Liu Sixuan, intern journalist at the Economic Observer

On August 20th, Alibaba disclosed its latest quarterly financial report, and thus all of BAT's interim results have been released. The three leading Chinese Internet companies have shown different approaches and attitudes towards cloud business: Alibaba values the profit prospects of cloud, and is the only one of the three companies to separate cloud profits and show them on their books; Baidu is still betting on the cloud when its net profit plummets by nearly 70%. While using Apollo Go to grow its computing power in its own scene, Baidu is exporting through Kunlun Core and industry programs; Tencent continues its tradition of not separately listing cloud revenue, integrating cloud into the ecosystem of gaming, advertising, and WeChat.

The change in demand structure has made selling computing power one of the most certain directions for manufacturers. According to the "China AI Cloud Market Share 2025" report released by international market research firm Omdia in May, the proportion of inference tasks in AI computing will increase from 55% in 2025 to over 80% in 2027, and the unit of measurement for computing power will shift from "GPU hours" to Token consumption.

Demand is soaring, supply remains tight, and computing power has shifted from self use costs to sellable goods. Cloud providers are trying to tell a new story in their financial reports.

Tencent: Integrating computing power into the ecosystem, selling alone is not the focus

Tencent's Q2 revenue was 20.48 billion yuan, a year-on-year increase of 11%, mainly driven by two cash cows: local market gaming revenue of 47.3 billion yuan, a year-on-year increase of 17%; The marketing service revenue was 43.6 billion yuan, a year-on-year increase of 22%. Relying on these two businesses, Tencent has the confidence to invest heavily in computing power - with a capital expenditure of 52.8 billion yuan in the second quarter, a year-on-year increase of 176%, nearly three times that of the same period last year; As a result, the free cash flow turned negative by 13.8 billion yuan, and if the advance payment for computing power procurement is excluded, it is a positive value of 37.6 billion yuan.

In the financial report, Tencent Cloud's business was included in the Financial Technology and Enterprise Services division display: the revenue of this division in the second quarter was 60.3 billion yuan, a year-on-year increase of 9%. The main body is payment and wealth management, and the specific figures of cloud business have not been clearly disclosed. Only the qualitative statement made by the management during the performance conference can reveal the growth rate of cloud revenue, which has increased from a high double-digit year-on-year growth rate in the first quarter to a low twenty digit growth rate in the second quarter, and is in the lowest range among the three companies.

This growth rate is closely related to Tencent's positioning in the cloud. The financial report mentioned that the growth of enterprise service revenue is driven by the growth of cloud service revenue and benefited from the increasing demand for AI related services; Ma Huateng further clarified in the performance announcement that a significant increase in computing power procurement will help us convert the use of applications and models into revenue. In other words, Tencent's logic is to first rely on AI efficiency improvement in games and advertising to earn profits, and then convert demand into cloud revenue through the use of applications and models. The call volume of the hybrid big model Hy3 ranks among the top in the world on platforms such as OpenRouter.

Cloud to Tencent is a natural result of application and model calls, and selling it separately is not the focus. During the conference call, regarding the allocation logic of capital expenditures between different businesses such as Tencent Cloud, James Mitchell, Chief Strategy Officer (CSO) of Tencent, responded that model training will be given top priority in the coming months, followed by providing inference computing power to "integrate with WorkBuddy's third-party model operation"; He also stated that among the existing channels for monetizing computing power, "relying on the token billing business generated by WorkBuddy" is creating the most long-term and stable commercial value for Tencent, and computing power resources are therefore tilted.

Baidu: New engine cannot drive old engine

Baidu's quarterly report showed multiple indicators declining. The total revenue for the second quarter was 31.3 billion yuan, a year-on-year decrease of 4%; The net profit attributable to Baidu was 2.3 billion yuan, a year-on-year decrease of 68%; Online marketing services continue to be weak, with revenue of 13.1 billion yuan, a year-on-year decrease of 19%.

At the end of February, when disclosing Baidu's 2025 annual performance, Robin Lee said that in the foreseeable future, core AI new business will become the main part of Baidu's overall business. Now it seems that the new engine is running steadily: the revenue of intelligent cloud infrastructure is 7.3 billion yuan, a year-on-year increase of 50%, of which GPU cloud revenue increased by 283% year-on-year, maintaining triple digit growth for four consecutive quarters; The core AI new business revenue was 12.5 billion yuan, a year-on-year increase of 25%, accounting for half of Baidu's general business revenue for two consecutive quarters. Robin Lee said on the conference call that the growth momentum of new core AI business "confirms Baidu's transformation from an Internet centric company to an AI led company".

However, the core AI new business is still in the stage of high investment and low gross profit, and the conversion of revenue growth to profit is weak; The combined pressure of online marketing has resulted in a significant decrease in overall net profit.

From the perspective of application scenarios, Baidu's computing power is mainly invested in two directions: self owned scenarios and external empowerment. On its own scene end, Apollo Go has covered 28 cities and accumulated more than 350 million kilometers of automated driving. Its data closed-loop, model training and cloud collaboration are all running on Baidu AI Cloud, and both provide mutual technical evolution support; On the external enabling end, according to two market share reports released by IDC, an international data company, in July and August, Baidu Intelligent Cloud ranks first in China's embodied intelligent AI cloud market and first in China's financial industry's generative AI market.

