Cloud Computing in the AI Era, the Prosperity of the Minority

"The largest super app in the AI era is an AI cloud platform centered on AI computing power."

Wu Yongming, CEO of Alibaba, put forward such a view at the Q1 2027 fiscal quarter earnings meeting. His logic is that in the long run, most of enterprises' AI workloads, including large-scale training, large-scale inference, customized inference, and enterprise Agent development and operation, all require full-stack AI cloud platform support.

If AI cloud is regarded as a super city, then training, inference and Agents are the "residents" living inside. The more perfect the cloud infrastructure is, the more it can attract new customers and improve the retention of existing customers at the same time, forming a positive cycle.

In the past, the cloud brought productivity tools to enterprises or individuals. Today, AI cloud delivers "the capability to get work done". Previously, cloud services completed the first revolution brought by virtualization, and in the AI era, it is undergoing the second revolution, upgrading from supplying resources to outputting intelligence.

With the advent of the AI era, the pattern of cloud services is also being reshaped. The competition dimension between enterprises has completely shifted from "competition over scale and price" to "competition over comprehensive intelligence capability", which puts extremely high requirements on the technical capability and capital strength of market participants.

A profound transformation of the cloud industry is kicking off, but very few players are eligible to take a seat at the table.

01

Game for the Few

Without the support of the cloud, AI is nothing but an expensive fireworks display.

In 2006, after Amazon launched cloud services available to both enterprises and individuals, public cloud kicked off its commercial practice in China. However, since 2017, Alibaba Cloud and Tencent Cloud have pushed the price war to its peak, directly leading to the decline of traditional CDN giant Wangsu Science & Technology.

What's more, Tencent Cloud once won the bid for the Xiamen government cloud project with a quotation of 1 cent. In the highly homogeneous IaaS market, low price has become the only means of competition.

Figure: Announcement of the winning bid for the Xiamen municipal government extranet cloud service project. (Screenshot from China Government Procurement Network)

After entering 2026, the subversion brought by AI to the cloud computing industry can be described as a whole new world that is unrecognizable from the past.

Now when enterprises move to the cloud, the first question has shifted from "how much cheaper can it be" to "can it run my model". The positioning of the cloud has changed from "an alternative to IT" to "the production soil for AI".

In the five-layer AI "cake" system proposed by Jensen Huang (energy, chip, infrastructure, model, application), cloud computing is the main delivery form of the "infrastructure layer", and it is also the core participant responsible for vertical integration among the five layers.

Now, the competition dimension between cloud players has turned into a contest of comprehensive capabilities such as computing power supply, model performance, platform engineering capability, ecological collaboration and industry implementation, and industry barriers have increased significantly.

Moreover, cloud computing vendors are also evolving into integrated complexes of "power + chip + model + cloud". They are forced to extend upstream along the industrial chain and need to become "all-round" players integrating self-developed chips, self-built power facilities, model platforms and cloud services.

Therefore, the battle among cloud vendors has become a veritable super capital war.

The capital expenditure of the giants is at an astronomical level. The total capital expenditure of the five major North American cloud vendors reached 411.5 billion U.S. dollars in 2025, the consensus expectation for 2026 is even as high as 773.9 billion U.S. dollars, and it is expected to reach 1.02 trillion U.S. dollars in 2027.

Investment of this scale means that only a very small number of players can stay at the table. In China, Alibaba plans to invest at least 380 billion yuan in the construction of cloud computing and AI infrastructure in the three years from 2026 to 2028, a figure exceeding the sum of the previous ten years.

Huge capital ammunition has formed a "monopoly dividend", which makes small and medium-sized vendors hesitate to participate in the AI race, and the gap between them and the top players is widening.

In 2025, the scale of China's AI cloud market reached 684 billion yuan, with Alibaba, ByteDance, Baidu, Tencent and eSurfing Cloud ranking the top five. Alibaba Cloud leads the market with a 38.1% share.

However, scale alone is not enough to gain a foothold in the AGI race, and the full-stack self-development capability of cloud vendors has become the key to market tests. Even a computing power cluster of enormous size is just "paper wealth" if it only purchases chips from external suppliers. Therefore, top cloud vendors are sparing no effort to develop self-developed chips.

Globally, Google, Microsoft, Amazon, Alibaba, Huawei and Baidu have all deployed in both cloud computing and AI chips, and some vendors have also self-developed large models.

With the large-scale rollout of these self-developed chips, the gross profit margin of cloud business will be significantly improved, which is an unmatched advantage for vendors that only purchase NVIDIA GPUs.

At present, "Model as a Service" (MaaS) has become a brand-new service tier following IaaS, PaaS and SaaS, which forms the barrier of the fourth business model.

In the initial stage of MaaS, fees are charged based on Token consumption. The call volume of large models on China's public cloud has exploded by nearly 17 times within one year, soaring from 114.2 trillion in 2024 to 1944 trillion in 2025.

With the popularity of AI Agents and Personal Agents, the billing unit will further evolve to billing by "task" or "business result", and the economic boundary of cloud computing is undergoing historic restructuring.

02

Two Giants Vying for Dominance

Among domestic enterprises, the four vendors of Alibaba Cloud, Tencent Cloud, eSurfing Cloud and Baidu Intelligent Cloud respectively represent different paradigms: internet platform type, ecological scenario type, national strategic type and technology deep cultivation type. Their competitiveness and characteristics are expanding along their respective genes and strategic depth.

