Author | Hu Min, Editor | Bao Yonggang
International market research firm Omdia recently released its China Cloud Market Report for Q3 2025.
Amid dense datasets, one figure stands out remarkably: Alibaba Cloud’s market share jumped from 33% to 34%, and then rose further to 36%.
This 3-percentage-point leap, seemingly calm on the surface, is actually a strong signal of a major industry shift.
According to the law of diminishing marginal returns, market share for leading vendors in mature markets should generally converge — the larger the base, the harder it becomes to grow.
Furthermore, internet traffic dividends have peaked, and most migration of traditional enterprises to the cloud is largely completed. Expanding market share further is akin to scraping grain from rock crevices on an already fully farmed field; the challenge is far greater than before.
Yet the reality is that Alibaba Cloud has not only held its 33% baseline but also captured an additional 3 percentage points of growth across three quarters.
What exactly drives this “anti-gravity” growth? And why is Alibaba Cloud seeing this expansion rather than other cloud vendors?
01. Where Exactly Does the Upward 3% Come From?
Alibaba Cloud’s market share growth is highly unlikely to stem solely from legacy system migration.
Zhou Yu, formerly an SME sales staff at a cloud provider, told Leiphone that the vendor’s official website used to attract 4,000–6,000 organic enterprise clients daily. In the post-pandemic period after 2022, this figure halved.
So where is this new source of business coming from? The likely answer is structural incremental growth powered by AI.
Still, many question whether AI truly moves the revenue needle for large cloud providers.
This skepticism is understandable. In many observers’ view, enterprise AI adoption simply means calling a few large-model APIs. The current API market has long turned into a red ocean. For Qwen, Hunyuan and ERNIE Bot, pricing per call has fallen to just a few cents, or even less than one cent.
If AI is reduced merely to “selling tokens,” even one billion API calls would be merely a drop in the bucket for Alibaba Cloud, whose annual revenue runs into hundreds of billions — far too little to deliver that 3-percentage-point market share gain.
But this skepticism overlooks a qualitative shift underway in enterprise AI deployment:
Over the past year, corporate AI demand has rapidly evolved from isolated pilot tests toward systemic business reconstruction.
Yi Zhang, Senior Analyst at Omdia, stated: “A single model’s performance can no longer satisfy enterprises’ needs in real-world business scenarios. The core challenge of scaling AI projects lies in how to collaboratively orchestrate models, data, tools, and workflows within complex systems to achieve reusability, manageability, and commercialization.”
In short, customers are no longer satisfied with simply “mounting” a chatbot. They are starting to deeply embed models within core business workflows.
When AI transforms from a plugin into the core engine of a business, cloud resource consumption patterns change entirely. It is no longer lightweight, pay-per-call invocation; instead, it evolves into high-density, continuous consumption of infrastructure including compute power, storage and databases.
This is the real secret behind the flywheel effect lifting Alibaba Cloud’s performance.
We can examine live real-world use cases, each supported by massive cloud infrastructure:
In scientific research, the steel material design large model co-launched by Alibaba Cloud and the New Materials Big Data Center may look like a simple Q&A interface on the surface, yet underneath it is a full-chain closed loop built on Qwen, covering data mining, performance prediction and knowledge Q&A. To hit prediction accuracy up to 90%, relevant material datasets must be cleaned, pre-trained, fine-tuned and continuously run for inference in the cloud. This consumes GPU compute power and directly boosts usage of object storage and cloud-native databases.
In consumer electronics, ECOVACS’ collaboration with Alibaba Cloud reaches deep into embodied intelligence. It is not merely about letting robot vacuums understand commands such as “go clean the kitchen”, but also about deploying models ranging from 0.7B to 7B parameters across edge and cloud environments simultaneously. To deliver real-time intelligent experiences for 38 million households globally, ECOVACS must build an ultra-high-concurrency cloud inference cluster. This places stringent demands on cloud network throughput and elastic computing stability.
In industrial internet, leading companies including Marriott, GAC and Haier are pursuing the same goal: turning Qwen into their proprietary “enterprise brain”. To operate this brain, businesses build dedicated Retrieval-Augmented Generation (RAG) systems, deploy vector databases to store massive knowledge bases, and adopt advanced security products to protect data assets.
This creates a powerful cloud flywheel:
First, AI fuels compute demand. Model training and inference have more than doubled GPU-related revenue. Analyst estimates show Alibaba Cloud’s GPU revenue grew by over 100% in H1 2025.
Second, compute drives data. Compute power ultimately serves data workloads. Massive intermediate datasets, chat logs and AI-generated content produced by AI workloads have spurred strong growth in database and storage product consumption.
Finally, data strengthens customer stickiness. When an enterprise’s core business logic and data assets reside on Alibaba Cloud’s AI infrastructure, migration costs become prohibitively high. Renewal rates and average revenue per user (ARPU) naturally rise as a result.
Omdia’s data also supports this trend: Driven by AI, mainland China’s cloud infrastructure services market reached $13.4 billion in Q3 2025, representing 24% year-on-year growth.
02. What Did Alibaba Cloud Do Right?
Fueled by strong AI tailwinds, Alibaba Cloud has traced an upward curve over three quarters. But widening the view reveals the flip side of this growth: brutal market-share divergence.
Data shows that as Alibaba Cloud’s share climbs, second and third-place vendors face headwinds. Huawei Cloud’s market share fell from 18% to 16%, while Tencent Cloud slipped from 10% in Q1 to 9%.
