to the financial report, the revenue of intelligent cloud in the first quarter was 4.7 billion yuan, a year-on-year increase of 12% - the proportion of revenue contributed by generative AI accounted for 6.9% of it.

However, the revenue of hundreds of millions of yuan is not enough to resolve Baidu's competitive anxiety. Competitors continue to catch up, according to the research agency "Similarweb" data in April, Wenxin Yiyan's average daily visits ranked second in China.

Has Baidu's AI medicine worked?

▲(图源/Similarweb)

In order to greatly increase cloud business revenue, it is necessary to find a way to start from the B layer. Especially this year, with the entry of all players in the basic layer of the large model, the AI competition has entered the stage of "customer acquisition" from volume computing power and volume technology.

In April this year, with the escalation of the competition, the Chinese and American industries staged a fierce "open and closed source" dispute at the same time. Baidu, which was the first to release and the first to charge fees in China, has begun to encounter more severe challenges to whether it can maintain its leading position.

In short, the "open and closed source" controversy means that the next stop of the big model competition basic logic - open source model, is conducive to elevating the ecology; The closed-source model, on the other hand, relies on API calls, which is more secure, but is more conducive to the platform's charging and cost recovery.

In March this year, for the closed-source OpenAI, Musk open-sourced his large model Grok-1. In April, Meta released a new generation of open-source Llama 3 models. Google, which originally chose closed source, has also recently launched the open source Gemma 2 model, and open source and closed source go hand in hand.

In China, the open source model is represented by Alibaba Cloud, and with the model of "cash + computing power", it has invested in star startups such as the dark side of the moon, Minimax, and Baichuan Intelligence. In the "Big Model Five Tigers", Zero One Ten Thousand Things and Baichuan Intelligence have also chosen open source. Standing with Baidu, there are only a few players such as Zhipu AI in the first-line companies.

Has Baidu's AI medicine worked?

Is Baidu's closed-source choice correct? Since the beginning of this year, almost every time Robin Li appears, he has stood in the center and called for the industry to "roll up". On April 11, he said in an internal speech: "The open source of large models is of little significance, the technology is backward, and the closed source can get through the business model." ”

But there are many opponents. Zhou Hongyi, the founder of 360, quickly retorted: "After open source has an ecosystem, the development will be very fast. The company that said this (referring to Baidu) itself has grown to this day with the help of open source. ”

Zhu Xiaohu, managing partner of GSR Ventures, also publicly said: When the iterative curve of closed-source technology slows down, open source will go up. He "will not invest in China's pedestal model startups".

According to insiders revealed to "City Boundary", Robin Li was very dissatisfied with Zhou Hongyi's expression within Baidu. However, how Baidu can open up the "customer acquisition" situation in the closed source is still full of challenges.

According to Baidu, as of April 9, 2024, Baidu's AI app store has more than 400,000 monthly active users; In an effort to retain customers, Baidu has also preset more than 10 mainstream open source models in the large model, including Llama, Baichuan, and ChatGLM; It is convenient for users to call more flexibly.

However, according to the "city boundary", the industry believes that the price of using the "agent" provided by Baidu's large model is still high, and for small companies, choosing many of Baidu's services requires an investment of between hundreds of thousands and millions, dissuading some startups that are interested in testing the waters of AI generation.

3. Difficult to solve traffic anxiety

Search products and advertising revenue are constantly diverted by short video platforms; AI needs to be more widely implemented, as well as cheaper customer acquisition costs, and Baidu's anxiety about traffic is particularly obvious at the moment.

Recently, Zhou Hongyi also said in an interview with the media: The large model needs to be pulled off the altar and enter hundreds of industries and households; It also needs to be smaller and smaller, and the model needs to be on a mobile phone, a computer, and a car to find a vertical scene. According to "City Boundary", Zhou Hongyi has set up a new AI-related live broadcast and marketing team for Zhou Hongyi within 360 to create a personality and drive traffic.

To a certain extent, the unexpected outbreak of the "Xuan Jing Incident" also reflects Baidu's urgent need for traffic and "wolf" combat effectiveness.

"City Boundary" understands from within Baidu that since the beginning of this year, Baidu's public relations department has been required by all employees to make short video accounts, and there is a heavier KPI assessment. However, Jingjing's radical approach has also encountered internal challenges. A Baidu employee told "City Boundary" that around April, Cui Shanshan personally asked Xuan Jing about the short video made by the team, "Why can all the rules exceed 10,000", and asked her to explain.

Although after the public opinion incident, Baidu once again reaffirmed the culture of "simple and trustworthy" at the annual advanced awards event; However, Baidu's traffic anxiety will not be easily resolved with Xuan Jing's resignation being "approved in seconds".

In the midst of internal and external troubles, Baidu has also formulated a more urgent profit schedule for the "second growth curve" that has given space for development.

Has Baidu's AI medicine worked?

On May 15, Baidu Apollo released the sixth-generation unmanned vehicle of "Radish Run" in Wuhan. Wang Yunpeng, president of Baidu's intelligent driving business group, said at the meeting: Tesla wants to achieve the current level of Baidu's fully unmanned operation, which needs to be as little as 3 years and as much as 5 years.

However, according to the "Delay" report, Baidu has set a profit schedule for "Carrot Run" internally. In April this year, Wang Yunpeng said in an internal letter: first there will be 100 million revenues, and then 100 million profits.

In addition, the outside world also expects Baidu to make greater efforts to buy back shares to boost market confidence.

According to the financial report, in the first quarter of 2024, Baidu's repurchase amount will be 229 million US dollars, while in the same period, Alibaba announced a repurchase of 4.8 billion US dollars in the financial report, and Tencent said that it will repurchase shares for 14.8 billion Hong Kong dollars. There is speculation that Baidu may want to leave more cash flow for AI investment at this time.

Author | Dong Wenshu Zhao Zikun

Edit | Li Yuan

Operations | Liu Shan

View original image 25K

  • Has Baidu's AI medicine worked?
  • Has Baidu's AI medicine worked?
  • Has Baidu's AI medicine worked?
  • Has Baidu's AI medicine worked?
  • Has Baidu's AI medicine worked?

Read on