AI Application Shift: Cloud Vendors Bridge the Silicon and Carbon Worlds

Author: Tide Research

The AI industry is undergoing a critical shift in narrative. For the past two years, the market has been pricing in the logic of "building power plants"—that is, computing infrastructure. Starting now, the market will begin pricing in the logic of "electrification"—that is, the commercialization of AI applications.

Cloud providers are the bridge connecting silicon-based computing to the carbon-based world, and the true explosion of AI applications has only just begun.

Over the past year and a half, the core narrative in AI investing has been "compute is king." From NVIDIA GPUs to optical modules, from data centers to liquid-cooled servers, upstream players in the supply chain have reaped substantial profits. Entering the second half of 2026, this landscape is beginning to shift.

The reversal signal comes from the large model's pricing strategy.

In May 2026, Doubao launched its paid version, and Kimi API released a new model with a significant price increase. Domestic large models have officially transitioned from "burning cash for free users" to "monetizing to realize commercial value."

Several sets of data can corroborate this turning point:

• Alibaba's Bailian MaaS platform has seen its token consumption increase sixfold over three months and is set to become Alibaba Cloud's highest-revenue product.
• Microsoft 365 Copilot paid seats exceed 30 million, with net new additions for the quarter more than doubling quarter-over-quarter.
• Google Cloud revenue grew 63% year-over-year, marking the strongest growth rate in recent years.
• Anthropic releases Claude Fable 5 and Mythos 5, marking another leap in model capabilities.

From late July to early August, Microsoft, Google, Amazon, and Meta sequentially released their earnings reports. All four reports conveyed the same message: cloud revenue growth is accelerating, capital expenditures are being revised upward, but market focus has shifted to returns.

Looking specifically at Q2 2026: The four major North American cloud providers continued to significantly increase their capital expenditures, with a combined full-year guidance of approximately $720–745 billion. However, even more notable than capital spending is the acceleration in revenue: Google Cloud +63%, Azure +40%, AWS +28%.

Microsoft’s actions are particularly noteworthy. Copilot’s business model has expanded from a simple "per-seat" pricing structure to a dual model of "per-seat + usage," indicating that AI applications are establishing sustainable revenue models.

Silicon-based remains unending, carbon-based is gradually rising—China Merchants Securities presented this title in its mid-year 2026 strategy report; "silicon-based" refers to the digital world composed of chips, computing power, and AI models; "carbon-based" refers to the physical world shaped by human activity, traditional industries, and physical consumption.

If the silicon-based world is the "power plant" and the carbon-based world is "every household," then cloud providers are the intermediary "power grid."

The cloud provides the foundational support for AI applications. Applications require elastic computing power, model services, storage, and networking—these cannot be solved simply by purchasing GPUs; they depend on the comprehensive infrastructure of a cloud platform.

More importantly, as AI transitions from "training" to "inference," the strategic importance of cloud platforms becomes even greater. Inference requires low latency, high concurrency, and elastic scalability—exactly the core strengths of the cloud.

Capital expenditure data further confirms this point. The combined 2026 capital expenditure guidance for the four major North American cloud providers totals $720–745 billion, while global capital spending by the eight largest CSPs is expected to exceed $710 billion, representing an annual increase of approximately 61%. Morgan Stanley has projected a forward-looking estimate of $1.4 trillion in capital expenditures driving a fourfold increase in computing power.

Among domestic cloud providers, Alibaba Cloud is the most aggressive in its布局. Alibaba has established a complete closed loop spanning "chips (T-Head)—cloud (Alibaba Cloud Intelligence)—models (Tongyi Lab)—and applications." The Zhenwu 810E general-purpose GPU has entered mass production and has been deployed in multiple ten-thousand-GPU clusters. Alibaba Cloud's AI revenue has grown by over 100%, and CEO Wu Yongming has explicitly stated that the Bailian MaaS platform will become Alibaba Cloud's largest revenue-generating product.

Major domestic cloud service providers have clearly stated that their total investment growth will exceed 20% by 2026, making capital expenditures in the hundreds of billions a certainty.

The current areas with the fastest commercial validation for AI applications include:

In light of the current market, three investment themes can be identified.

Main Theme 1: Cloud Service Providers — The Most Certain "Shovel Sellers"

The more prosperous AI applications become, the greater the computing power consumption by cloud service providers. In North America, this includes AWS, Azure, and GCP; in China, it includes Alibaba Cloud, Huawei Cloud, Tencent Cloud, and their respective industrial chains.

Core logic: Cloud revenue acceleration → valuation shifts from "resource-based pricing" to "value-based pricing" → long-term gross margin expansion for cloud providers.

Secondary Line Two: AI Application Platform, Seeking a "Killer App"

Focus on companies that have achieved a commercialized closed loop, particularly in areas such as AI programming, AI marketing, and multimodal content generation. Key screening criteria: AI revenue占比 > 10% and growth rate > 50%.

Core logic: Enhanced large model capabilities → Qualitative improvement in user experience → Surge in paid conversion rates → Performance realization.

Main Line 3: Hashrate Leasing and AI Infrastructure—The "Shadow Stocks" in a Booming Sector

The surge in AI applications will drive a sharp increase in demand for inference computing power, benefiting infrastructure segments such as computing power leasing simultaneously.

Core logic: Surge in token usage → Demand for inference computing power exceeds supply → Increased demand for computing power leasing.

In the first half of 2026, the AI sector experienced a "K-shaped divergence": AI infrastructure and hardware prices rose steadily before sharply retracing, while AI applications and software sectors lagged behind.

In August, several factors were at play simultaneously. The quarterly earnings reports from North American cloud providers signaled a combination of accelerating cloud revenue and upward revisions to capital expenditures, indicating that the business cycle for AI investment is closing. Large models are transitioning from free to paid services, and the industry is finally beginning to answer the question: "How does AI make money?"

New AI applications are being released in rapid succession, from Claude Fable 5 to Alibaba’s Qwen consumer apps, with continuous product iteration. On July 30, the Political Bureau meeting emphasized "deeply implementing the AI+ initiative," further strengthening policy support.

The core task facing investors for the second half of this year and potentially the coming years is to identify companies that can truly turn AI technology into revenue and profits.

[Disclaimer]

This report is produced by Chaoxiang Research and is based on publicly available information and institutional research reports. It is intended solely for professional investors and does not constitute any investment advice.