Baidu: GPU Cloud Revenue Nearly Quadruples as AI Business Hits Half of Revenue, But Robotaxi Hits Domestic Regulatory Snag and Core Ad Business Keeps Shrinking
Q2 2026 earnings call, August 18, 2026
Baidu's second-quarter results laid bare the central tension in its multi-year AI pivot: the infrastructure and application businesses it has built around Kunlunxin chips, cloud compute and generative AI tools are scaling at a blistering pace, even as the legacy search-and-advertising engine that still funds much of the company continues to erode. Total revenue came in at RMB 31.3 billion, down 4% year-over-year and 2% sequentially, while net income attributable to Baidu fell sharply to RMB 2.3 billion from a year-ago period that had benefited from a much larger RMB 4.9 billion of other income. Non-GAAP operating margin held at 12%, but the swing in other income — driven by lower fair value gains on long-term investments and foreign exchange losses — flatters the underlying operating trend less than it might appear.
GPU Cloud growth accelerates to 283%, AI now half the business
The standout data point of the quarter was AI Cloud Infra revenue growth of 50% year-over-year, which management said continues to outpace the broader industry. Within that segment, GPU Cloud revenue grew 283% year-over-year, up from an already elevated 184% in the first quarter — the fourth consecutive quarter of triple-digit growth. CFO Haijian He noted that Baidu Core AI-powered Business generated RMB 12.5 billion in the quarter and now represents half of Baidu General Business revenue, "reinforcing AI-powered business as a key driver of Baidu's long-term growth." AI Cloud Group head Dou Shen attributed the acceleration to three forces: computing supply remaining constrained industry-wide even as demand for both training and inference keeps rising, existing large customers in gaming, e-commerce and lifestyle content increasing spend, and rapid broadening of the customer base across verticals — embodied AI revenue alone grew roughly sixfold year-over-year. Shen said the mix shift toward GPU Cloud, which carries a better margin profile than traditional CPU cloud, is "contributing to a healthier revenue mix and strengthening the long-term profitability of our cloud business," and that management expects AI Cloud Infra growth to stay strong in the second half "with the potential for further acceleration." Separately, revenue from external customers' token usage on the Qianfan MaaS platform grew more than ninefold year-over-year, though management conceded MaaS still represents "a relatively small share" of Cloud Infra revenue today.
Robotaxi hit a regulatory speed bump domestically
Apollo Go delivered around 1 million fully driverless rides in the quarter, taking cumulative rides past 23 million, but Robin Li's team disclosed that ride volume was "temporarily affected by operational adjustments in certain domestic cities due to regulatory considerations." The company said it used the period to conduct "a systematic review to further strengthen the robustness of our autonomous driving systems and the rigor of our operational processes," with operations in the affected cities only beginning to resume in August. This is a rare admission of a domestic setback for a business Baidu has positioned as a flagship proof point of its AI strategy, and it signals that China's tightening regulatory posture toward autonomous vehicles — including a new national safety standard for Level 3 and Level 4 systems that Apollo Go contributed technical input to — carries near-term execution risk alongside its long-term legitimizing effect. Internationally, the picture was more constructive. Apollo Go began fully driverless commercial operations in Dubai in July, claiming the largest scale among fully driverless robotaxi services in that city, and started open-road testing in London with Uber and Lyft as partners following a milestone driverless testing permit in Hong Kong — the first globally in a right-hand-drive, left-hand-traffic market. Li argued the international opportunity outside the U.S. and China is larger than the domestic market itself, and that "our low-cost vehicles and proven operating model have the potential to deliver even stronger unit economics" in overseas markets with higher fares than China's.
Management concedes ERNIE fell behind, points to new talent and application-first strategy
Asked directly by JPMorgan's Alex Yao how ERNIE stacks up against a wave of new multitrillion-parameter frontier models, Li offered an unusually candid response for a CEO discussing his own foundation model. He said Baidu had brought in "top AI talent" after reorganizing model teams into separate foundation-model and application groups, and stated plainly that the company is "confident in accelerating AI's iteration and bring ERNIE back into the top tier of foundation models" — phrasing that implicitly acknowledges ERNIE has slipped from that tier. Rather than chasing benchmark leadership across the board, Li said Baidu will pursue an "application-driven approach," concentrating ERNIE's development on the capabilities that matter most for AI search, digital humans, the Miaoda coding platform, Famou Agent and the DuMate general-purpose agent, using product feedback loops to direct model training.
Advertising business explicitly guided lower into second half
Julius Rong Luo, who runs Baidu's Mobile Ecosystem Group, gave one of the more direct near-term negatives on the call: online marketing revenue will "remain under pressure in the second half," driven by the company's deliberate choice to hold back monetization of AI search while it prioritizes product quality and user experience. Luo noted that ERNIE Assistant daily active users grew 83% year-over-year in June with conversation rounds more than tripling, and that its Task Agent topped two third-party agent benchmarks — but he was equally clear that "competition for users' time and attention has intensified further" as AI chatbots proliferate, and that monetization will only follow once product-market fit is more fully established.
Hong Kong dual primary listing on track for this year
Responding to a question from UBS's Wei Xiong, He confirmed Baidu's board approved conversion to a dual primary listing in Hong Kong in July, the company has filed its application and received the Hong Kong Stock Exchange's acknowledgment, and it will hold an extraordinary general meeting on August 26 to secure shareholder approval, with conversion expected to become effective within the year. He said Baidu is also "actively preparing" for potential Southbound Stock Connect inclusion following conversion, which management expects would "meaningfully expand participation from Mainland China investors" and diversify the shareholder base — though he cautioned this remains subject to exchange eligibility rules and review.
Kunlunxin chip listing still pending, roadmap extends to M300
On the long-anticipated spin-off listing of Kunlunxin, Dou Shen offered no new timeline, telling Goldman Sachs' Lincoln Kong only that "the listing process for Kunlunxin is still ongoing, and we will update the market as soon as we have more to share." On the business itself, management said the chip line has broadened compatibility to newer Chinese foundation models including Kimi K3, GLM5.2, Minimax M3 and Hunyuan 3, and is advancing a roadmap that includes the M100 for large-scale inference and an upcoming M300 series — continuity that underscores Kunlunxin's role as a domestic alternative amid persistent supply constraints on foreign AI chips in China.
Applications gaining traction but still early
Beyond the headline infrastructure numbers, Baidu pointed to early commercial signs across its application suite: one large Chinese internet client expanded digital-human livestreaming deployment to 2.5 times prior levels after a single quarter of use, Miaoda's monthly active users rose 67% from March to June following the launch of a mobile app-generation feature, and Famou Agent began to see "early commercial traction" after its 2.0 launch. These remain modest relative to Baidu's roughly RMB 125 billion in trailing annual revenue, but they represent the diversification layer management is counting on to eventually offset the structural decline in traditional search advertising.