By Xie Zefeng | Edited by Yang Xuran
"The biggest super-application of the AI era is an AI cloud platform centered on AI computing power."
Wu Yongming, CEO of Alibaba, presented this view during the Q1 earnings call for fiscal year 2027. His logic is that, in the long run, most enterprise AI workloads—including large-scale training, large-scale inference, customized inference, and enterprise Agent development and operations—require support from a full-stack AI cloud platform.
If AI cloud is a super-city, then training, inference, and Agents are the "residents" living within it. The more complete the cloud infrastructure, the more new customers it attracts, while simultaneously improving retention among existing clients, creating a positive cycle.
In the past, cloud services provided tools for enterprises or individuals to get work done; today, AI cloud delivers the "capability to get work done." Previously, cloud services underwent their first revolution driven by virtualization. In the AI era, they are experiencing a second revolution, upgrading from supplying resources to outputting intelligence.
As the AI era arrives, the landscape of cloud services is being reshaped. The competitive dimension among enterprises has shifted completely from "a battle of scale and price" to "a contest of comprehensive intelligent capabilities," imposing extremely high requirements on the technical capabilities and capital strength of market participants.
A profound transformation in the cloud industry is unfolding, but very few players have the qualification to join the table.
This article is an in-depth value piece from the Juchao WAVE content team. We welcome your attention across multiple platforms.
A Game for the Few
Without cloud support, AI is merely an expensive fireworks display.
In 2006, after Amazon launched cloud services accessible to both enterprises and individuals, public cloud began its commercial practice in China. However, since 2017, Alibaba Cloud and Tencent Cloud pushed the price war to its peak, directly leading to the decline of traditional CDN giant Wangsu Science & Technology.
Even more notably, Tencent Cloud once won the Xiamen government cloud project with a bid of one cent. In the highly homogenized IaaS market, low pricing became the sole means of competition.
Entering 2026, the disruption brought by AI to the cloud computing industry can be described as worlds apart.
Today, when enterprises move to the cloud, the first question has shifted from "how much can we save?" to "can it run my model?" The positioning of the cloud has changed from an "IT alternative" to the "production soil for AI."
In Jensen Huang's "Five-Layer Cake" framework for AI (Energy, Chips, Infrastructure, Models, Applications), cloud computing is the primary delivery form of the "Infrastructure Layer" and also serves as the core participant responsible for vertical integration across all five layers.
Nowadays, the competitive dimensions among cloud players have turned to a contest of comprehensive capabilities including computing power supply, model performance, platform engineering, ecosystem synergy, and industry implementation. Industry barriers have significantly increased.
Moreover, cloud vendors are evolving into integrated entities combining "Power + Chip + Model + Cloud." They are forced to extend upstream in the industrial chain, needing to become "all-around" players who self-develop chips, build their own power facilities, develop model platforms, and provide cloud services.
Consequently, battles among cloud vendors have become true super-capital wars.
The capital expenditures of giants are astronomical. The combined CapEx of the top five North American cloud vendors reached USD 411.5 billion in 2025, with consensus expectations for 2026 rising to USD 773.9 billion, and projected to reach USD 1.02 trillion in 2027.
Investments of this magnitude ensure that only a very few players remain at the table. Domestically, Alibaba plans to invest at least CNY 380 billion over three years from 2026 to 2028 for cloud computing and AI infrastructure construction, a figure exceeding the total investment of the previous decade.
Huge capital ammunition creates a "monopoly dividend," causing small and medium-sized vendors to hesitate in the AI race, widening the gap with leaders.
In 2025, the Chinese AI cloud market size reached CNY 68.4 billion, with Alibaba, ByteDance, Baidu, Tencent, and Tianyi Cloud ranking in the top five. Alibaba Cloud leads with a 38.1% share.
However, scale alone cannot secure a foothold in the AGI race; full-stack self-research capability has become the key test for the market. Merely purchasing chips means that no matter how large the computing cluster, it remains "paper wealth." Therefore, top-tier cloud vendors are desperately engaging in chip self-development.
Globally, Google, Microsoft, Amazon, Alibaba, Huawei, and Baidu are deploying in both cloud computing and AI chips simultaneously, with some vendors also self-developing large models.
As these self-developed chips scale up, gross margins for cloud businesses will see significant recovery, an advantage unmatched by vendors relying solely on NVIDIA GPUs.
Currently, "Model-as-a-Service" (MaaS) has become a new service tier following IaaS, PaaS, and SaaS, constituting a fourth barrier to business models.
MaaS initially charges based on Token consumption. Public cloud large model invocation volume in China surged nearly 17-fold in one year, skyrocketing from 114.2 trillion in 2024 to 1,944 trillion in 2025.
