Baidu created a "Daniel Wu"

Author | Zhou Zhiyu

Editor | Zhang Xiaoling

When a digitally replicated "Daniel Wu" begins to serve as your exclusive English coach, a brand new species is born. This is not a trailer for a science fiction movie, but a new business card presented by Baidu for its intelligent cloud strategy at the 2025 Baidu Cloud Intelligence Conference.

Behind "Daniel Wu" is the integrated application of Baidu's self-developed end-to-end voice semantic large model and the "Hui Bo Xing" digital human technology. It is a clear signal: Baidu is fully promoting AI from the background to the foreground, evolving from complex "technically usable" to universally perceptible "product usability."

This exquisite digital human is just the tip of the iceberg. Its birth stems from a profound self-revolution within Baidu Intelligent Cloud.

At the conference on August 28, Baidu Group Executive Vice President Shen Dou set the tone for this transformation: "The intelligent economic era calls for AI cloud with intelligence as a priority." He bluntly stated that AI cloud is shifting from a cost center that incurs expenses to a new profit center that generates revenue. Consequently, Baidu Intelligent Cloud's strategy has shifted from scale dominance to efficiency priority.

Behind this rhetoric is a reconstruction of the value system of cloud computing. In the past, cloud vendors competed on who had more servers and cheaper storage, which was a war about scale. Baidu believes that the second half of the AI era is a competition over whose intelligence is smarter, more efficient, and better able to create value for customers' profit statements. This is a war about efficiency.

To win this new war, Baidu has unveiled its two core engines—Bai He and Qian Fan.

Bai He 5.0 is the driving engine of this revolution, a powerful foundation built by Baidu Intelligent Cloud for the AI era. In today's mainstream computing paradigm of integrated reinforcement learning training and inference, mere accumulation of computing power is a thing of the past; the core lies in "efficiency-to-cost ratio." The upgrade of Bai He is a comprehensive solution to the efficiency challenges of AI computing power across four major areas: network, computing power, inference systems, and integrated training and inference.

The newly launched Kunlun Chip Super Node significantly enhances computing power density and inference efficiency, powerful enough to allow anyone to run trillion-parameter open-source models in just a few minutes using a single cloud instance. This extreme computing power has also attracted cutting-edge institutions like the Beijing Humanoid Robot Innovation Center, whose released embodied world model is built on Bai He, resulting in a twofold increase in research and development efficiency. At the same time, through partnerships with Intel and others, integrating the new generation Xeon 6 processors and Gaudi accelerator cards into the platform, Bai He showcases an open and powerful hardware ecosystem.

While Bai He 5.0 provides surging energy, Qian Fan 4.0 serves as the precise and busy "Agent Factory" on the ground. Baidu does not hide its ambitions, directly defining it as the industry-leading "Agent Infra," a platform specifically designed for nurturing and deploying Agents.

The brilliance of this factory lies in its attempt to solve the most challenging issues enterprises face when building their own Agents: making models understand you better, enabling Agents to be perceptive, turning data into fuel, and allowing the Agent legion to strike The Qianfan model library has been upgraded to over 150 models, encompassing cutting-edge video generation models like "Baidu Steam Engine," and introducing specialized models such as Qianfan Huijin for the financial industry and Qianfan VL for visual understanding, capable of surpassing the performance of trillion-parameter models on specific tasks with a scale of tens of billions of parameters.

At the same time, the newly released RFT (Reinforced Feedback Tuning) toolchain allows enterprises to achieve ideal model tuning results with only one-tenth of the previous data volume (reducing from thousands of entries to hundreds), significantly lowering the technical and data barriers.

In the past, large models were like bookworms living in historical data. The "Baidu AI Search MCP Server," exclusively launched by Qianfan, acts like an antenna connecting Agents to the real world, enabling them to access real-time, authoritative information and escape the illusion of nonsense.

To address the data governance challenges of the large model era, Baidu officially launched the Qianfan Data Intelligence Platform, DataBuilder. It helps users efficiently manage multimodal data, achieving astonishing results such as a 600% increase in processing efficiency, a 30% reduction in computing costs, and up to an 80% decrease in retrieval and storage costs.

In the face of complex tasks, Qianfan has added a multi-agent collaboration mode, allowing different Agents to work together like an organization, breaking through the capability bottleneck of a single intelligent agent.

As this factory operates at full capacity, we see the birth of more new species beyond the "Daniel Wu Digital English Coach." In the factory workshop of SANY Energy, a visual large model named "Yijian" is becoming the "AI Master" on the production line. By simply uploading a standard operation video, an SOP inspection task can be generated in minutes. If a worker makes an operational error, the system will immediately alert, and managers can see the details in real-time, significantly reducing the time for manual on-site inspections.

More radical changes are happening in the coding world. Baidu provides weapons for two types of people: for ordinary people who do not understand coding, it launched the "Miaoda" platform. A development task that used to cost tens of thousands and take weeks can now be compressed to 10 minutes and under 5 yuan.

Since its launch, users have generated over 250,000 applications on the platform, with daily active applications growing over 500%. For professional programmers, it offers "Wenxin Quick Code," a tool positioned as an enterprise-level intelligent R&D productivity engine.

According to Zang Zhi, director of Baidu's Engineering Efficiency Department, AI can now contribute over 30%–40% of code output in enterprise R&D, and in the future, it will free up 80%–90% of developers' energy for creative work.

This series of radical investments and transformations raises a crucial question: Where does the confidence come from?

The answer lies in Baidu's financial report. Just before this conference, Baidu's mid-year report showed that the AI new business built on intelligent cloud and autonomous driving has seen quarterly revenue surpass the 10 billion yuan mark for the first time, growing at a rapid pace of 34% year-on-year. This rapidly growing business has built a solid safety net for Baidu's intelligent cloud evolution and transformation.

The market is also betting real money on Baidu's new path. This is not a PPT vision but a commercial practice already recognized by the core forces of the Chinese economy Data shows that over 65% of central enterprises, 80% of systemically important banks, and 95% of mainstream car manufacturers have chosen Baidu Intelligent Cloud when implementing large models. The case of Zhilian Recruitment is even more direct: based on the Qianfan platform, its job matching solution not only reduced reasoning costs by 70%, but also significantly shortened response time from the original 14 seconds, while improving transaction conversion rates.

Shen Dou said that AI is entering a true "super cycle." Baidu's gamble is precisely on the future form of this cycle. From the fleeting glimpse of "Daniel Wu" at the card table to the massive AI infrastructure roaring behind him, Baidu is proving through action that it no longer wants to be a patcher of the old map; what it wants to be is a rule maker of the new world