Big tech giants are betting heavily on AI, where does the capital flow?

In recent days of August, Alibaba, Tencent, Baidu, JD.com and Kuaishou have successively released their financial reports covering the period from April to June. The AI-related components in these reports have become the focus of the entire industry.

Before making horizontal comparisons, we need to unify the statistical calibers first. Alibaba's fiscal year ends on March 31 of each year, so the period from April to June 2026 corresponds to the first quarter of its 2027 fiscal year; the performance data for the same period disclosed by Tencent, Baidu, JD.com and Kuaishou all refer to the second quarter of 2026. Despite the different names of the financial reports, the five companies actually all disclosed operating data for the same three-month period ending on June 30, 2026, which can be compared under a unified time standard.

The five companies are all well-known listed internet enterprises, whose businesses cover social networking, e-commerce, cloud computing, games, search, short videos and supply chain, connecting hundreds of millions of users and a large number of enterprise customers, with fully complete financial disclosure. From April to June 2026 alone, the total revenue of the five companies reached about 887 billion yuan, which is sufficient to form a representative sample for observing the AI competition in China's internet industry.

The external environment is also changing rapidly. The 57th Statistical Report on the Development of China's Internet released by CNNIC in 2026 shows that by December 2025, the penetration rate of generative artificial intelligence in China reached 42.8%, continuing to rise from 36.5% half a year earlier.

Generative AI has moved from being a new tool for a small group of people to wider daily usage scenarios. While the user base continues to expand, the AI competition faced by internet companies is becoming increasingly concrete. The five financial reports just provide a window to peek into the respective development stages that leading tech companies have reached in the same quarter.

AI Competition Is Extremely Capital-Intensive

In the second quarter of 2026, the AI competition among Chinese internet companies saw a very intuitive change: massive funds are flowing into computing power.

Alibaba recorded a revenue of 2.6895 trillion yuan in the quarter ending June 30, a year-on-year increase of 9%, with capital expenditure reaching 676.8 billion yuan, up 75% year-on-year. Its capital expenditure has accounted for a quarter of its revenue in the same period. Alibaba explained that the expenditure growth reflects the company's continuous increase in investment in AI infrastructure to meet the expanding customer demand.

Tencent achieved a revenue of 2048 billion yuan in the same period, a year-on-year increase of 11%, with capital expenditure reaching 527.8 billion yuan, up 176% year-on-year, which is also close to a quarter of its quarterly revenue. Tencent also disclosed that its operating cash flow includes a large amount of AI-related prepayments for model upgrades, WorkBuddy, CodeBuddy, WeChat AI and infrastructure required for cloud services.

For Alibaba and Tencent alone, their combined capital expenditure in one quarter has exceeded 1.2 trillion yuan. Not all of the 1.2 trillion yuan can be attributed to AI, as the two companies still need to invest in cloud services, data centers and other technical facilities, but the financial reports have directly linked the rapid expansion of expenditure to the demand for AI computing power. By 2026, in addition to model capabilities, capital capacity has begun to determine how long an internet company can stay in the competition game.

Baidu provides another perspective. In the second quarter, Baidu's revenue was 313.3 billion yuan, down 4% year-on-year, and its capital expenditure reached 113.9 billion yuan, accounting for about 36% of its quarterly revenue, a proportion higher than that of Alibaba and Tencent. Its operating cash flow in the same period was only 34 billion yuan, and its free cash flow was negative 79.5 billion yuan. Baidu's online marketing revenue dropped by another 19%, its traditional cash cow is under pressure, and AI still requires continuous investment in talents and infrastructure.

For companies with smaller revenue scales, the same amount of computing power investment brings heavier pressure. This adds a new threshold to the AI arms race: the technical team needs to keep up with the pace, and the cash flow also needs to be sustainable.

Behind the simultaneous increased investment of leading tech companies, the demand for computing power is still growing at a high speed. IDC data shows that in 2025, the market size of China's intelligent computing cloud infrastructure reached 486.7 billion yuan, a year-on-year increase of 128%, and it is expected to exceed 1 trillion yuan in the next two years. Globally, IDC predicts that AI infrastructure expenditure will reach 4870 billion US dollars in 2026, up about 53% year-on-year. In the first quarter of 2026, global server market expenditure increased by 30.7% year-on-year, mainly driven by the continuous large-scale deployment of GPU servers.