With the popularity of AI Agents and Personal agents, billing units will further evolve to charge by "task" or "business outcome," marking a historic restructuring of the economic boundaries of cloud computing.
Battle of the Two Giants
Among domestic enterprises, Alibaba Cloud, Tencent Cloud, Tianyi Cloud, and Baidu Smart Cloud represent different paradigms: internet platform-based, ecosystem scenario-based, national strategy-based, and technology-deepening-based, respectively. Their competitiveness and characteristics unfold along their respective genetic codes and strategic depths.
As the pioneer of China's cloud industry, Alibaba Cloud is the largest domestically and a leader in full-stack self-research. Its strategic layout resembles that of Google across the ocean. It dominates both the IaaS market and the incremental track of AI cloud. In 2025, Alibaba's growth rate for public cloud training and inference of large models reached 116%, with a market share of 42.2%.
Alibaba's strength lies in its full-stack investment in "Large Language Models + Chips + Cloud." T-Head's AI chip, Zhenwu, has been deployed at scale on Alibaba Cloud. Self-developed AI supernodes support efficient inference for large models with over 2 trillion parameters. Qwen3.8's performance ranking is approaching domestic SOTA levels such as Kimi K3 and GLM 5.3.
In the latest fiscal quarter, while revenue from AI & Cloud business increased, profit margins continued to rise. Both overall and external revenue growth rates for Alibaba Cloud were 45%. Among them, AI-related revenue reached CNY 12.4 billion, increasing by over 150% year-on-year, accounting for nearly 26% of total segment revenue. Profit margins also entered double digits, approaching 12%.
At the FY2026 exchange meeting, Wu Yongming stated that AI ARR had surpassed CNY 8 billion and would exceed CNY 30 billion by year-end. "The advantage of high profit margins is gradually becoming prominent and will serve as a pillar for healthy, high-quality revenue growth for Alibaba in the future."
In March, Alibaba Cloud announced price increases of 5%-34% for certain AI computing products, breaking the industry convention of "only decreasing, never increasing" prices for cloud services. This reflects the exponential growth in Token invocation volume due to surging AI demand and the resonance of upstream hardware price hikes. Deeper still, it signifies the return of pricing power for top-tier cloud vendors, shifting from scale-cost competition to technology and value-driven strategies.
This combination of "full-stack AI capabilities" and the endowment of being a strong player in both "computing power + large models" makes Alibaba Cloud one of the biggest beneficiaries in the AI era.
Tencent Cloud, another giant, relies on its incomparable ecosystem advantage. Its competitiveness in the AI era lies not in head-to-head battles of cloud infrastructure scale, but in the unique multiplier effect of its "C-end Ecosystem × Scenario Application."
The WeChat ecosystem is a natural digestion pool for AI computing power. With a massive user base of 1.4 billion and richly diverse embedded businesses, it provides vast soil for AI cloud.
After enduring cash burn and streamlining, Pony Ma confirmed at the earnings call that the cloud business achieved scaled profitability in 2025, a rarity among domestic cloud vendors. Tencent Cloud's strategy is "not pursuing the number one market share" but focusing on scenario applications.
For instance, Tencent dominates the AI office scenario. The AI office agent WorkBuddy recorded 20.97 million monthly visits (June), ranking first. The AI coding agent CodeBuddy ranks among the top. Moreover, the gross margin for paid users of Work Buddy is comparable to that of Tencent Cloud overall.
The AI assistant Xiao Wei focuses on the AI-driven reshaping of the internal WeChat ecosystem. Martin Lau positions it as a key carrier for the WeChat ecosystem entering the "AI era," using the analogy of WeChat's "10x value amplification" relative to QQ to illustrate the potential of this leap.
Compared to companies like Alibaba and Google, Tencent's large model capabilities remain a concern for the market. To address this, Tencent is going full throttle, spending CNY 52.8 billion (actual CapEx) in the second quarter, plus a prepayment of CNY 51.4 billion for computing power procurement, totaling over CNY 100 billion. Following this trend, annual CapEx could surge to CNY200 billion.
Currently, pushing Hunyuan to achieve SOTA status is a clear strategic priority. The internal mission is "Hunyuan 5 must become a SOTA model."
Dislocated Competition
Beyond the two internet giants, Tianyi Cloud and Baidu Smart Cloud are also important players in the current AI cloud market.
China Telecom's Tianyi Cloud can be viewed as the "National Cloud Team," shouldering critical tasks in information security. The barriers it has built in tracks such as government affairs, state-owned enterprises, and Xinchuang (domestic innovation) are difficult for internet cloud vendors to replicate in the short term. The deep integration of state-owned identity, network resources, localized services, and full-stack domestication capabilities enables Tianyi Cloud to fulfill the mission of "national infrastructure."
