Baidus Journey to Full-Stack Dominance: The Long Game of AI Bets

百度长线战略：从早期的人工智能投资到全栈式主导，百度通过不断投入研发，成功将人工智能技术应用于多个领域，实现了从搜索引擎到智能助手的跨越，百度已占据行业领先地位，展现出其在人工智能领域的全面优势。

At the 2025 Baidu World Conference, the spotlight fell once again on the early movers of China’s AI era.

Robin Li, Baidu’s founder, used the occasion to underscore a shift that has been quietly reshaping the company’s trajectory: artificial intelligence has evolved far beyond chatbots and automated responses. Digital humans, intelligent code agents, and self-evolving systems capable of discovering global optimum solutions across general scenarios are no longer theoretical exercises—they are increasingly part of Baidu’s operational reality.

“When AI capabilities become an innate part of our workflow, intelligence ceases to be a cost and instead becomes a form of productivity,” Li said. “What we care most about is how AI can be organically integrated into everything we do, so that AI becomes a native driving force for both enterprise development and personal growth.”

For much of the past decade, the buzzwords dominating China’s tech landscape were “traffic” and “monetization.” Li argues that the next ten years will be defined instead by “intelligence” and “long-termism.” In other words, the true winner of a technology cycle is determined less by short-term profits than by foundational investments and patient accumulation.

The Path of AI Long-Termism

Baidu’s journey in AI exemplifies the “first-mover, long-horizon” approach. Its earliest investments date back to 2013, years before AI became a household concept. That year, Baidu established its Institute of Deep Learning (IDL), signaling a commitment to bring AI out of academia and into the commercial sphere. While the broader internet ecosystem focused on mobile traffic and short-term growth, Baidu quietly built infrastructure and expertise with no immediate commercial payoff.

In 2015, Baidu launched its autonomous driving initiative, the Apollo Program, becoming the first Chinese company and the fourth globally to invest heavily in Level 4 autonomous driving technology. Simultaneously, the company introduced Baidu Brain, an ambitious platform integrating foundational AI capabilities such as speech recognition, computer vision, and knowledge graphs. At the time, Baidu Brain was dismissed by some as an academic curiosity with limited commercial relevance.

Baidu’s ambitions extended to hardware as well. In 2018, it unveiled its in-house AI chip, Kunlun, against a backdrop of global dominance by NVIDIA and a weak domestic chip ecosystem. In 2019, the company rolled out Wenxin 1.0, its first large language model, just before the world took notice of ChatGPT. Yet early on, the broader market remained skeptical of Baidu’s “tech-first” strategy, questioning the commercial viability of chips, models, and computing infrastructure while competitors chased immediate revenue from mobile traffic and app monetization.

For years, Baidu was the subject of a familiar refrain: “early to rise, late to arrive at the market.” The company’s focus on long-term technological bets seemed at odds with the speed of business. Yet, it was precisely this steadfastness that positioned Baidu for the explosive growth of the AI era. By 2025, strategies once considered idealistic have become mainstream expectations for tech giants worldwide.

Baidu’s technological architecture is now fully integrated, combining chips, computing frameworks, models, and applications into a self-consistent ecosystem. Its ERNIE large model sits at the core, powered by Kunlun chips and supported by the Baige AI computing platform, which offers stable, scalable services to enterprise partners. Qianfan, Baidu’s AI agent development platform, has cultivated a developer ecosystem that extends Baidu’s capabilities into real-world enterprise applications.

On the consumer side, Baidu has launched AI-powered transformations across its core products. Baidu Search now produces AI-generated content across the majority of search results, with first-result rich media coverage exceeding 70%. Baidu Smart Cloud reported a 27% year-on-year revenue increase, ranking among the top providers in China’s AI public cloud market. Apollo Go autonomous mobility services now operate in 22 cities globally, highlighting Baidu’s leadership in both enterprise and consumer-facing AI solutions.

“The company’s early-mover advantage is validated not by hype but by results,” Li said. “This is the power of patience, the payoff of investing in fundamentals rather than chasing trends.”

Autonomous Driving and AI Chips: Strategic Trump Cards

Two pillars underscore Baidu’s unique competitive moat: autonomous driving and AI chip development.

