China Great Wall: Continuous leaps in model capabilities are expected to drive the prosperity of the AI application ecosystem. The cloud computing industry is entering a second phase of acceleration.

China Great Wall released a research report stating that with the improvement of model capabilities and the rapid decline in AI usage costs, the bank believes that the prosperity of the cloud computing and AI application industries is on an upward trend. Cloud computing: The Jevons Paradox will drive total cloud demand into an explosive growth phase, and cloud computing is expected to maintain a trend of rising volume and price; AI applications: Enterprise-level AI applications and C-end Agent intelligent agents are in an accelerating implementation phase, among which enterprise AI budget release has gradually moved from pilot demonstration to large-scale implementation. China Great Wall's main viewpoints are as follows: The capabilities of domestic open-source models are rapidly catching up with overseas closed-source models, which is expected to accelerate the implementation of AI on the enterprise side Since June, many leading domestic model manufacturers have intensively released new-generation large models, showing trends of larger parameter scales, stronger Agent capabilities, more complete open-source ecosystems, and deeper multimodal applications. As domestic model manufacturers accelerate model iteration, the capabilities of domestic open-source models are rapidly improving, and the gap with leading overseas closed-source models (such as Anthropic, OpenAI, etc.) continues to narrow. Compared with closed-source models, open-source models (1) have higher cost-effectiveness; (2) can optimize models at a deeper level, promoting autonomous and controllable model behavior; (3) can be deployed on public clouds, private clouds, locally, at the edge, and in hybrid deployments, giving customers multiple choices. The bank believes that as the capabilities of open-source models catch up and they maintain high cost-effectiveness, the central token price of top models is expected to continue to move downward, thereby promoting high-quality foundation models from usable to easy to use, accelerating enterprise-level AI deployment and commercialization, and driving the demand side to expand from frontier model laboratories to multiple vertical fields such as industry, life sciences, finance, and the public sector. Cloud computing: The AI cloud computing industry in North America and China has entered a phase of rising volume and price The 26Q2 results of North American/domestic CSP and NeoCloud companies have all been released. In 26Q2, the year-over-year revenue growth rates of Azure/AWS/Google/Coreweave/Nebius/Alibaba Cloud businesses were 43%/37%/82%/112%/454%/45%, respectively, all maintaining an accelerating growth trend. The bank believes that the cloud computing industry has entered a phase of rising volume and price, mainly because (1) the capabilities of open-source models are rapidly improving, and cloud vendors can provide customers with corresponding models according to their requirements for quality, latency, cost, etc. The model tiering strategy drives industry discourse power to shift from leading model manufacturers to downstream CSPs; (2) the actual cost of AI usage (cost/Mtok) has dropped significantly. After achieving economic benefits on the cost side, the Jevons Paradox will drive total cloud demand into an explosive growth phase; (3) the parameter scale of open-source models continues to rise. For example, the K3 model released by KIMI on July 17 has reached 2.8 trillion parameters, and the resource threshold for a single deployment has simultaneously increased. It is uneconomical for enterprises to build their own computing power, so they are more inclined to rent cloud vendor infrastructure, driving a resonance in cloud vendors' IaaS+MaaS revenue. In 2026 H1, major domestic and international cloud vendors successively raised computing service prices, and price wars became a thing of the past. The cloud business revenue growth rates of North American cloud vendors, NeoCloud, and domestic cloud vendors continue to accelerate, and the logic of rising volume and price has been verified in financial statements. AI applications: After Vibe Coding, pay attention to opportunities on the application side such as AI4S, FDE, and AI office Referring to the mobile internet era, AI applications will become direct beneficiaries of falling token prices and improving model capabilities. In 26H1, downstream monetization scenarios were mainly concentrated in Vibe Coding. As model capabilities improve, AI is expected to empower more downstream application scenarios: (1) AI4S: AI is used to accelerate the process of scientific research and discovery. It has already been implemented in many fields such as life sciences, physics, and materials science, creating a complete data closed loop of "AI prediction - automated validation - data feedback"; (2) FDE: The implementation of models in enterprise business needs to overcome difficulties such as system integration, permission sorting, process transformation, compliance auditing, and organizational adaptation. FDE accelerates the implementation of models in enterprise business processes. Facing customers, FDE transforms business experience and process rules scattered in employees' minds into knowledge bases, workflows, and application capabilities that AI can call; facing the platform, FDE precipitates industry scenarios, system interfaces, testing methods, and failure cases into reusable product components, continuously iterating products, thereby reducing the delivery cost and cycle of subsequent FDE projects; (3) AI office: Enterprises and employees have strong willingness and motivation to pay for efficiency. AI improves work efficiency, and users have a relatively high willingness to pay; in addition, the bank believes that high-frequency calls of AI in office scenarios can continuously obtain enterprise private domain data (such as approval flows and documents) for fine-tuning models, forming a positive data flywheel cycle in which the more users use it, the more the model understands the enterprise, and the more users cannot do without it. Related targets Cloud computing/computing power leasing: Alibaba, KINGSOFT CLOUD, GBA AI COMP, Jiangsu Lettall Electronic, Fengzhushou, Nexwise Intelligence China; AI applications: TENCENT, Kuaishou, MARKETINGFORCE, DEEPZERO, XUNCE, Zhongke Wenge, DC HOLDINGS, MOBVISTA, HAIZHI TECH GP, VOBILE GROUP, Focus Technology, COL Global Co., Ltd., BlueFocus Intelligent Communications Group, Giant Network Group, Kingnet Network. Risk warnings: AI large model iteration falling short of expectations; AI application commercialization falling short of expectations; intensifying industry competition; impact of regulatory policies.