China AI Model Monitoring: Accelerated Upgrades and Soaring Usage Signal a New Upward Cycle for Cloud Computing

China's AI model market is undergoing a period of intensive technological iteration and explosive growth in usage, which is driving cloud computing infrastructure investment into a new expansion cycle.

According to the latest monthly monitoring report on Chinese AI models released by Bank of America Merrill Lynch, three main trends emerged simultaneously in August: rapid model upgrades, tightening computing power demand, and a significant increase in cloud capital expenditure. These trends collectively outline the accelerated leap of China's AI industry from model competition to commercial monetization.

In terms of the latest developments, cloud giants such as Tencent and Alibaba, along with independent AI labs like DeepSeek and Zhipu, all launched new versions of their models in August, with some vendors simultaneously raising token prices. Meanwhile, Tencent and Alibaba's capital expenditures in the second quarter reached RMB 53 billion and RMB 68 billion respectively, and their cloud revenue growth also accelerated significantly—Tencent's year-on-year growth was just over 20%, while Alibaba's reached 45%.

Bank of America Merrill Lynch projects that the combined capital expenditures of China's four leading cloud platforms will increase by 98% and 44% year-on-year in 2026 and 2027, respectively. In the past 30 days, market consensus forecasts for Tencent and Alibaba's capital expenditures over the next 12 months have been revised upwards by 25% and 29%, respectively, indicating that the upside potential of cloud services is being repriced by the market.

Model Upgrade Wave: Gap between China and the US Narrows, Domestic Models Lead in Usage

In August, the Chinese AI model market saw another round of intensive upgrades. On the cloud platform, Tencent launched Hy4 preview, and Alibaba released Qwen 3.8 Max and a Flash version; on the independent platform, Zhipu launched GLM5.3 Flash.

In terms of capability assessment, the gap between domestic models and the world's top levels continues to narrow. According to Artificial Analysis data, Anthropic's Claude Opus 5 ranks first in the intelligence index with a score of 63, followed by Claude Fable 5 (62 points) and OpenAI's GPT-5.6 Sol (61 points). Among Chinese models, Moonshot AI's Kimi K3 ranks fifth globally with a score of 60, reaching 95% of Claude Opus 5's score, while Zhipu GLM-5.3 also ranks sixth with a score of 60. In terms of agentic index, GLM-5.3 and Claude Opus 5 are tied for first place with a score of 59; Alibaba's Qwen 3.8 Max ranks fifth with a score of 58.

In the front-end web development programming benchmark (Code Arena WebDev), Claude Opus 5 ranked first with 1691 points, followed by Kimi K3 (1674 points) and Qwen-3.8 Max (1669 points), while Tencent Hy4 preview ranked sixth with 1633 points.

Usage data also confirms the strong momentum of domestic models. OpenRouter data shows that as of August 17th, DeepSeek V4 Flash ranked first with 31.6 trillion tokens, followed by Tencent Hy3 with 26.2 trillion tokens, and Xiaomi MiMo-V2.5 ranked third with 19.1 trillion tokens. According to vendor statistics, DeepSeek leads with a 22% weekly token usage share, an increase of 4.7 percentage points compared to the same period in July.

On the Vercel AI platform, DeepSeek's monthly token usage share reached 29.7%, a further increase from 26.0% in July, while Anthropic ranked second with 24.7%. It's worth noting that despite DeepSeek's lead in usage volume, Anthropic still dominates in spending share with 64.8%, reflecting a significant difference in pricing between the Chinese and American models.

Pricing divergence: Tight computing power drives price increases in some areas, while Chinese models maintain their cost-effectiveness advantage.

In August, the LLM token spending index fell 26% month-over-month to $1.07 (from $1.45 in July), but has since stabilized.

Bank of America Merrill Lynch points out that DeepSeek raised its token price in August, but the bank attributes this to tight computing power supply rather than a strategic shift—that is, it's not abandoning the "high-performance, low-cost" approach, but rather a short-term supply-demand imbalance. Meanwhile, smaller, lower-priced models such as GLM5.3 Flash from Zhipu continue to emerge, providing the market with more cost-effective options.

