
AI Application Reversal: Silicon-Based Connects Carbon-Based, Cloud Providers Take the Lead
Author: TechFlow Research
The AI industry is undergoing a critical narrative shift. In the past two years, the market traded on the logic of "building power plants," i.e., compute infrastructure; from now on, the market will begin to trade on the logic of "appliance popularization," i.e., AI application commercialization.
Cloud providers are the bridge connecting silicon-based compute and the carbon-based world, and the true explosion of AI applications has only just begun.
AI Application Reversal
Over the past year and a half, the core narrative of AI investment was "compute is king." From NVIDIA GPUs to optical modules, from data centers to liquid-cooled servers, the upstream industry chain made huge profits. Entering the second half of 2026, the landscape began to loosen.
The turning signal comes from the pricing strategy of large models.
In May 2026, Doubao launched a paid version, and Kimi API released a new model and significantly increased prices. Domestic large models officially switched from "burning money for free to grab users" to "charging to realize commercial value."
Several sets of data can corroborate this inflection point:
- Alibaba Bailian MaaS Platform Token consumption increased 6-fold in 3 months, about to become Alibaba Cloud's largest revenue product
- Microsoft 365 Copilot paid seats exceeded 30 million, with quarterly net additions doubling quarter-over-quarter or more
- Google Cloud revenue grew 63% year-over-year, the most impressive growth rate in recent years
- Anthropic released Claude Fable 5 and Mythos 5, model capabilities leaped again
From late July to early August, Microsoft, Google, Amazon, and Meta released financial reports sequentially. The four reports conveyed the same signal: cloud revenue accelerated growth, capital expenditure continued to be revised upward, but market focus has shifted to the return side.
Specifically: In Q2 2026, capital expenditure of the four major North American cloud providers continued to grow highly, with full-year guidance totaling approximately $720-745 billion. But more notable than capital expenditure is the acceleration on the revenue side: Google Cloud +63%, Azure +40%, AWS +28%.
Microsoft's moves are particularly noteworthy. Copilot's business model expanded from simple "charging per seat" to dual charging of "seat + usage," meaning AI applications are running through a sustainable profit model.
Silicon Connecting Carbon: The Strategic Position of Cloud Providers
"Silicon is not yet over, carbon is gradually rising," Changjiang Securities systematically gave this title in its 2026 mid-year strategy. Silicon-based refers to the digital world composed of chips, compute, 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 the "power plant," and the carbon-based world is "thousands of households," then cloud providers are the "power grid" in between.
Cloud is the underlying support for AI application operation. Applications require elastic compute, model services, storage, and networking; these cannot be solved by simply buying GPUs directly and must rely on the complete infrastructure of cloud platforms.
More critically, when AI moves from "training" to "inference," the strategic position of cloud platforms becomes even stronger. Inference requires low latency, high concurrency, and elastic scaling, which are precisely the core advantages of cloud.
Capital expenditure data also confirms this. The 2026 capital expenditure guidance for the four major North American cloud providers totals $720-745 billion, and capital expenditure for the global top eight CSPs is expected to exceed $710 billion, a year-over-year increase of about 61%. Morgan Stanley gave the forward-looking judgment of "$1.4 trillion capital expenditure, 4x compute improvement."
Among domestic cloud providers, Alibaba Cloud is the most aggressive layouter. Alibaba has formed a complete closed loop of "Chip (Pingtouge) — Cloud (Alibaba Cloud Intelligence) — Model (Tongyi Lab) — Application." The Zhenwu 810E general-purpose GPU has entered mass production, and multiple ten-thousand-card cluster deployments have been achieved. Alibaba Cloud AI revenue achieved triple-digit growth, and CEO Eddie Wu explicitly stated that the Bailian MaaS platform will become Alibaba Cloud's largest revenue product.
Major domestic cloud service providers have clarified that the total investment growth rate for 2026 will exceed 20%, making hundred-billion-level capital expenditure a certainty.
Gold Rush Map for AI Applications
The fields where AI application commercialization verification is currently fastest include:

Combining the current market, three investment main lines can be categorized.
Main Line 1: Cloud Service Providers, the "Shovel Sellers" with the Highest Certainty
The more prosperous AI applications are, the greater the compute consumption of cloud service providers. In North America, look to AWS/Azure/GCP; domestically, look to Alibaba Cloud, Huawei Cloud, Tencent Cloud, and their industry chains.
Core Logic: Cloud revenue acceleration → Valuation shifts from "resource pricing" to "value pricing" → Cloud provider gross margin improves long-term.
Main Line 2: AI Application Platforms, Seeking "Killer" Products
Focus on companies that have already achieved a commercialization closed loop, especially in tracks such as AI programming, AI marketing, and multimodal content generation. Key screening criteria: AI revenue proportion > 10% and growth rate > 50%.
Core Logic: Large model capability improvement → Application experience qualitative change → Paid conversion rate leap → Performance realization.
Main Line 3: Compute Leasing and AI Infrastructure, Prosperous "Shadow Stocks"
The explosion of AI applications will drive a surge in inference compute demand, and infrastructure links such as compute leasing will benefit simultaneously.
Core Logic: Token call volume explosion → Inference compute supply falls short of demand → Compute leasing prosperity rises.
Opportunities Greater Than Risks
In the first half of 2026, the AI sector experienced a "K-shaped differentiation": compute hardware corrected significantly after continuous rise, while AI application and software sectors lagged relatively.
Entering August, several variables are acting simultaneously. North American cloud provider financial reports gave a combined signal of cloud revenue acceleration combined with capital expenditure upward revision; the commercial closed loop of AI investment is closing. Large models moving from free to charging, the industry is finally starting to answer "How does AI make money."
New AI applications are being released intensively, from Claude Fable 5 to Alibaba Tongyi C-end applications, product power is continuously iterating. On July 30, the Politburo meeting emphasized "deeply implementing the 'AI+' Action," and the policy side is also increasing efforts.
The core task for the second half of the year or even the next few years lies before investors: Find those companies that can truly turn AI technology into revenue and profit.
[Disclaimer]
This report is produced by TechFlow Research, content is organized based on public information and institutional research reports, for reference by professional investors only, does not constitute any investment advice.














