China's artificial intelligence model market is witnessing a period of rapid technological iteration coupled with explosive usage growth, propelling investments in cloud computing infrastructure into a fresh expansion phase.

Data from a recent monthly report on China's AI models indicates three parallel trends are unfolding: accelerated model upgrades, tightening compute demand, and a sharp uptick in cloud capital expenditure. Together, these trends paint a picture of China's AI industry swiftly transitioning from a competition focused on model development to one centered on commercial monetization.

Among the latest developments, cloud giants like Tencent and Alibaba, alongside independent AI labs such as DeepSeek and Zhipu AI, all launched new model versions in August. Some of these companies also raised their token pricing during this period. Concurrently, Tencent and Alibaba reported second-quarter capital expenditures of RMB 53 billion and RMB 68 billion, respectively. Their cloud revenue growth also accelerated notably, with Tencent's cloud revenue growing by just over 20% year-on-year, while Alibaba's surged by 45%.

Projections for China's top four cloud platforms suggest their combined capital expenditure will grow by 98% in 2026 and a further 44% in 2027. Over the past 30 days, market consensus for Tencent's and Alibaba's capital expenditure over the next 12 months has been revised upward by 25% and 29%, respectively, indicating that the market is repricing the upside potential within the cloud sector.

A Wave of Model Upgrades: Narrowing Gap with the US, Domestic Models Lead in Usage

August saw another dense wave of upgrades in China's AI model market. On the cloud vendor front, Tencent introduced its Hy4 preview model, and Alibaba launched the Qwen 3.8 Max and Flash versions. Among independent labs, Zhipu AI rolled out GLM5.3 Flash.

In terms of capability benchmarks, the performance gap between domestic models and global leaders continues to shrink. According to Artificial Analysis data, Anthropic's Claude Opus 5 tops the intelligence index with a score of 63, followed by Claude Fable 5 (62) and OpenAI's GPT-5.6 Sol (61). Among Chinese models, Moonshot AI's Kimi K3 ranks fifth globally with a score of 60, reaching 95% of Claude Opus 5's performance, while Zhipu AI's GLM-5.3 also scores 60, tying for sixth place. In the Agentic Index, GLM-5.3 shares the top spot with Claude Opus 5, both scoring 59, while Alibaba's Qwen 3.8 Max ranks fifth with a score of 58.

In the Code Arena WebDev benchmark focused on front-end development, Claude Opus 5 leads with 1691 points, followed closely by Kimi K3 (1674) and Qwen-3.8 Max (1669). Tencent's Hy4 preview ranks sixth with 1633 points.

Usage data further confirms the strong momentum of domestic models. OpenRouter figures show that for the month-to-date period ending August 17, DeepSeek V4 Flash led with 31.6 trillion tokens, Tencent's Hy3 followed with 26.2 trillion tokens, and Xiaomi's MiMo-V2.5 ranked third with 19.1 trillion tokens. On a vendor basis, DeepSeek led with a 22% share of weekly token usage, up 4.7 percentage points from early July. On the Vercel AI platform, DeepSeek's average monthly token usage share rose to 29.7% in August from 26.0% in July, with Anthropic in second place at 24.7%. Notably, despite DeepSeek's lead in usage volume, Anthropic still dominates the expenditure share at 64.8%, highlighting the significant pricing gap between US and Chinese models.

Diverging Pricing Dynamics: Tight Compute Prompts Select Price Hikes, Chinese Models Maintain Cost Advantage

In August, the LLM token expenditure index fell 26% month-on-month to $1.07 (from $1.45 in July), though it appears to be stabilizing. DeepSeek raised its token prices in August, but this is attributed to tight compute supply rather than a strategic shift away from its "high-performance, low-cost" approach, reflecting a short-term supply-demand imbalance rather than a change in direction. Meanwhile, the emergence of small, low-cost models like Zhipu AI's GLM5.3 Flash continues to offer more cost-effective options to the market.

The absolute pricing differential between US and Chinese models remains substantial. Claude Opus 5 is priced at $5 per million input tokens and $25 per million output tokens, while GPT-5.6 Sol is priced at $4 and $20. Among Chinese models, Kimi K3 commands the highest prices at $3 input and $15 output, Qwen-3.8 Max is priced at $2 and $6, and DeepSeek V4.0 is priced at just $0.7 and $2.

