There is no standard answer for a giant's transformation; Baidu has found its own rhythm.

In July this year, Sun Tianxiang, born in 1997, officially took over as the head of Baidu's Foundational Model R&D Department and joined the Model Committee. When the news broke, the industry's reaction was calmer than expected.

Over the past year or so, similar personnel adjustments have become so frequent at Baidu that people have gradually grown accustomed to the changes happening at this veteran internet company.

The changes are occurring systematically, from the grassroots to the core decision-making level. From frontline technical team leaders to business group managers, from the finance system to human resources, younger faces are stepping into the spotlight.

Outsiders like to summarize this perception by saying, "Baidu has become younger," but youth is not just about replacing a few individuals. It signifies a reconstruction of decision-making logic, a shift in resource allocation, and even a deliberate departure from the company's past successful paths.

The significance of this is often underestimated in the context of China's internet industry.

Most major companies that experienced glory during the PC era now bear, to varying degrees, the imprints of that time: rigid reporting hierarchies, path dependency on mature businesses, and a reaction speed to new things that is cautious to the point of sluggishness.

The "old-school vibe" people joke about is essentially the organizational inertia formed when a company stays in its comfort zone for too long.

Baidu is attempting to break this inertia. But what is the cost of breaking it, and where does this path ultimately lead? No one has a standard answer.

Just as Microsoft pivoted to the cloud and Google went all-in on AI back in the day, all giants initially feel their way forward when they turn.

01
Letting the Young Take Center Stage

Sun Tianxiang's appointment as the head of Baidu's Foundational Model R&D Department was a personnel move worth pondering in the industry in the first half of this year.

Those in the circle are not unfamiliar with Sun Tianxiang. During his Ph.D. at Fudan University, he led the complete R&D engineering of the MOSS large model, making him a bona fide young technical backbone in the field of large model engineering.

Even so, directly taking charge of a core model R&D department at a trillion-dollar company exceeded many people's expectations.

Large models are the technological lifeblood of all tech companies today and are typically overseen by highly experienced technical executives. Yet Baidu entrusted this responsibility to a young person under 30.

But this is not an isolated case at Baidu.

By the end of 2025, Baidu split its two major model R&D lines, establishing a Foundational Model R&D Department and an Applied Model R&D Department, led by Wu Tian and Jia Lei respectively, both reporting directly to Li Yanhong.

Wu Tian is a technical backbone who grew up within Baidu, while Jia Lei is a seasoned expert in voice technology. By mid-2026, Sun Tianxiang took over the foundational model line, and Wu Tian shifted her focus to broader technological strategy.

The rejuvenation of business lines is also progressing.

Ping Xiaoli, born in 1985, now oversees the entire commercialization system of the mobile ecosystem, including the merged E-commerce and Business Division.

This female general who rose from intern to leadership is a witness to Baidu's journey from the search era to the AI era and is now a core executor of Li Yanhong's AI commercialization efforts.

Even the backend support systems are adjusting simultaneously.

He Haijian, former CFO of Kingsoft Cloud, took over as Group CFO, bringing financial experience from the cloud industry; He Junjie, the former acting CFO, rotated to oversee human resources, placing a manager who better understands the business in charge of organizational matters, aiming to align the HR system with the new business rhythm; Cui Shanshan, who has worked at Baidu for over two decades, stepped back from specific operational tasks to focus on the Culture Committee.

Through these advances and retreats, Baidu has achieved a smooth transition in its management ranks.

Many people dismiss these adjustments as merely a "blood transfusion," but it's not that superficial.

The core is not about age numbers but changes in decision-making logic.

Li Yanhong has repeatedly mentioned internally that people's creative peak is around thirty, and he himself is already outdated. While such a statement might sound like a platitude from another entrepreneur, at Baidu, this logic is genuinely being implemented.

When there are disagreements on technical routes, the ideas of young frontline researchers are often prioritized for trial, with adjustments made if they fail, rather than decisions being imposed top-down.

This is easier said than done. In most mature companies, management logic dictates that seniority determines discourse power (influence) and hierarchy determines decision-making power.

A young employee's idea must pass through multiple layers of reporting before reaching the core level, often missing the optimal time window long before.

