Mainland China’s cloud infrastructure market has returned to high-growth territory, with spend climbing to $12.4 billion in the second quarter of 2025, according to new figures from research firm Omdia. Year-on-year growth hit 21%, the first time the market has broken the 20% threshold since early 2024, underscoring how AI infrastructure demand is reshaping capital allocation and product strategy among Chinese hyperscalers.
AI remains the dominant growth catalyst. As foundation models improve and tooling matures, Chinese enterprises are moving beyond simple API-style “call a model and get a response” use cases. Instead, they are beginning to embed AI into specific business processes, build industry-focused models on proprietary data, and experiment with AI agents that can take actions across workflows rather than just answer questions. This shift is driving heavier, more persistent infrastructure consumption – particularly GPU capacity, high-bandwidth storage and higher-level AI platforms.
Omdia notes that some organizations are now redirecting parts of their traditional, CPU-based application logic and data flows into model inference engines. For cloud providers, that translates into demand not only for raw compute, but also for optimized orchestration, observability and cost management around AI-heavy workloads. It also increases the strategic value of owning the model, platform, and infrastructure stack end-to-end.
Vendors appear prepared to trade near-term margins for strategic position. Omdia’s analysts characterize this phase as a capability-building cycle rather than an optimization cycle: hyperscalers are investing ahead of revenue to secure GPU supply, expand data center footprints, and build AI platform ecosystems that will be difficult for smaller competitors to replicate over the next decade.
Alibaba Cloud, Huawei Cloud and Tencent Cloud remain the dominant players in Mainland China’s cloud infrastructure services market, with Q2 2025 shares of 34%, 17% and 10% respectively. Alibaba Cloud posted the fastest growth among the top three, with revenue up 26% year-on-year and AI-related revenue growing at triple-digit rates for the eighth consecutive quarter. AI has shifted from “adjacent line of business” to the central engine of Alibaba’s cloud strategy.
To support that trajectory, Alibaba Cloud boosted capital expenditure to CNY 38.6 billion ($5.4 billion) in the quarter and has announced an aggressive plan to invest CNY 380 billion (about $52.9 billion) over the next three years in cloud and AI infrastructure. That roadmap includes new regions in Brazil, France and the Netherlands, signaling that its AI infrastructure ambitions extend well beyond the domestic market.
On the product side, Alibaba has been rapidly iterating its Qwen3 model family. Following upgrades to multimodal capabilities and development toolchains in mid-2025, the company introduced Qwen3-Max in September, a frontier-scale model with more than one trillion parameters designed for complex analysis and agent-oriented workloads. Alongside the model, Alibaba launched Agent Bay, a cloud-based execution environment and toolset aimed at letting enterprises and partners build AI agents capable of performing real-world operational tasks, not just conversational support.
Huawei Cloud remained the second-largest provider with a 17% share and 17% year-on-year revenue growth in Q2. Its strategy hinges on tightly integrated “full-stack” AI, from chips and networking through to industry solutions. The company has expanded its CloudMatrix architecture from 384 to 8,192 GPUs, and layered on services such as AI Token Service and EMS to optimize inference-level compute utilization and cost.
For application delivery, Huawei is pushing ModelArts Versatile, its enterprise-grade agent platform. Recent enhancements add full lifecycle management for agents and integrate Model Context Protocol (MCP) tooling, with Huawei claiming up to 40% gains in development efficiency and roughly 30% reductions in deployment cost for AI-native applications. Sector focus remains strong: manufacturing, finance, public sector and automotive are all target verticals for AI-native cloud solutions.
Tencent Cloud, with a 10% share of the infrastructure market, is leveraging its broader ecosystem from gaming, social and media to differentiate its AI and cloud offerings. Revenue growth accelerated in Q2 as customers increased their use of GPUs and AI API tokens. In August, Tencent released four compact Hunyuan models (0.5B, 1.8B, 4B and 7B parameters), fully open-sourced on GitHub and HuggingFace, and optimized for deployment across a range of compute environments, from edge nodes to multi-GPU clusters.
In September, Tencent introduced Agent Development Platform 3.0, a full-stack upgrade that covers model invocation, agent design, orchestration and operations. ADP 3.0 focuses on reasoning, knowledge integration, workflow automation, multi-agent collaboration and enterprise-grade management, aiming to lower the barrier for partners building task-centric AI applications on Tencent’s infrastructure. Internationally, Tencent Cloud has committed $150 million to its first Middle East data center in Saudi Arabia and is planning a third data center in Osaka, reinforcing its role as a regional rather than solely domestic player.
