Author: Chao Xiang Research

The AI industry is undergoing a critical narrative shift. For the past two years, the market has been trading on the logic of "building power plants," that is, computing infrastructure; starting now, the market will begin to trade on the logic of "electrical appliance popularization," that is, the commercialization of AI applications.

Cloud vendors are the bridge connecting silicon-based computing power with the carbon-based world, and the true explosion of AI applications has just begun.

AI Application Reversal

In the past one and a half years, the core narrative of AI investment has been "computing power is king." From Nvidia GPUs to optical modules, from data centers to liquid-cooled servers, the upstream of the industry chain has reaped enormous profits. As we enter the second half of 2026, the landscape is beginning to loosen.

The turning signal comes from the pricing strategy of large models.

In May 2026, Doubao launched a paid version, Kimi API released new models and significantly raised prices. Domestic large models officially shifted from "burning money for free to attract users" to "charging to realize commercial value."

Several sets of data can corroborate this turning point:

• Ali Bailian MaaS platform token consumption increased sixfold in three months, soon to become Alibaba Cloud's largest revenue product
• Microsoft 365 Copilot paid seats surpassed 30 million, with quarterly net additions more than doubling sequentially
• Google Cloud revenue grew 63% year on year, the brightest growth rate in recent years
• Anthropic released Claude Fable 5 and Mythos 5, with model capabilities leapfrogging once again

From the end of July to early August, Microsoft, Google, Amazon, and Meta successively released financial reports. The four reports conveyed the same signal: cloud revenue is accelerating growth, capital expenditures continue to be revised upwards, but the market's focus has shifted to the returns.

Specifically, in Q2 2026, the capital expenditures of the four major North American cloud vendors continued to grow significantly, with a full-year guidance totaling approximately $720 billion to $745 billion. However, more noticeable than capital expenditures is the acceleration on the revenue side: Google Cloud +63%, Azure +40%, AWS +28%.

Microsoft's actions are particularly worth noting. The Copilot business model has expanded from simply "charging by seat" to a dual charging model of "seat + usage," indicating that AI applications are establishing a sustainable profit model.

Silicon Connects Carbon: The Strategic Position of Cloud Vendors

Silicon-based is unceasing, carbon-based is gradually rising, and Changjiang Securities systematically provided this title in the mid-2026 strategy: silicon-based refers to the digital world composed of chips, computing power, and AI models; carbon-based refers to the physical world composed of human activities, traditional industries, and physical consumption.

If the silicon-based world is "power plants," the carbon-based world is "thousands of households," while cloud vendors are the intermediary "power grid."

The cloud is the underlying support for the operation of AI applications. Applications require elastic computing power, model services, storage, and networks, which cannot simply be resolved by purchasing GPUs directly; they must rely on the complete infrastructure of cloud platforms.

More critically, as AI moves from "training" to "inference," the strategic position of cloud platforms has become even stronger. Inference requires low latency, high concurrency, and elastic scaling, which are precisely the core advantages of the cloud.

Capital expenditure data also corroborates this point. The four major North American cloud vendors have a capital expenditure guidance of $720 billion to $745 billion in 2026, while the capital expenditures of the eight major global CSPs are expected to exceed $710 billion, with a year-on-year increase of about 61%. Morgan Stanley provided a forward-looking estimate of "$1.4 trillion in capital expenditures, with a fourfold increase in computing power."

Among domestic cloud vendors, Alibaba Cloud is the most aggressive player. Alibaba has formed a complete closed loop of "chips (Pingtouge) - cloud (Alibaba Cloud Intelligence) - models (Tongyi Laboratory) - applications." The Zhenwu 810E general-purpose GPU has entered large-scale production, achieving multiple ten-thousand-card cluster deployments. Alibaba Cloud's AI revenue has achieved triple-digit growth, and CEO Wu Yongming has explicitly stated that the Bailian MaaS platform will become Alibaba Cloud's largest revenue product.

Domestic major cloud service vendors have clearly stated that the total investment growth in 2026 will exceed 20%, and the trillion-class capital expenditure has become a foregone conclusion.

Gold Mining Map of AI Applications

The current fields with the fastest commercialization verification of AI applications include:

Combining the current market, three investment main lines can be categorized.

Main Line One: Cloud Service Providers, the most certain "shovel sellers"

The more prosperous AI applications become, the greater the computing power consumption of cloud service providers. In North America, look at AWS/Azure/GCP; in China, look at Alibaba Cloud, Huawei Cloud, Tencent Cloud, and their industry chain.

Core logic: Cloud revenue acceleration → valuation shifting from "resource pricing" to "value pricing" → long-term increase in the gross margin of cloud vendors.

Main Line Two: AI Application Platforms, seeking "killer" products

Focus on companies that have realized a commercial closed loop, especially in AI programming, AI marketing, multimodal content generation, and other tracks. Key screening criteria: AI revenue ratio > 10% and growth rate > 50%.

Core logic: Improvement in large model capabilities → qualitative change in application experience → leap in conversion rates → performance realization.

Main Line Three: Computing Power Leasing and AI Infrastructure, prosperous "shadow stocks"

The explosion of AI applications will drive a surge in demand for inference computing power, benefiting segments like computing power leasing.

Core logic: Token call volume explosion → shortage of inference computing power supply → prosperity of computing power leasing on the rise.

Opportunities Greater Than Risks

In the first half of 2026, the AI sector saw a "K-shaped differentiation": after the continuous rise of computing hardware, there was a significant correction, while AI applications and software sectors lagged behind.

As we enter August, several variables are acting simultaneously. The financial reports of North American cloud vendors provided a combination signal of accelerating cloud revenue alongside upward revisions of capital expenditures, suggesting the commercial closed loop of AI investment is closing. Large models transitioned from free to paid, and the industry is finally starting to answer "how does AI make money."

New AI application products are being densely released, from Claude Fable 5 to Alibaba's Qianwen C-end application, with product capabilities continuing to iterate. The Politburo meeting on July 30 emphasized "in-depth implementation of artificial intelligence + actions," and the policy side is also ramping up.

The core task in the second half of the year and even for the next few years lies ahead of investors: find those companies that can truly turn AI technology into revenue and profit.

[Disclaimer]

This report is produced by Chao Xiang Research, based on publicly available information and institutional research reports, provided for professional investors' reference only and does not constitute any investment advice.