Many companies are not lacking in young talent; rather, the voices of the young are not heard, and their ideas do not materialize.

Baidu still retains many traits of early-stage internet companies. People call each other by their first names without excessive hierarchical titles; flexible working hours are in place without formalistic attendance checks; Li Yanhong himself, when present at the company, always dresses casually, and employees call him Robin without restraint.

The phrase "simple and reliable" has been repeated for over two decades. While many companies have turned their culture into slogans on walls, Baidu continues to live by it.

Of course, young people have energy and sensitivity to new technologies, but they inevitably have shortcomings in industry and management experience, leading to occasional reversals in decision-making.

Over the past six months or so, Baidu's presence in consumer-facing large model products has fluctuated, which is, to some extent, a necessary period of adjustment during the transition.

Blood transfusion is not the goal in itself; whether a more efficient decision-making mechanism and more creative business strategies can emerge after the blood transfusion is the real test.

02
The Value of Full-Stack "Dull Efforts" Begins to Show

If organizational rejuvenation is Baidu's facade, full-stack technological capabilities are its substance.

Most people's impression of Baidu's technology still stay at (lingers at) search or even earlier, the Tieba era. But if you look at the timeline, Baidu may have been the earliest company in China to systematically invest in AI, without parallel.

In 2016, Li Yanhong publicly stated that Baidu would thoroughly transform into an AI company and subsequently submitted AI-related proposals at the Two Sessions for multiple years in a row. At that time, mobile internet was still the absolute mainstream, and the short video track was just emerging. The vast majority of companies were focused on consumer-side traffic, while AI remained a distant frontier concept.

No one could have imagined that seven years later, there would be a large model revolution, and Baidu dove right in.

Over those seven years, from the underlying Kunlunxin computing power to the PaddlePaddle deep learning framework, then to the Wenxin large model, and upward through Baidu Smart Cloud for industrial implementation, extending to autonomous driving and AI-native applications, Baidu has built almost the entire technological chain itself.

This model requires heavy investment, has long cycles, and delivers slow results, making it seem out of place in the internet industry, which emphasizes rapid iteration.

Many felt it was unnecessary, believing that using mature computing power and frameworks for application-layer innovation would be faster and yield quicker results.

But as 2026 arrived, this "dull effort" began to slowly pay off.

The Q1 financial report contained a crucial milestone: for the first time, revenue from Baidu's core AI new businesses accounted for over 50%. Among this, revenue from Baidu Smart Cloud infrastructure grew nearly 80% year-on-year, while GPU cloud revenue surged 184% year-on-year.

Behind these numbers are real industrial orders.

According to first-half statistics released by AI cloud industry agency Intelligent Hyperparameters, the disclosed winning bid amounts for large model-related projects from China's five major AI cloud vendors exceeded 2 billion yuan, with Baidu Smart Cloud alone securing over 60%, leading the industry in winning bid amounts for multiple consecutive periods.

The structure of these orders reveals some insights.

There are both local multi-billion-yuan AI infrastructure projects and, more commonly, vertical projects scattered across industries like finance, energy, and manufacturing. These cases reveal the true logic of AI implementation: the biggest cost is not going live but ongoing operations.

Many vendors can create demonstration-level solutions, but once deployed in production environments, issues such as high operation and maintenance costs, slow iteration speeds, and rapid performance degradation emerge.

In contrast, Baidu secures continuous repeat orders not through one-time project deliveries but through its ability to accompany clients in refining scenarios over the long term.

Shen Dao mentioned in an interview a practice within Baidu: each employee is given a monthly allowance of 1,000 yuan to freely use any mainstream large model on the market, with no performance requirements—just to let everyone experience the value of AI firsthand.

"Forcing AI adoption rarely yields good results; we focus more on positive guidance and incentives."

This detail speaks volumes about Baidu's technological culture.

It trusts the judgment of its technical staff, allowing frontline personnel to first understand, use, and create with AI before naturally integrating it into business operations.

To be precise, this is a continuation of an engineer-driven culture, a common trait among technology-driven companies like Google and Microsoft.

03
No Perfect Transformation Script Exists in the Face of Path Dependency

When discussing Baidu, an old topic cannot be avoided: starting early but arriving late.