IDC data shows that Tianyi Cloud continues to lead the Chinese government public cloud market with a 26.1% market share, ranking first for four consecutive years (2021–2024). In the second half of 2025, Tianyi Cloud broke into the top three of the public cloud IaaS market, ranking second with a 12.1% share, surpassing Huawei Cloud, Mobile Cloud, and others.
In the first half of 2026, Tianyi Cloud's revenue reached CNY 61.8 billion, a year-on-year increase of 7.8%. The company continues to deepen its "five-in-one" intelligent cloud system construction, reshaping its operational development model with Tokens as the main line. The total computing power of the "Xirang" platform exceeds 118.8 EFLOPS, with proprietary capabilities reaching 70.8 EFLOPS.
Tianyi Cloud builds differentiated barriers through "National Cloud + National Chip + National Model." From Zijinshan servers, Zijin DPU, CTyunOS operating system to TeleDB database, Tianyi Cloud has connected the full-stack domestic software and hardware matrix. This advantage gives it distinctive characteristics in government and enterprise scenarios, differentiating it from internet cloud vendors.
Furthermore, as a major operator, its cloud services possess the inherent advantage of "computing-network convergence." Nationwide localized service capabilities, network systems, and barriers in the government/enterprise sector allow it to deeply bind with customers who have extremely high requirements for security and compliance.
However, the incremental space in the current government/enterprise market is narrowing. In the AI cloud market, Tianyi Cloud's share lags significantly behind vendors like Alibaba. Whether it can shore up weaknesses in the PaaS and SaaS layers and transition from "general computing" to "intelligent computing" is key to determining its competitiveness in the next stage.
Baidu Smart Cloud is a "full-stack self-research" vendor of the AI-native school.
In the traditional public cloud IaaS market, it does not rank in the first tier, but in the AI cloud market, it has risen to the top three thanks to its full-stack self-research capabilities. Therefore, it may not be the "largest cloud," but it might be the "most AI-centric cloud."
It has constructed a four-layer full stack of "Chip-Cloud-Model-Agent," which is almost unique domestically: "Chip (Kunlunxin)—Cloud (Baidu Smart Cloud)—Model (ERNIE Bot)—Agent (Super Intelligent Agent)."
Alibaba has chips and cloud, but its model layer coexists with multiple open-source models. Tencent Cloud relies on ecosystems but depends on externally sourced underlying chips. Baidu is one of the few players with full-chain self-research from chips to applications.
This layout addresses the efficiency losses caused by the fragmentation between chips, frameworks, models, and applications in traditional cloud computing. Shen Dou likens this "instruction-set-level native alignment" to Baidu Smart Cloud's "new full-stack" technical architecture in the Agent era.
In engineering terms, it means making the model understand the strengths of the chip, making the cloud understand the shape of the model, making the Agent understand the boundaries of the cloud, and having all four layers share the same optimization language. Only this native alignment can maximize Token computation efficiency.
In the core track of AI public cloud services, Baidu Smart Cloud is a long-term leader and the "King of Winning Bids" for large model implementation. In the first half of the year, disclosed winning bid amounts for Baidu Smart Cloud's large model projects reached CNY 1.386 billion, several times higher than other cloud vendors.
In the second quarter of this year, AI-related revenue reached CNY 12.4 billion, maintaining a share of over 50% for two consecutive quarters. AI cloud infrastructure continued its strong momentum, with GPU cloud growing 283% year-on-year.
The vertical integration from Kunlunxin to ERNIE Bot allows Baidu to achieve substantial cost advantages. Kunlunxin P800 has already enabled training of ERNIE Bot 5.1 on a cluster of 30,000 cards, and the majority of inference tasks have been migrated to the self-developed chip system.
However, Baidu's shortcomings are equally clear. Its overall scale and ecosystem depth are slightly insufficient, its foundation in general IaaS is weak, and traditional cloud ecosystem gaps are obvious. It failed to enter the top five in the traditional public cloud IaaS market. The leap from "AI Cloud" to "Strongest Comprehensive Cloud" is a challenge Baidu Smart Cloud must face. Indeed, this is also a critical question for whether Baidu's ecosystem can prosper again in the AI era.
In Conclusion
In the AI era, the cloud is no longer just an upgraded server room, nor a wholesale provider of computing power.
It is the infrastructure mentioned by Jensen Huang and the super-application emphasized by Wu Yongming. Almost all AI application costs are determined by it, and all progress in intelligence is constrained by its delivery capabilities.
Models iterate every three months, chips update annually, and Agent inference traffic surges in tidal patterns. Facing the severe uncertainty of the AI world, the cloud is the only buffer carrier for computing power.
Undoubtedly, the coming period will be a few years where Chinese cloud vendors experience prosperity amidst intense competition, thriving business formats, and coexisting developmental anxiety.