This scissors gap between rising and falling market share reveals a critical industry signal: In the AI era, cloud computing’s competitive threshold has been completely reshaped, with market demand rapidly concentrating toward top-tier players.
Why is the market tilting this way? AI engineering carries extreme complexity and capital requirements that directly segment industry participants.
Alibaba Cloud got three key things right to build its 36% market moat:
First, a massive RMB 380 billion investment addresses compute hunger.
AI training and inference consume compute resources voraciously. What customers now seek is not a handful of servers, but GPU clusters comprising thousands or tens of thousands of cards.
Over the past decade, Alibaba cumulatively invested roughly RMB 130 billion in infrastructure. At the start of 2025, however, Alibaba CEO Eddie Wu announced plans to invest RMB 380 billion in AI infrastructure over the following three years. During the November 2025 earnings call, an executive even noted this figure “might be conservative.”
This means Alibaba’s annual infrastructure investment may now exceed its total spend across the prior decade.
This near-transformative investment has widened gaps with competitors. When clients require tens-of-thousands-GPU clusters, Alibaba Cloud is among the very few vendors able to deliver immediate availability and guarantee stable operation. By contrast, rivals constrained by supply-chain limits or more moderate investment strategies cannot match this supply capacity.
Second, cost advantages from full-stack optimization.
Alibaba Cloud ranks among the world’s few cloud providers with fully in-house full-stack development spanning chips (T-Head), cloud platform and large models (Tongyi Lab).
This is not merely a technical showcase; it delivers tangible commercial benefits. Its underlying Cloud Infrastructure Processing Unit (CIPU) optimizes network latency, while the Platform for AI (PAI) manages VRAM scheduling to push GPU utilization to its limits.
Simply put, running the same model on Alibaba Cloud can deliver an overall cost 10%–30% lower than competing platforms. For AI firms burning tens of millions monthly on compute, that 10%–30% gap can mean the difference between survival and failure.
Third, entry advantages from its open-source ecosystem.
If hardware technology represents hard power, Qwen’s open-source strategy is Alibaba Cloud’s most impactful strategic play.
Data shows cumulative downloads across the entire Qwen model family have exceeded one billion. Behind this number sits a large community of developers and enterprise users.
Open source may appear free, yet it fuels cloud revenue growth.
As millions of developers grow accustomed to working with Qwen models, and large numbers of small and mid-sized enterprises build applications atop Qwen, they discover that deploying and running Qwen on Alibaba Cloud delivers the smoothest experience and optimal performance.
Qwen’s one billion downloads effectively filter and nurture globally active potential AI customers for Alibaba Cloud. Through open-source models, Alibaba Cloud has built an App Store-style closed-loop ecosystem via ModelScope: developers source models within the community and run fine-tuning and inference workloads on Alibaba Cloud.
This creates another positive flywheel: stronger models drive more downloads; a more vibrant ecosystem increases AI compute consumption on the cloud.
03. Mirror Validation Across the Ocean
This head-player siphon effect is by no means unique to China’s market. Across the Pacific, Google Cloud is following an identical playbook.
In Q4 2025, Google Cloud revenue surged 48% year-on-year to $17.664 billion. Its annual revenue surpassed $70 billion for the first time, with backlog hitting $240 billion, more than doubling year-on-year. Management explicitly attributed this growth “primarily to a surge in demand for enterprise AI infrastructure and AI solutions.”
Three years ago, Google Cloud was widely seen as the laggard in the global cloud market, with its share stuck near 10% for a long time, viewed externally as a permanent third place unable to catch up with AWS and Azure. Within just two years, it completed its leap from follower to leader.
The logic behind this leap mirrors Alibaba Cloud’s.
First, compute positioning. Google is one of very few global cloud vendors with full-stack in-house development covering chips, systems, models and platforms. Its Tensor Processing Unit (TPU) is evolving from an internal chip into a differentiated offering for external customers. Media reports state Anthropic plans to purchase nearly one million units, and Meta is in advanced procurement negotiations. While competitors wait for NVIDIA hardware deliveries, Google can offer immediate availability of tens-of-thousands-GPU clusters to customers.
Second, cost compression. Full-stack in-house development directly boosts effective compute output per unit resource. The TPU’s deep optimization for the Transformer architecture drives far lower overall costs for large-model training compared with general GPU alternatives. For AI companies spending tens of millions monthly on compute, this translates to substantial cash savings.
Third, ecosystem lock-in. Gemini’s monthly active users exceed 750 million, and over 70% of existing cloud customers use Google’s AI services. This is not just an add-on feature; it is the anchor retaining customers on Google Cloud. When an enterprise’s data assets, model fine-tuning workflows and inference pipelines are built on Google’s AI infrastructure, switching costs become prohibitively high.
This is no coincidence. It represents parallel validation of fundamental industry laws across two markets —
Alibaba Cloud’s rise from 33% to 36% market share and Google Cloud’s climb from 10% to 14% (with growth still accelerating) illustrate the same industry trend: In the AI era, compute power and large models act as new triggers for the Matthew effect. Players that establish leadership in these two areas can trigger the positive cycle of market siphoning.
Over the next three years, the key suspense within China’s cloud market may no longer revolve around who holds the number-one spot, but how long the second-tier vendors can keep pace.
Note: Zhou Yu is a pseudonym.