ROI CapEx
Search documents
西南证券:AI模型迭代聚焦工程能力 AI应用落地锚定高ROI场景
智通财经网· 2026-01-13 09:17
Core Insights - The report from Southwest Securities highlights that by 2025, overseas cloud providers will emphasize "cloud service shortages" and "expanding data centers based on demand signals," while increasingly focusing on the commercialization of AI applications. AI investment is transitioning from FOMO CapEx to ROI CapEx [1][2] Group 1: AI Investment Trends - Overseas tech giants are expected to see significant capital expenditure growth in 2024-2025, with increased investment from AI startups and an upward revision of future spending expectations. The industry is currently facing cash flow pressures, prompting tech firms to explore various data center construction methods and financing options [1] - In 2024, some overseas cloud providers indicate that the risk of under-investment in AI is far greater than the risk of over-investment, with AI investments accompanied by FOMO sentiment [2] Group 2: Data Center Efficiency - Data centers are facing power capacity limitations, leading cloud providers to emphasize maximizing tokens output efficiency per watt. This involves optimizing hardware components such as chips, storage, and communication, as well as software stacks and system architectures to enhance computational efficiency [3] - Cloud providers are increasingly focusing on the versatility and flexibility of data center construction to accommodate various generations of GPUs and electrical components, allowing for a flexible switch between training and inference workloads [3] Group 3: AI Model Development - The engineering capabilities of large AI models are continuously improving, with a growing demand for commercializing AI products. Future iterations of AI models will focus on long text, multimodal capabilities, logical reasoning, and tool usage [4] - As the Scaling Law extends from pre-training to reinforcement learning and continuous learning, the requirements for training datasets will evolve, leading to differentiated model capabilities and the emergence of various AI use cases [4] Group 4: Cloud Business Growth - By 2025, overseas AI cloud services are expected to enter a "super large orders + long-term infrastructure" phase, with cloud order amounts ranging from billions to hundreds of billions. The remaining contract amounts for overseas cloud providers are also experiencing rapid growth, indicating a potential acceleration in AI cloud service expansion [5] - The revenue growth of cloud businesses will heavily depend on the pace of capacity rollout, with expectations for accelerated growth as computational capacity is gradually released by 2026 [5]