As the foundation of Baidu Cloud's AI computing power, Kunlun Chip is also a key focus for investors. Robin Lee said on the conference call that Kunlun Chip has completed the research, development and commercialization of three generation AI chips, and further covered the new version of the mainstream model in the second quarter. In addition, Baidu executives responded to the progress of Kunlun Chip's listing during the conference call, stating that the process is being pushed forward and will be synchronized with the market as soon as there is more information to disclose. At the end of June, foreign media reported that Kunlun Chip prioritized institutions willing to purchase its chips when determining IPO investors, and some participating institutions were required to purchase Kunlun Chip chips worth 3 to 7 times their subscription amount.

Alibaba: Cloud makes money, Qian Wen burns money

In the first quarter of Alibaba's fiscal year 2027 (corresponding to the second quarter of 2026), the revenue was 268.953 billion yuan, a year-on-year increase of 9%, making it the largest among the three groups in terms of revenue scale.

Alibaba Cloud's external commercialization revenue increased by 45% year-on-year, reaching a new high in 22 quarters; AI related product revenue accounts for 35% of external cloud revenue, maintaining double-digit growth for the 12th consecutive quarter.

One change this quarter is Alibaba's restructuring of its divisions, merging Cloud Intelligence Group and self-developed chip company Pingtou Ge into "AI Cloud and Computing Power Services", with a single quarter revenue of 48.437 billion yuan, a year-on-year increase of 45%; After adjustment, EBITA (earnings before interest, tax, depreciation, and amortization) increased significantly by 133% year-on-year, and the profit margin rose to 12%. Through this restructuring, Alibaba has for the first time separated the profits from selling computing power from its cash burning model business and presented them separately.

On the other end of the financial report, some cloud based businesses did not deliver impressive numbers. The newly established "AI Laboratory and Applications" division, covering Qianwen App, Qianwen Office, and Model Laboratory, incurred a quarterly adjusted EBITA loss of 13.861 billion yuan, approximately 2.5 times the profit of the cloud division. The group's operating profit decreased by 57% year-on-year. That is to say, the money earned by the cloud is not enough to fill the gap burned out by the model.

However, some losses did not slow down the implementation of investments. This quarter, the latest generation AI processor, the Zhenwu M890, was commercialized through Alibaba Cloud among over 650 external customers; The delivery cycle of large-scale AI data centers has been compressed to 100 days; The Qwen3.8-Max with 2.4 trillion parameters has been announced as open source, marking the first time that Alibaba Qianwen has made its Max level flagship model publicly available for weighting.

Their Respective Battlefields

Intensifying investment in computing power is a common move among the three companies in the past fiscal quarter.

Tencent's capital expenditure in the second quarter was 52.8 billion yuan, a year-on-year increase of 176%. Tencent President Liu Chiping explained during the performance conference call that it was a large-scale investment in computing power; Alibaba's revenue reached 67.678 billion yuan, a year-on-year increase of 75%. Wu Yongming stated that "the certainty of investment return on AI computing power is very high"; Baidu's revenue reached 11.39 billion yuan, a year-on-year increase of about 200%. Baidu's Chief Financial Officer (CFO), He Haijian, stated that the company is still in the AI investment cycle and will firmly invest in its performance. However, when compared horizontally, there are slight differences in the strategic weights and expected returns given by various companies in the cloud business.

The most direct difference among the three accounting books is how clouds are written into financial reports. Alibaba listed cloud and computing power separately and disclosed profits, Baidu listed cloud as an intelligent cloud infrastructure, and Tencent continued its tradition of integrating cloud into its fintech and enterprise services division. The tightness of disclosure criteria corresponds precisely to the weight of cloud in their respective strategies: Alibaba regards cloud as an independently calculable second curve, Baidu regards cloud as a new growth point after advertising stall, and Tencent makes cloud serve its own ecosystem more.

What will support this heavy investment? Tencent has two cash cows, gaming and advertising, to support it, and the computing power expenditure is still within an affordable range; Alibaba Cloud has been able to make profits on its own, but the profits earned immediately filled in Qianwen's losses; Baidu is short of cash cow to rely on, and its net profit has dropped significantly. The investment of computing power is recycled through the sale of Kunlun core, while it is undertaken by its own scenes such as Apollo Go.

Cloud is a heavy asset business, and whether the investment can be recovered is directly related to the long-term financial performance in the future.

There is a slight difference in the clarity of the three responses regarding this point:

Alibaba's Wu Yongming stated during a conference call that assuming the current average gross profit level of AI products can be maintained, coupled with the improvement in gross profit margin brought about by the increase in the proportion of self-developed chips, "CAPEX (capital expenditure) can be recouped within three years", and it is expected to be shortened to 2.5 or even 2 years in the future.

In response to market concerns about negative free cash flow, Tencent's Liu Chiping explained that the capital expenditure on AI native businesses is mainly a "one-time AI centralized investment" this year and next year, and computing power expenditure will not remain high in the long run.

Baidu's He Haijian stated that the return cycle of different AI investment projects varies, and some projects require longer time to realize. Baidu has a good prediction of investment returns and is confident in continuously improving investment efficiency.

(Author Liu Sixuan)