A notable element in Omdia’s analysis is the growing role of partners. In Q2 2025, partner-driven cloud revenue accounted for around 25% of mainland China’s cloud infrastructure services market. As AI use cases become more domain-specific, cloud providers increasingly rely on independent software vendors, consulting partners and vertical specialists to translate foundational AI capabilities into sector-appropriate solutions, from manufacturing quality inspection to financial risk modeling.
The emergence of AI agent platforms from Alibaba, Huawei and Tencent also reinforces this ecosystem dynamic. Rather than treating enterprises purely as model consumers, these platforms aim to pull partners into higher-value roles: assembling toolchains, integrating with existing systems, and designing agents that can trigger transactions, update records or orchestrate multi-step workflows. For MSPs and ISVs in China and beyond, this opens new service lines but also increases dependence on provider-specific AI stacks.
Omdia defines “cloud infrastructure services” broadly as the combined market for bare metal as a service (BMaaS), infrastructure as a service (IaaS), platform as a service (PaaS), container as a service (CaaS) and serverless offerings hosted by third part. The Q2 2025 figures refer specifically to services delivered from data centers located in mainland China, rather than global revenue from Chinese-headquartered providers.
For global technology leaders, China’s return to above-20% cloud infrastructure growth underlines two related trends. First, AI-led cloud spending is no longer a short-term spike but appears to be entering a structural phase, with hyperscalers willing to commit tens of billions of dollars to capacity expansion and platform buildout. Second, the “AI stack” is moving up a level, from base models and APIs to agent platforms and domain solutions, which will have implications for how enterprises, partners and regulators engage with Chinese cloud ecosystems in the coming years.
Executive Insights FAQ
How should global CIOs interpret China’s return to 20%+ cloud infrastructure growth?
For CIOs, China’s numbers are a signal that AI infrastructure demand can sustain high double-digit cloud growth even in a relatively mature market. It suggests that as enterprises embed AI into workflows and build their own models, baseline IaaS and PaaS consumption can re-accelerate. Leaders outside China can expect similar patterns: once foundational AI capabilities stabilize and governance matures, infrastructure demand may shift from experimental peaks to more predictable, business-driven growth.
What is the practical significance of AI agents versus traditional chat-style AI for enterprises?
Agents matter because they move from “answering questions” to “taking actions.” Instead of a chatbot surfacing insights that a human must manually apply, an AI agent can trigger workflows in CRM, ERP or custom apps, coordinate with other agents and handle multi-step tasks. The platforms Alibaba, Huawei and Tencent are launching are designed to industrialize that pattern. For enterprises, this could mean faster automation of knowledge-heavy processes, but also a need for tighter controls, observability and integration testing as agents gain more operational authority.
How will hyperscalers’ massive AI capex influence pricing and access to GPU resources?
Large capex commitments help secure GPU supply and expand capacity, which should ease some of the acute shortages seen in earlier phases of the AI boom. Over time, enterprises may see more flexible pricing tiers, region choices and reservation models for AI compute. However, the scale of investment also locks customers deeper into provider ecosystems; preferential pricing may be tied to using a provider’s own models, storage and networking. Buyers should expect aggressive “land and expand” offers around AI infrastructure in exchange for longer-term commitments.
What opportunities does the 25% partner-driven revenue share create for MSPs and ISVs?
As roughly a quarter of China’s cloud infrastructure revenue now flows through partners, it underscores the importance of specialization. MSPs and ISVs that can combine cloud infrastructure, domain knowledge and AI capabilities are well-positioned to capture higher-margin work around design, integration, and ongoing optimization of AI workloads. The agent platforms from major providers offer a route to build recurring services – such as managing agent fleets, tuning workflows and monitoring business KPIs – rather than one-off implementation projects.
What should boards and risk leaders watch as AI-led cloud adoption accelerates in China?
Boards should focus on three areas: concentration risk, regulatory alignment and data governance. As AI workloads consolidate on a small number of hyperscalers, enterprises become more exposed to changes in terms, pricing or export controls. Regulatory frameworks for AI, data localization and cybersecurity in China are evolving quickly and may not mirror regimes in North America or Europe. Finally, as more business logic moves into opaque models and agents, boards will need assurance that auditability, fallback mechanisms and human oversight remain in place, particularly in regulated industries.