As a pathfinder of China's cloud industry, Alibaba Cloud is the largest in China and also the leader in full-stack self-development, and its strategic layout is similar to that of Google across the Pacific. It takes the lead in both the IaaS market and the AI cloud incremental track. In 2025, the growth rate of Alibaba's public cloud for large model training and inference reached 116%, and its market share even reached 42.2%.

Alibaba's strength lies in its full-stack investment in "large language model + chip + cloud". The Pingtouge AI chip Zhenwu has been deployed on a large scale on Alibaba Cloud, and self-developed AI super nodes support efficient inference of large models with over 2 trillion parameters. The performance of Qwen 3.8 has once again approached the domestic SOTA level of Kimi K3, GLM 5.3 and others.

In the latest fiscal quarter, while the revenue of the AI & cloud business increased, its profit margin also continued to rise. The overall and external revenue growth rate of Alibaba Cloud both reached 45%, of which AI-related revenue reached 12.4 billion yuan, a year-on-year increase of over 150%, accounting for nearly 26% of the total revenue of the segment, and the profit margin also stood at double digits, close to 12%.

Wu Yongming said at the 2026 fiscal year exchange meeting that AI ARR has exceeded 8 billion yuan, and will exceed 30 billion yuan by the end of the year. "The advantage of high profit margin is gradually emerging, and it will be the pillar of Alibaba's healthy and high-quality revenue growth in the future."

In March, Alibaba Cloud also announced a 5%-34% price increase for some AI computing products, breaking the long-standing industry convention of cloud services that "prices only fall and never rise". This reflects the resonance of exponential growth in Token call volume caused by the AI demand explosion and the price rise of upstream hardware, but at a deeper level, it is the return of pricing power of top cloud vendors shifting from scale cost competition to technology and value-driven development.

This endowment of "full-stack AI capability" superimposed on the dual strengths of "computing power + large model" makes Alibaba Cloud still one of the biggest beneficiaries in the AI era.

The other giant, Tencent Cloud, still relies on its incomparable ecological advantages. The competitiveness of Tencent Cloud in the AI era does not lie in the scale confrontation of cloud infrastructure, but in the unique multiplier effect of its "C-end ecosystem × scenario application".

The WeChat ecosystem is a natural consumer pool for AI computing power. WeChat has an ecological thickness of 1.4 billion massive users and extremely rich diverse embedded businesses, all of which provide vast soil for AI cloud.

After the period of burning money and streamlining operations, Pony Ma confirmed at the earnings meeting that the cloud business achieved large-scale profitability in 2025, which is still rare among domestic cloud vendors. Tencent Cloud's strategy is "not to pursue the first place in market share", but to focus on scenario applications.

For example, Tencent takes the lead in the AI office scenario. The monthly visits of AI office agent WorkBuddy reached 20.97 million times (in June), ranking first, and AI programming agent CodeBuddy is at the forefront. Moreover, the gross profit margin of Work Buddy's paying users is already comparable to that of Tencent Cloud as a whole.

AI Assistant Xiaowei focuses on the AI-oriented reshaping of WeChat's internal ecosystem. Martin Lau positions it as a key carrier for the WeChat ecosystem to enter the "AI era", and uses the analogy of WeChat's "10-fold value amplification" relative to QQ to illustrate the potential of this leap.

Compared with enterprises such as Alibaba and Google, Tencent's large model capability is still a concern for the market. To this end, Tencent is going all out, spending 52.8 billion yuan (actual capital expenditure) in the second quarter, and there is also a prepayment for computing power procurement as high as 51.4 billion yuan, totaling over 100 billion yuan. According to this trend, the annual CapEx may rush to 200 billion yuan.

At present, it is a clear strategic priority for Tencent's Hunyuan to impact SOTA, and the internal task is to "make Hunyuan 5 a SOTA model".

03

Displaced Competition

Apart from the two internet giants, eSurfing Cloud and Baidu Intelligent Cloud are also important players in the current AI cloud market.

China Telecom's eSurfing Cloud can be regarded as the national cloud team, shouldering the important task of information security. The barriers it has built in the tracks of government affairs, central SOEs and local SOEs, and information technology application innovation are difficult for internet cloud vendors to replicate in the short term. The deep integration of state-owned assets identity, network resources, localized services and full-stack domestic independent controllable capabilities enables eSurfing Cloud to undertake the mission of "national infrastructure".

IDC data shows that eSurfing Cloud continues to lead China's government affairs public cloud market with a 26.1% market share, and has ranked first for four consecutive years (2021-2024). In the second half of 2025, eSurfing Cloud broke into the top three of public cloud IaaS, ranking second with a 12.1% share, leaving Huawei Cloud and China Mobile Cloud behind.

In the first half of 2026, eSurfing Cloud's revenue reached 61.8 billion yuan, a year-on-year increase of 7.8%. The company continues to deepen the construction of the "five-in-one" intelligent cloud system, reshaping its business development model with Token as the main line. The total computing power of the "Xirang" computing power platform has exceeded 118.8 EFLOPS, and its self-owned capability reaches 70.8 EFLOPS.

eSurfing Cloud builds differentiated barriers with "national cloud + national chip + national model". From the Purple Mountain server, Zijin DPU, CTyunOS operating system to TeleDB database, eSurfing Cloud has got through the full-stack domestic independent controllable matrix of software and hardware. This advantage makes it have differentiated characteristics distinct from internet cloud vendors in government, enterprise and information technology application innovation scenarios.

Moreover, as a large-scale operator, its cloud service has the innate advantage of "computing and network integration". Its nationwide localized service capabilities, network system and government-enterprise track barriers can deeply bind customers with extremely high requirements for security and compliance.