On February 18, Baidu, a prominent Chinese technology company known for its search engine, revealed its financial performance for the fourth quarter and the entire year of 2024. The company's annual total revenue reached an impressive 133.1 billion yuan, while the net profit attributed to Baidu's core business soared to 23.4 billion yuan, marking a 21% year-on-year increase. For the fourth quarter alone, the revenue was quantified at 34.1 billion yuan, with core income standing at 27.7 billion yuan.

Among the various business sectors, Baidu's intelligent cloud services emerged as a noteworthy performer. In the fourth quarter, the intelligent cloud segment experienced a remarkable growth rate of 26% compared to the previous year, with the revenues from AI-related services nearly tripling. This significant expansion has garnered recognition from the market regarding the value of Baidu's intelligent cloud business.

According to Baidu's founder and CEO, Robin Li, 2024 represents a pivotal year for the company as it transitions from an Internet-focused approach to an artificial intelligence-driven strategy. As their full-stack AI technology gains widespread acceptance in the market, the momentum of intelligent cloud growth continues to surge. Li emphasized the importance of AI-driven reconstruction of the search ecosystem, enhancing user experience, and underscored the conviction that AI-related investments would yield greater returns by 2025.

The unexpected success of Baidu's AI and cloud business is undoubtedly the highlight of their performance in 2024, showcasing impressive metrics that underscored the strength of these two sectors. For instance, the AI revenue from Baidu's intelligent cloud division showed a staggering increase of nearly threefold year-on-year. By December 2024, the daily API call volume of the Wenxin series models had skyrocketed to an astounding 1.65 billion calls, a phenomenal increase compared to just 50 million calls during the same period in the previous year. Additionally, the month-over-month active user count for Baidu's AI features within its document library hit 94 million, witnessing an impressive 216% year-over-year growth and an 83% increase quarter-on-quarter.

Regarding market dynamics, statistics revealed that the number of project bids related to large models in China surged to 910 in 2024, representing a 4.5-fold increase from 2023, with the total bid amount reaching 2.52 billion yuan, four times that of the previous year. Significantly, Baidu took the lead among bidding units across three dimensions: the number of bids, the breadth of industry coverage, and the number of central and state-owned enterprise projects, with clients including key national companies like China National Electricity and the All-China Federation of Trade Unions. Furthermore, in the recently published general large model vendor bidding rankings, Baidu secured two first-place positions.

This growth is closely linked with Baidu's extensive AI technology deployment. The company had pioneered the Wenxin large model back in 2019, a time when the concept of generative AI was still largely unfamiliar to the masses. As the global tide of generative AI gained momentum in 2023, Baidu rapidly introduced the Wenxin Yiyuan and continued to stay ahead of competitors by constructing a powerful matrix of large models. By the third quarter of 2024, Baidu launched two new models – the flagship version 4.0 and a faster version known as Turbo – along with enhanced lightweight models called Speed Pro and Lite Pro. Recently, it was announced that Wenxin Yiyuan would be offered for free, with plans to release the Wenxin large model 4.5 series over the coming months, officially transitioning to open source by June 30.

In the realm of application deployment, Baidu is making strides in addressing technological bottlenecks associated with implementing large models. For instance, the recently unveiled iRAG technology significantly mitigates the hallucination effect often seen in image generation, consequently enhancing productivity in image creation. Furthermore, Baidu’s Qianfan ModelBuilder toolchain is modernizing model training methodologies. The Qianfan large model platform now supports various Wenxin models, alongside hundreds of mainstream large models from both domestic and international sources. This platform guarantees comprehensive post-training services, including SFT, data labeling, and model preparation, enabling enterprises to efficiently fine-tune dedicated large models tailored to specific industry use cases.

It's important to emphasize Baidu's unique position as one of the few global players possessing comprehensive full-stack AI capabilities. The company's self-developed four-layer AI architecture and end-to-end optimization capabilities allow Baidu to achieve high reliability, security, and performance for its models at a lower cost. For instance, the recently launched Kunlun chip third-generation WAN card cluster marks a pivotal achievement, being the first such self-developed cluster in China officially operational, with plans to scale up to a 30,000-card cluster.

Moreover, Baidu's proficiency in managing and deploying WAN card clusters means that it can further drive down model training costs. The company has demonstrated a mature capability to manage and oversee ten-thousand WAN card clusters, achieving over 99.5% effective training time. Additionally, the platform controls the efficiency loss in mixed-chip training to below 5%, placing it among the industry's leaders. Baidu's PaddlePaddle framework stands out as the most widely utilized open-source framework among Chinese developers, while its rich matrix of Wenxin models will see the introduction of several new releases throughout the year, enhancing multi-modal capabilities significantly.

With a bold approach in applying AI technology, Baidu’s search engine and online document library boast nearly 100 million AI users, illustrating the company’s unwavering commitment to innovating its AI applications.

During Baidu's earnings conference on February 18, Robin Li shared groundbreaking updates, mentioning that the Wenxin model 4.5 would be open-sourced and hailed it as the most formidable large model Baidu has ever created. Previously, Wenxin Yiyuan operated under a monetized model akin to ChatGPT, but Baidu has recently declared that the entire service would now be free, reflecting a strategic shift that prioritizes accessibility and engagement.

This development signals a broader recognition of the potential for open-source models, particularly platforms like DeepSeek, to significantly accelerate the proliferation and practical application of AI technology. Li highlighted that the success of DeepSeek illustrated the advantages of making the finest models accessible, enhancing experimentation and expanding their impact across diverse scenarios.

Baidu's shift towards open-sourcing large models will reduce entry barriers for developers and enterprises, allowing for low-cost and no-threshold access to advanced AI technologies. This strategy not only enhances Wenxin's market presence but also forms a dual driving force for its business model: "open-source attraction + cloud service monetization." By attracting a wider pool of developers into the ecosystem, Baidu aims to build a robust feedback loop where data accumulation will directly influence model optimization, fostering user loyalty and solidifying market standing, culminating in a virtuous cycle of technology openness, ecosystem growth, and business realization.

As this self-reinforcing dynamic takes shape, it undoubtedly positions Baidu favorably for substantial advancements in AI applications, laying a solid foundation for future business model innovations, and potentially establishing a replicable paradigm within the industry.