In terms of absolute pricing, the price difference between Chinese and American models remains significant. Claude Opus 5's input/output token prices are $5 and $25 per million tokens, respectively, while GPT-5.6 Sol's are $4 and $20. Among Chinese models, Kimi K3 has the highest pricing, with inputs and outputs priced at $3 and $15, respectively; Qwen-3.8 Max is $2 and $6, while DeepSeek V4.0 is only $0.7 and $2.

In terms of unit task cost, the Chinese model also has a clear advantage. According to Artificial Analysis data, the cost per intelligent index task for Claude Fable 5 is as high as $3.14, for Claude Opus 5 it is $2.34, while for Qwen 3.8 Max it is only $0.91, for Kimi K3 it is $0.84, and for GLM-5.3 it is $0.68.

Accelerated Capital Expenditure: Cloud Infrastructure Enters a New Expansion Cycle

The explosive growth in the use of AI models is directly driving investment in cloud infrastructure.

According to data from Bank of America Merrill Lynch, China's leading internet platforms have been continuously increasing their investment in AI infrastructure since 2024. The combined capital expenditure of the four major cloud platforms approached RMB 400 billion in 2025, and is expected to reach RMB 726 billion in 2026, and further exceed RMB 1 trillion in 2027.

Specifically, for individual companies, Alibaba's capital expenditure for fiscal year 2026 is projected 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 estimates for Tencent and Alibaba's capital expenditure over the next 12 months have been revised upwards by 25% and 29%, respectively, with depreciation expectations also being revised upwards, indicating that the market is incorporating higher investment intensity into its valuation system.

In terms of cloud revenue, the accelerating trend has been confirmed in the second-quarter financial report: Tencent Cloud's revenue growth rate reached just over 20% year-on-year, and Alibaba Cloud's revenue growth rate was 45% year-on-year.

Cloud Market Outlook: AI-Driven Growth, MaaS Becomes the Fastest Expanding Track

Bank of America Merrill Lynch predicts that China's cloud market will grow from RMB 389 billion in 2025 to RMB 1.729 trillion in 2030, with a compound annual growth rate of approximately 38%, of which AI-related cloud services (MaaS and AI infrastructure) will be the core driver.

The AI cloud market share is projected to increase from 17% in 2025 to 66% in 2030. AI MaaS (Model as a Service) is considered the fastest-growing segment, expected to grow from RMB 21 billion in 2025 to RMB 676 billion in 2030, representing a CAGR of 81%. AI infrastructure cloud (including GPU computing power, training, and inference systems) is projected to grow from RMB 45 billion in 2025 to RMB 470 billion in 2030, with a CAGR of 51%.

In terms of market structure, Bank of America Merrill Lynch expects internet giants with mature cloud businesses, such as Alibaba, Tencent, and ByteDance, to benefit more from the growth in AI enterprise service spending; while independent AI labs such as Zhipu and MiniMax, with their smaller revenue base and stronger overseas market presence, are expected to achieve faster growth between 2026 and 2030.

AI Applications: Doubao continues to lead in user scale, DeepSeek sees increased user engagement.

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

In terms of user stickiness, DeepSeek's average daily usage time per user increased slightly to 18 minutes and 8 sessions per day in the week of August 3, while Doubao's average daily usage time per user and session frequency declined significantly between June and July.

Looking ahead, Bank of America Merrill Lynch has identified several key events to watch: Alibaba may release a new model during the Yunqi Conference from September 22nd to 24th; MiniMax is expected to launch the M3.1 and M3 Pro in September or October; DeepSeek V5 and upgraded versions of ByteDance's Seed series are expected to debut in the second half of 2026; and Tencent's Hy4 official version and Hy5 preview version are expected to launch in the fourth quarter. Furthermore, Zhipu and MiniMax will have their 12-month lock-up periods expire on January 7th and 8th, 2027, respectively, at which time 40% and 90% of their shares will be released, which investors should pay attention to.

~~~~~~~~~~~~~~~~~~~~~~~

The above content is from Zhuifeng Trading Platform .

For more detailed analysis, including real-time updates and firsthand research, please join the [ Trading Channel Annual Membership ].

Risk Warning and DisclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.