In the third quarter, Baidu has already tasted the sweetness of AI | Frontline

Text | Zhou Xinyu

Editor | Su Jianxun

In 2023, Baidu released its third quarter financial report-this is also the first quarterly answer sheet submitted by Baidu after obtaining the large model service license in August.

The financial report shows that in the third quarter of 2023, Baidu Inc’s total revenue was 34.4 billion yuan, a year-on-year increase of 6%; Baidu Core’s revenue was 26.6 billion yuan, a year-on-year increase of 5%.

At the same time, thanks to continued cost reduction and efficiency improvement, net profit attributable to Baidu Group turned a profit year-on-year, with net profit reaching 6.7 billion yuan; net profit (non-GAAP) attributable to Baidu Core reached 7 billion yuan, It increased by 21% year-on-year, and profit margin increased by nearly 4% year-on-year.

Behind the good financial performance, Baidu’s rapid progress in AI commercialization in the third quarter is inseparable.

On the one hand, Wenxin Big Model 4.0 is online. Baidu has transformed all the original mobile ecosystem products such as search, office, and marketing based on AI and launched intelligent assistants.

On the other hand, as one of the first large model manufacturers to obtain service licenses in August, Baidu launched the first batch of self-developed AI native applications based on Wenxin large models, which have been implemented in Baidu Netdisk, Ruliu, GBI, Zhicai, etc. Scenes.

The results of the AI-based transformation of products are reflected in increased users’ willingness to pay, as well as in significantly increased online revenue. Take Baidu Wenku as an example. Baidu data shows that after the launch of functions such as smart text and PPT generation, Baidu Wenku’s payment rate increased significantly. The financial report shows that Baidu’s core online marketing revenue was 19.7 billion yuan, a year-on-year increase of 5%.

During the third quarter earnings conference call, Robin Li, founder, chairman and CEO of Baidu, also said that the advertising system connected to Wenxin’s large model will bring hundreds of millions of yuan in incremental revenue in the fourth quarter.

For many years, Baidu has been placing its hopes on smart cloud services for its second growth curve besides online revenue, but the results have been mediocre. The explosion of large models this year has made AI services a turning point for Baidu to leverage intelligent cloud services.

For cloud vendors, the emergence of large models means that the era of selling computing power alone has passed. From training, deployment, to tool chain provision, customers need cloud vendors to proactively lower the threshold for large models and directly provide model and AI application capabilities or services.

In the third quarter, Baidu Intelligent Cloud’s action was to launch AI application development tools and build a developer ecosystem. On the one hand, the Qianfan large model platform has launched an AI native application development platform to provide application development frameworks and tools; on the other hand, Qianfan has launched an AI native application store to share revenue with the application developers who settle in the store.

Up to now, 42 domestic and foreign mainstream large models, including Llama2, have settled on the Qianfan platform, and more than 20,000 companies have developed applications by calling APIs of different models. The financial report shows that Baidu’s non-online marketing revenue in this quarter was 6.9 billion yuan, a year-on-year increase of 6%.

However, if cloud vendors want to reap the dividends of large models, they still have to go through a long dormant period.

On the one hand, the high training and inference costs of large models mean that cloud vendors have to burn more money while serving customers. On the other hand, due to limited model capabilities and long delivery cycles, the industry implementation of large models is still in its infancy, and it is difficult to bring the first pot of gold to cloud vendors.

Recruiting talents and purchasing GPUs are long-term investments for Baidu. The third quarter balance sheet shows that short-term investment in this quarter (156.359 billion yuan) is almost the same as the previous two quarters, which is more than 30 billion yuan more than in the last quarter of 2022 (120.839 billion yuan).

In order to enrich its profit model, Baidu launched Wenxinyiyan Professional Edition in November, which charges 59.9 yuan per month. At the same time, the scale effect has also brought about a reduction in model calling costs – in the same month, Qianfan Large Model Platform updated the token calculation logic. The number of tokens corresponding to the same number of Chinese characters was reduced by more than 20% in the new version compared with the old version. For customers, this means cheaper model service.

This quarter, large models have allowed Baidu to initially taste the sweetness of growth. But if you want to achieve more stable growth, continue to iterate model capabilities and tap user needs, it is a long-term front for Baidu.

Welcome to communicate!