Generative AI is also changing the cost structure of internet companies. In the past, a mature software product could serve hundreds of millions of users at low cost. Today, even a single question-and-answer session or a piece of video generation consumes computing resources, let alone code generation and agent execution. As the number of users surges, the number of invocations also increases, which further raises the demand for servers, power, networks and data centers. Internet companies are shifting from light-asset to heavy-asset models, and GPU and computing power have truly entered the core of competition.

The impact of capital expenditure will also continue into the future. After servers and data centers form long-term assets, depreciation and operating costs will be recorded in the income statement for multiple quarters. Alibaba and Tencent have e-commerce, games, advertising and cloud businesses that continuously generate cash flow, while Baidu needs to maintain AI investment when its traditional advertising business declines. Facing the same round of technological revolution, the costs borne by several companies are not the same.

At the current stage of AI competition, it is no longer sufficient to only refer to model rankings. Whether a company can invest tens of billions of yuan in computing power for several consecutive years and withstand the pressure on profits and free cash flow has become a new competitive condition. Technological leadership may change with one model iteration, while infrastructure, cash flow and capital bearing capacity require years of accumulation. Ordinary medium-sized enterprises that participate in competition at this scale and level are very likely to be eliminated.

AI Has Become a Revenue Item

In the second quarter of 2026, AI has moved from the "expenditure sheet" to the "revenue sheet". Among the five companies, Baidu, Alibaba and Kuaishou have sent the clearest signals: what kind of AI capabilities enterprises are willing to pay for can already be found in the financial reports.

The change of Baidu is the most representative. In the second quarter, Baidu's core AI-driven business revenue reached 125 billion yuan, a year-on-year increase of 25%, accounting for half of Baidu's core AI business revenue. Breaking down the data, the revenue of AI cloud infrastructure was 73 billion yuan, up 50% year-on-year; among which, GPU cloud revenue increased by 283% year-on-year, higher than the 184% growth rate of the previous quarter. The revenue of AI applications was 25 billion yuan, up 3% year-on-year.

In the same company and the same quarter, there is a large gap between the two sets of figures. The areas where enterprise customers are currently most willing to spend money are still concentrated in computing power, cloud infrastructure and model operation environments. Model training, Agent operation and enterprise system access to AI all rely on stable computing resources.

Alibaba's financial report gives a similar answer. In the second quarter, the revenue of AI cloud and computing power services reached 484.37 billion yuan, with both total revenue and revenue from external customers increasing by 45% year-on-year. Among them, the revenue of AI-related products reached 123.76 billion yuan, achieving three-digit year-on-year growth for the 12th consecutive quarter.

Alibaba also cited Omdia data that Alibaba Cloud held a 38.1% share of China's AI cloud market in 2025. The products sold by the cloud business are increasingly equipped with GPU computing power, model services, Agent operation environments and heterogeneous chip scheduling capabilities. AI has reinstalled a growth engine for cloud computing.

It is more noteworthy that the revenue of Alibaba's AI-related products has accounted for about a quarter of the revenue of its AI cloud and computing power services business. When the demand for computing power continues to grow, cloud platforms, self-developed models, chips and enterprise customers can be connected in one business chain. This adds a very practical evaluation criterion to the large model competition: how much are customers actually willing to pay?

Kuaishou has taken another path. The revenue of Keling AI in the second quarter exceeded 850 million yuan, up more than 200% year-on-year. Kuaishou's total revenue in the same period was 355 billion yuan, and Keling's revenue has accounted for more than 2% of the group's quarterly revenue.

Video generation is naturally connected with advertising, e-commerce, film and television, short dramas and content creation. Users invoking models usually correspond to clear production tasks. The cost of generating a video and the shooting and production expenses it can save are easy to calculate, and the payment link is also shorter.

Industry data can also confirm the change of commercialization direction. IDC data shows that in 2025, the market size of China's AI application public cloud services reached 137.3 billion yuan, higher than the 79.4 billion yuan of the large model training and inference public cloud market. The logic of enterprises purchasing AI is shifting from simply acquiring model capabilities to solving specific business problems. Models are still important, and continuous revenue is increasingly close to computing power supply, industry applications and production tools.

Putting Baidu, Alibaba and Kuaishou together, AI has currently formed two clear revenue paths: one is to sell computing power, cloud resources and model services to enterprises, and the other is to turn model capabilities into vertical production tools. By 2026, AI commercialization has become increasingly specific: in addition to how advanced the technology is, it also depends on who can truly convert a single model invocation, a unit of GPU computing power, and a generated video into revenue.

AI Has Integrated Into Core Business

After the funds are spent and revenue can be seen, AI competition still needs to answer a more difficult question: can the models integrate into existing businesses and change the operation modes of advertising, transactions, procurement, healthcare and logistics?