Robobus, Baidu’s fully autonomous ride-hailing fleet, has evolved from laboratory prototypes to operational scale. The system now manages more than 250,000 weekly rides, covering over 240 million kilometers, with fully driverless operations accounting for more than 140 million kilometers. Globally, Robobus has provided more than 17 million mobility services, surpassing Waymo as the largest autonomous ride-hailing operation worldwide. Beyond serving as a technical showcase, Robobus demonstrates AI’s capacity to impact urban life, from transportation patterns to city planning.

Baidu’s chip strategy, embodied in Kunlun, has been similarly prescient. Initiated in 2011 with early FPGA-based accelerators and later expanded to GPUs, Baidu sought to control the foundation of AI computing. By 2025, Kunlun supports China’s first fully in-house 30,000-card AI cluster, capable of training multi-billion-parameter models. The cluster has earned a five-star certification from the China Academy of Information and Communications Technology (CAICT), marking a milestone in China’s self-reliant AI computing capability.

The company’s latest chip generation, Kunlun M100 and M300, targets large-scale inference and training of ultra-large, multimodal models. Complementing these chips are Tianchi supernodes, designed to train trillion-parameter models. Tens of thousands of Kunlun chips are already deployed, forming the backbone of Baidu Smart Cloud’s AI computing infrastructure, which has held the top market share in China for six consecutive years.

Baidu has also moved from “enabling” AI to “internalizing” it across enterprise operations. The company’s AI strategy is no longer limited to consumer-facing applications; it now permeates production, decision-making, supply chain optimization, financial risk control, and scientific research.

At the 2025 conference, Baidu unveiled Baidu Fumo, the world’s first commercially available “self-evolving” AI agent. Fumo can autonomously optimize complex systems—from port logistics to energy design—by iterating solutions in response to real-world dynamics, surpassing even top algorithm experts in some tasks. This represents a shift from AI as a tool to AI as a self-contained operational capability.

Meanwhile, Baidu’s AI-powered ecosystem has expanded dramatically. PaddlePaddle, the company’s deep learning platform, now counts over 23 million developers. The Qianfan Platform hosts 460,000 companies and 1.3 million AI agent applications, forming a rapidly growing industrial ecosystem. Large models have moved from value-added features to central productivity drivers, enabling Baidu to operationalize “AI to B” as a model for enterprise digital transformation.

Redefining Search and Human-AI Collaboration

Baidu Search has undergone a radical transformation. No longer a text-and-link platform, it now emphasizes rich media content, powered by AI. Baidu has opened its search API to 625 device manufacturers, including Samsung, Honor, and vivo, making AI search capabilities foundational across industries. Monthly active users of Baidu AI Search have reached 382 million, topping China’s market for three consecutive quarters.

Baidu Wenku and Baidu Netdisk have integrated GenFlow 3.0, the first universal intelligent agent system across platforms, featuring global memory, task collaboration, and autonomous invocation. Baidu Huiboxing has become a full-scenario digital human platform, with real-time interactive capabilities now deployed across multiple industries. These developments highlight a shift from consumer-facing AI to enterprise productivity, signaling that the long-term value of AI lies in industrial internalization rather than short-term user engagement.

For years, Baidu’s AI strategy was criticized as idealistic or detached from market realities. Today, the company exemplifies the payoff of long-term vision. Twelve years of incremental investment have culminated in a full-stack AI ecosystem, combining hardware, software, models, and industrial applications.

Baidu’s evolution demonstrates that in technology, timing and patience are as critical as ingenuity. By starting early and persisting through a decade-long period of skepticism, Baidu has established both technological leadership and market credibility. Its model illustrates how long-termism—eschewing short-term gratification in favor of sustained foundational investment—can produce compounding returns, particularly in AI.

As AI becomes the backbone of industrial upgrading, companies worldwide are racing to achieve “AI-native” status. Baidu, with its decade-long preparation, is uniquely positioned to serve as the foundational infrastructure provider for this next wave of digital transformation. The company’s AI strategy—from chips to models to applications—is no longer experimental; it is standard-setting.

Robin Li concluded the conference with a reflection that encapsulates Baidu’s journey: “Long-termism is not a slogan—it’s an organizational capability. It requires both a scientist’s mindset to withstand solitude and an engineer’s spirit to turn technology into practical application. That is how our AI system was forged.”

Baidu has not merely caught up with the AI era—it has awaited it. The company’s early vision, full-stack strategy, and steadfast belief in technology are finally being rewarded. In an age where computing power is the new oil and AI the engine of industrial innovation, Baidu has stepped fully into its role as a pioneer, proving that patience, persistence, and a focus on fundamentals pay off in the long run.