This is almost the most firmly attached label to Baidu.

Tieba emerged early but later declined; food delivery started early but was eventually sold off; AI began even earlier, but the buzz around consumer-facing large models was surpassed by startups.

Many thus conclude: Baidu has strong technology but seems less adept at product development.

Li Yanhong directly addressed this question (doubt) at the 2025 Baidu World Conference.

He did not refute it, saying he had repeatedly pondered the question himself, and then explained that Baidu does not follow the fast-track path of seizing trends but adheres to a slow-variable route guided by technological development laws, ultimately pursuing the emergence of effects rather than models.

This answer can certainly be interpreted as a public relations line, but when viewed over a longer time horizon, things may not be so black and white.

All major enterprises that have stood atop the pinnacle face the problem of path dependency.

Microsoft, clinging to the cash cows of Windows and Office, watched helplessly as Google and Apple rose during the internet and mobile eras; Google, Lying down to earn money (lying down to make money) through search advertising, repeatedly faltered in areas like social and cloud services; Baidu, relying on the search advertising profit model, similarly missed the super-app window of mobile internet.

This is not a capability issue of any individual but an organizational structural dilemma.

When a business contributes billions in annual profits, any innovation that might impact it will be marginalized during resource allocation.

It's not that trends are unseen; rather, the incremental gains from trends fall far short of the potential losses to existing businesses in the short term.

Rational calculations ultimately lead to conservative choices.

Microsoft's transformation provides a classic reference. After Satya Nadella took over in 2014, he proposed "mobile-first, cloud-first," with the most symbolic move being the launch of Office for iPad.

Today, this seems like a natural step, but at the time, it meant Microsoft publicly acknowledged that Office's value exceeded that of the Windows platform, equivalent to shaking the company's decades-old value foundation with its own hands.

In the initial years of transformation, Microsoft's performance did not immediately take off; it even endured the pain of layoffs and asset write-downs.

The contraction of the Windows business and the dissolution of old licensing models dragged on short-term financial performance. It was not until four or five years later, when the growth curves of Azure and Office 365 truly took off, that the market woke up and revalued Microsoft.

Google's story provides a negative reference.

As the birthplace of AI technology, Transformer originated from Google, and most authors of the Transformer paper came from Google Brain.

Yet Google, holding a strong hand, reacted slower than startups in the large model era.

DeepMind and Google Brain operated independently for a long time, duplicating efforts until ChatGPT's emergence forced a hasty merger. Internal silos, disputes over computing power allocation, fragmented product lines, and the exodus of top talent in droves followed.

This is the AI dilemma of large enterprises: they lack neither technological reserves, funds, nor talent; what they lack is the organizational capability to rally forces and the resolve for self-revolution.

Returning to Baidu's context, it faces the same contradictions as all giants: search advertising remains the most stable cash flow, while AI new businesses are still in the investment phase. To invest requires sacrificing short-term profits; to prioritize profits requires slowing investment, and the ROI balance tests not just judgment but also resolve.

Baidu has chosen to go deeper into technology and industry.

Its consumer-facing products indeed generate less buzz than some startups, but on the industrial side, the order growth of Smart Cloud is real; in underlying technology, the complete chain from chips to frameworks is something most startups lack.

This choice comes with its costs. Full-stack investment is heavy, has long cycles, and delivers slow results, leaving little patience from capital markets and even less from public opinion.

People are accustomed to measuring an AI company's success by consumer-side monthly active users (MAUs) and daily active users (DAUs), yet much of Baidu's progress happens out of sight. These efforts may not be sexy or easily spread (spread), but they form the bedrock of AI's true implementation.

In a sense, what Baidu is doing today resembles what Microsoft did with the cloud back then. Initially underestimated, with a long investment phase and constant external doubts, but once it crosses the tipping point, the barriers formed by full-stack capabilities become highly resilient.

Microsoft took five or six years to complete its transformation; Baidu's AI transformation, counting from 2016, has also spanned a decade, with AI revenue only surpassing 50% for the first time in 2026.

Is this speed fast? Not by internet standards. But in the scale of technological revolutions, a decade is actually quite short.

— END —
For more, click below to follow