On a per-task cost basis, Chinese models also demonstrate a clear advantage. According to Artificial Analysis data, Claude Fable 5 has a cost of up to $3.14 per intelligence-index task, and Claude Opus 5 costs $2.34. In contrast, Qwen 3.8 Max is priced at just $0.91, Kimi K3 at $0.84, and GLM-5.3 at $0.68.

Capital Expenditure Acceleration: Cloud Infrastructure Enters a New Expansion Cycle

The surge in AI model usage is directly driving investment in cloud infrastructure. Leading Chinese internet platforms have been consistently increasing their AI infrastructure investments since 2024. The combined capital expenditure of the four major cloud platforms is projected to approach RMB 400 billion in 2025, reach RMB 726 billion in 2026, and surpass RMB 1 trillion in 2027.

Looking at individual companies, Alibaba's capital expenditure for fiscal 2026 is estimated at RMB 68 billion, ByteDance at RMB 350 billion, Tencent at RMB 210 billion, and Baidu at RMB 30 billion. Over the past 30 days, market consensus for Tencent's and Alibaba's 12-month forward capex has been revised upward by 25% and 29%, respectively, with depreciation expectations also revised higher. This indicates the market is incorporating higher investment intensity into its valuation models.

The acceleration in cloud revenue is already evident in second-quarter earnings: Tencent's cloud revenue grew by just over 20% year-on-year, while Alibaba's cloud revenue grew by 45%.

Cloud Market Outlook: AI-Driven Growth, MaaS Emerges as the Fastest-Expanding Segment

The size of China's cloud market is projected to grow from RMB 389 billion in 2025 to RMB 1.729 trillion by 2030, representing a compound annual growth rate (CAGR) of approximately 38%. AI-related cloud services (including MaaS and AI infrastructure) are expected to be the core growth driver. The share of AI cloud revenue is forecast to increase from 17% in 2025 to 66% by 2030.

Within this, AI MaaS (Model-as-a-Service) is viewed as the fastest-growing segment, expected to surge from RMB 21 billion in 2025 to RMB 676 billion by 2030, a CAGR of 81%. AI infrastructure cloud services (including GPU compute, training, and inference systems) are projected to grow from RMB 45 billion in 2025 to RMB 470 billion by 2030, a CAGR of 51%.

In terms of market structure, internet giants with mature cloud businesses, such as Alibaba, Tencent, and ByteDance, are expected to benefit most from rising enterprise AI spending. In contrast, independent AI labs like Zhipu AI and MiniMax, which have smaller revenue bases and stronger overseas market presence, are likely to achieve faster growth rates between 2026 and 2030.

Consumer AI Applications: Doubao Leads in User Scale, DeepSeek Improves Engagement

On the consumer front, ByteDance's Doubao reached 171.3 million weekly active users (DAU) during the week of August 3, accumulating 11.5 billion minutes of total usage time, maintaining its absolute leadership among Chinese AI chat applications. DeepSeek ranked second with 30.7 million DAU and 3.8 billion minutes of usage. Alibaba's Qwen ranked third with 28 million DAU, while Tencent's Yuanbao placed fourth with 8.7 million DAU.

In terms of user engagement, DeepSeek's average daily usage time per user saw a slight increase to 18 minutes during the week of August 3, with 8 sessions per day. In contrast, Doubao's average daily usage time and session frequency experienced a notable decline between June and July.

Looking ahead, several key events warrant attention. Alibaba may release a new model during the Apsara Conference scheduled for September 22-24. MiniMax is expected to launch its M3.1 and M3 Pro models between September and October. DeepSeek V5 and upgraded versions of ByteDance's Seed series are anticipated in the second half of 2026. Tencent's Hy4 official version and Hy5 preview are expected in the fourth quarter. Additionally, Zhipu AI and MiniMax will see their 12-month lock-up periods expire on January 7 and January 8, 2027, respectively, with 40% and 90% of their shares becoming eligible for trading, a development that investors should monitor closely.