Search is slowing, AI is burning cash: Baidu’s “transformation” with no way back.

China’s 2026 Spring Festival marketing has clearly shifted to AI apps. Baidu’s ERNIE rolled out RMB 500M in cash red packets and appeared on a major TV gala; Tencent’s Yuanbao launched a RMB 1B red-packet campaign; ByteDance’s Volcano Engine and Doubao partnered with CCTV’s Spring Festival Gala—turning the annual “red-packet war” into a prime window for mass AI user acquisition.

Behind the headline campaigns, Baidu made two major moves starting in 2026: its AI chip unit Kunlunxin filed a confidential IPO application in Hong Kong, and Baidu merged its document library and cloud drive into a new consumer-focused “Personal Super Intelligence Group (PSIG),” reporting directly to Robin Li. The first move points to a more independent, capitalized compute strategy; the second signals a tighter focus on subscription-driven consumer AI products.

These changes come as Baidu’s legacy engine weakens. In 2025 Q3, Baidu booked RMB 16.2B in long-term asset impairments, describing it as retiring infrastructure that can’t meet AI-era compute demands. At the same time, Baidu highlighted AI momentum: AI Cloud revenue rose 33% YoY, AI app revenue hit RMB 2.6B, and “AI-native marketing” surged 262% YoY. The transition is becoming explicit: search ads are slowing, while models, apps, and compute move to center stage.

Search remains the core dilemma. Q3 online marketing revenue fell 18% YoY to RMB 15.3B as user attention shifts to platforms like Douyin, WeChat, and Xiaohongshu. Baidu’s “AI-rebuilt search” improves user experience by answering directly with rich media—yet it undermines the click-driven ad model that historically funded search, creating a fundamental monetization paradox.

Meanwhile, Baidu’s two biggest cash burners face pressure. In large models, open-source and low-cost entrants (e.g., DeepSeek) forced Baidu to open-source and cut prices while competing against fast-growing rivals like Doubao. In autonomous driving, Apollo Go shows progress but scaling remains capital-intensive, especially as ad-driven cash flow softens and competition tightens.

To sustain the pivot, Baidu is centralizing control, reshuffling teams, and doubling down on its full-stack AI strategy—chips (Kunlunxin), framework, models (ERNIE), and cloud applications—to build end-to-end efficiency and a defensible cost structure in enterprise and government markets. In short, Baidu is being pushed from a “search company” into a “full-stack AI company,” in a transformation with little room to retreat.