Author: Wenzhi
Editor: Xiaodi
Baidu in 2023: Buff is full, AI takes off.
Baidu's annual revenue in 2023 reached 134.6 billion, hailed as a textbook example of AI making money.
On February 28, 2024, Baidu officially released its financial report for the fourth quarter of 2023 and the whole year, shocking the world. To everyone's surprise, while others were burning money to develop large models, Baidu was making a lot of money.
According to the financial report, Baidu's total revenue in 2023 reached 134.6 billion yuan, with a net profit attributable to Baidu of 28.7 billion yuan, a year-on-year growth rate of 39%, marking the largest increase in net profit in three years. Li Yanhong publicly disclosed for the first time that the incremental revenue brought to Baidu Intelligent Cloud by large models reached 660 million yuan, and even stated that this incremental revenue is expected to reach tens of billions of yuan in 2024.
At the same time, Baidu's R&D expenses in 2023 reached 24.2 billion yuan, an increase of 4% year-on-year. This growth is mainly due to Baidu's massive support for generative AI research and development, including a significant increase in server depreciation expenses and server hosting fees.
In addition to the increase in R&D expenses, Baidu also disclosed a key piece of information in the financial report, which is the amount of share repurchases. According to the financial report, since the fourth quarter of 2023, Baidu's share repurchase amount was 318 million US dollars, and the cumulative repurchase amount under the 2023 share repurchase plan reached 669 million US dollars.
In the view of Baidu's founder and CEO, Li Yanhong, the steady development of Baidu's core business has laid the foundation for future innovation. In particular, the continued investment in generative AI and basic model fields has yielded positive returns, and Baidu will continue to steadfastly invest in research and development.
It can be foreseen that 2024 will be a crucial year for Baidu's AI layout and a significant year for the large model race.
Crushing OpenAI, Google, and becoming the world's top AI money-making enterprise
Looking globally, how are AI giants such as OpenAI, Meta, Microsoft, and Google performing in terms of profits?
Recently, during the financing process, Sam Otterman revealed that OpenAI's monthly revenue is roughly between 10-15 billion US dollars, but the monthly cost expenditure exceeds 30 billion US dollars. This puts OpenAI in an extremely loss-making state, with a loss of over 10 billion US dollars per year.
Similarly, Meta, which also released its financial report in February, has seen a net profit surge of 201%, and revenue growth hit a new high. However, the departments truly involved in AI and large models at Meta have a net loss of over 34%, showing negative growth, which is more pronounced than last year.
Microsoft also recently released its financial report, and from the relevant data, despite the overall strong financial situation, the parts of Microsoft truly involved in AI and large model fields are actually not profitable and are still in a slight loss. However, Microsoft can use other businesses such as cloud services to offset the losses in the AI field.
In stark contrast, Google is in a slightly profitable state. Strictly speaking, Google is similar to Baidu, both relying on AI and large model services to drive the growth of cloud services and generate profit. Therefore, from the overall environment, the only company truly making money from AI and large models, and making the most money, is Baidu.
In fact, many companies have coincidentally discovered that it is difficult to develop large models if they do not have a strong financial foundation. This is because strong large models inevitably entail huge cost consumption, coupled with the Western GPU ban, which further exacerbates the financial difficulties of companies in this environment.
With strong technological capabilities, Baidu has chosen a more efficient path.
Even before ChatGPT exploded onto the market, Baidu had already comprehensively laid out the four essential elements of large models: chips, frameworks, large models, and terminal applications. It is precisely because of this that the various costs of Baidu's large models are continuously and significantly reduced.
Data shows that within less than a year of the launch of Wenzhi's large model, the inference cost has already been reduced by 99%, directly dropping to 1% of the original cost. The model's inference performance has increased by 50 times compared to the past. In addition to inference, training efficiency has also greatly improved. Through the collaborative optimization of PaddlePaddle and Wenzhi, the weekly average training efficiency of Wenzhi's large model exceeds 98%, the effective bandwidth of training GPUs reaches 95%, and the training algorithm efficiency has increased by 3.6 times, significantly reducing waste during the training process.
From this perspective, Baidu's success is not achieved through a sprint, but a marathon.
The AI boom is in full swing, and in 2024, "profit" means "win"
Regarding Baidu's 2023 financial report, Li Yanhong summarized it in two words: "steady development."
As an early entrant in the domestic large model field, Baidu has shown steady progress in both user technology and API usage. From the official release of Wenzhi's large model 1.0 version in March 2019, after four years of research and development iterations, to the release of Wenzhi's large model 4.0 version in October 2023, as of the end of last year, the user base of Wenzhi's large model has exceeded 100 million. Recently, Samsung, Honor, and Autohome have also announced strategic partnerships with Baidu.
As a large model rooted in the Chinese market, Wenzhi's large model has achieved excellent results in generating content in the Chinese context, giving it a first-mover advantage. In December alone, approximately 26,000 companies called Wenzhi's large model. If estimated at 50 million calls, this means that companies call Wenzhi's large model over 34,000 times per minute.
As early as November last year, Li Yanhong stated that the true value of the large model era lies in native applications. Therefore, Baidu's exploration of AI native applications has never stopped. Baidu has used large models to reshape all its applications and promote the popularization of AI applications in various industries. PaddlePaddle is a typical example, with over 200,000 companies and over 800,000 applications on the platform, effectively promoting the popularization of China's AI industry. In addition, Baidu's cloud service tools help companies quickly deploy and use large models, and through competitions, incentivize AI application innovation, all of which are widely recognized.
It can be said that this series of strategic decisions has enabled Baidu to achieve today's results and seize the initiative in AI. This also precisely demonstrates that Baidu's current strategy is correct, proving that Li Yanhong has indeed found a viable path for Baidu.
As the saying goes, "Bend down to plant a seed, stand up to feel the vastness of the world." What achievements Baidu will make in the future in the field of AI is worth looking forward to.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
