In terms of commercialization for (AI) agents, the global industry is still in its early stages, with business models continuously evolving. Currently, Token-based billing is more common, which essentially charges users for the capabilities of the foundational model. Over time, as AI agents approach human-level working capabilities, we believe future business models will become more diversified. Users will pay for the AI agents or AI applications themselves, and the scale of this market will far exceed that of Token-based billing.
On the evening of May 18th, during the Q1 2026 earnings call, Baidu founder Robin Li stated that AI applications have always been a strategic focus for Baidu. Within this product matrix, which includes "Major Search," digital humans, the "Miaoda" programming product, and the "Famou" intelligent agent, the search business leans more towards an advertising-based business model, while the latter three products are more inclined towards subscription-based, usage-based, or Token-billing models.
Released before U.S. market hours that day, Baidu's Q1 report showed total quarterly revenue of RMB 321 billion, a decrease of approximately 1.16% year-over-year and 2% quarter-over-quarter, yet surpassing market expectations of RMB 314.9 billion. Baidu's core business revenue was RMB 260 billion, up 2% year-over-year and flat quarter-over-quarter. Within this, AI business revenue reached RMB 136 billion, accounting for about 52% and showing growth for multiple consecutive quarters. iQiyi revenue was RMB 62 billion.
Regarding profits, Q1 net income attributable to Baidu was RMB 3.4 billion, a decrease of 56.2% year-over-year. Non-GAAP net income attributable to Baidu was RMB 4.3 billion, down 33.5% year-over-year, resulting in a non-GAAP net profit margin of 14% for Baidu.
During the earnings call, analysts inquired about Baidu executives' views on the competitive landscape and growth prospects of the domestic AI chip market in China.
Shen Dou, President of Baidu's Intelligent Cloud Group, believes that the market for China's locally self-developed AI chips is still in its early stages but is developing rapidly. With the continuous emergence of agent applications, vertical applications across various industries, and new forms of applications, inference is becoming an increasingly important component of computing demand. "On the supply side, domestic AI chips still face certain challenges in the short term, mainly reflected in production capacity and supply chain maturity, which is partly due to demand growth outpacing supply capabilities. However, in the long run, we believe the capabilities of domestically self-developed chips will continue to improve."
Baidu also disclosed progress in computing hardware in its Q1 report. The Kunlunxin P800 has completed large-scale validation and has delivered multiple ten-thousand-card clusters, meeting the requirements for large-scale training of cutting-edge large models in terms of computational precision, operator stability, framework adaptation, and long-term operation. Currently, the fully domestic Kunlunxin cluster has completed the training of the significant version 5.1 of the ERNIE model.
Furthermore, the Tianchi 256-card super node based on Kunlunxin has been successfully powered on and is scheduled for launch in June. It offers a 25% improvement in throughput performance compared to the previous generation and has completed adaptation for mainstream models including ERNIE, DeepSeek, GLM, and MiniMax, achieving a 50% increase in inference efficiency.
On May 8th, according to the China Securities Regulatory Commission website, Kunlunxin (Beijing) Technology Co., Ltd., a Baidu subsidiary, plans an initial public offering on the Science and Technology Innovation Board (STAR Market), with China International Capital Corporation Limited acting as the advising institution. On the same day, sources familiar with the matter revealed that Kunlunxin is still actively progressing with its Hong Kong listing process.
Regarding competition, Shen Dou stated that competition in AI chips depends not just on "whether chips can be delivered," but more importantly on whether these chips can operate stably and efficiently in diverse real-world scenarios. While domestic chips still have a certain gap compared to the world's most advanced products in some cutting-edge training scenarios, they already possess strong competitiveness in inference scenarios. "Currently, we have seen sustained growth in customer demand for Kunlunxin and an expanding industry coverage. We believe Kunlunxin holds a favorable competitive position for future industry development opportunities."
In terms of cloud business performance, the financial report shows that in Q1, intelligent cloud infrastructure revenue was RMB 8.8 billion, a 79% year-over-year increase, with GPU cloud revenue growing 184% year-over-year. Shen Dou noted that GPU cloud offers higher profit margins than traditional CPU cloud. Firstly, GPU cloud is technically more complex with higher barriers to entry. Secondly, self-developed chips and full-stack AI capabilities provide room for further cost optimization, and the continuous optimization of the customer structure also helps improve profit margins.
Regarding Robotaxi performance, in Q1, Apollo Go provided 3.2 million fully driverless ride-hailing services, with total order volume growing over 120% year-over-year. In March, the peak weekly order volume exceeded 350,000. As of April, Apollo Go has cumulatively provided over 22 million global ride-hailing services. As of May, Apollo Go's footprint has covered 27 cities globally.
Robin Li said that the current business focus for Apollo Go remains on continuously scaling up. "Apollo Go has already achieved unit economic breakeven in China's largest city of operation. Although our pricing remains relatively low, as we gradually expand globally, the overall pricing environment will become more favorable. Our overseas business will possess stronger profit potential after achieving scale."
The financial report shows that in the Middle East, Apollo Go's fully driverless ride-hailing service is already operational in multiple areas of Dubai, with the ApolloGo APP launched in March. In Europe, Apollo Go will conduct public road testing in Switzerland and will also begin testing in London, UK, alongside Uber and Lyft. Currently, Apollo Go's total autonomous driving mileage has exceeded 330 million kilometers, of which fully driverless mileage accounts for 220 million kilometers.
According to DZH VIP, as of the close of U.S. trading on May 18th, Baidu's stock rose 1.76% to $137.71 per share.
Comments