Among the five financial reports for the second quarter of 2026, Tencent and JD.com have sent clear signals. The economic value of AI is increasingly hidden in the original revenue items and business processes. If people only look for "AI revenue" separately, it is easy to underestimate its impact on an internet company.

Tencent is a typical case. In the second quarter, Tencent's marketing service revenue reached 435.65 billion yuan, up 22% year-on-year, nearly twice the group's 11% revenue growth rate. The company directly attributed the growth to the AI-driven upgrade of the advertising recommendation model, the upgrade of the AIM+ automatic delivery tool, and the enhanced closed-loop marketing capability of the WeChat ecosystem.

The application of AI in the advertising business is all about specific trivial matters: accurately calculating what content to push to users, helping advertisers select the most valuable traffic, and then directly connecting small stores and mini-games to the delivery link. No matter how advanced the technology is, the underlying logic is still the most practical account book: whether the click-through rate is high, whether the order is closed, and whether the budget increases accordingly.

The separate revenue of Yuanbao can only explain part of Tencent's AI value. AI makes advertising recommendations more accurate, makes merchants more willing to invest, and the new revenue is still recorded under the "marketing services" item. It can also be seen that the proportion of Tencent's marketing services in total revenue in the second quarter has risen to 21% from 19% in the same period of the previous year.

In the future, when observing the AI commercialization of leading tech companies, simply counting the revenue of AI products will be increasingly inaccurate. The increments in advertising efficiency, transaction conversion, game development and enterprise services are also worth calculating.

JD.com provides another sample. In the second quarter, JD.com's revenue was 3464 billion yuan, down 2.9% year-on-year, and the operating profit of JD Retail still reached 135 billion yuan, with an operating profit margin of 4.6%. AI has been extensively integrated into the product, supply chain and service systems. During the 618 shopping festival, JoyInside has cooperated with nearly 200 brands, and the total number of connected devices has increased by more than 3 times compared with the previous year's Double 11 shopping festival. In the first half of 2026, JD Industrial deployed more than 70 AI Agents in the business chain from procurement to fulfillment. The tasks faced by the models have extended from generating text to product selection, price comparison, order placement and fulfillment. The value of AI has evolved from adding an intelligent assistant to improving operational efficiency.

The healthcare scenario is more intuitive. After JD Health upgraded the "Dawei Doctor" service, it connected online consultation, home testing, nursing and medicine purchase into the same service process. During the 618 shopping festival, the number of users served by "Dawei Doctor" was nearly 4 times that of the same period last year. After users put forward demands, AI also needs to invoke services, connect products and arrange subsequent links. After Agents enter the business system, answering questions is only the first step, and transactions and fulfillment determine whether the value can be realized.

Industry trends are also moving in the same direction. IDC data shows that 27% of Chinese enterprises have put AI Agents into production. Another survey of global enterprises shows that 50% of organizations have deployed Agents in multiple business fields, and another 27% are running Agents in at least one business field. Enterprises' requirements for AI are shifting from "what content can be generated" to "what tasks can be completed". Procurement, supply chain, customer service, marketing, R&D and finance have all become scenarios where Agents are widely deployed.

Tencent owns WeChat, advertising, games and payment businesses, while JD.com owns products, supply chain, logistics, healthcare and industrial customers. As model capabilities become increasingly similar, the data, user entrances and transaction processes accumulated by existing businesses have become important again. A mature business network is difficult to replicate in a short period of time. The more specific tasks AI performs in the physical world, the richer the data feedback generated, and the easier it is for model capabilities to be converted into revenue, efficiency and user stickiness.

Overall, the five companies are most likely to continue to move forward along the businesses they are most familiar with. Alibaba focuses on cloud and e-commerce, Tencent holds the traffic entrance of social networking and games, Baidu sticks to its traditional search business, Kuaishou relies on its user base and short videos, and JD.com continues to focus on the supply chain. AI has not eliminated the original differences, but instead amplified the existing advantages of each company.

The next round of competition among major companies in the AI wave will eventually return to the essence of business. After AI integrates into the original ecosystem of leading tech companies, how much more revenue is earned from advertising, how much cost is saved in the supply chain, how much transaction efficiency is improved, and whether users are retained, are the real performance reports they need to submit. This also means that the practice of AI in China is moving towards a broader application stage.

This article is from the WeChat official account "Xiu Tai", author: Shi Can, and is published with authorization from 36Kr.