Zhu Shuai of CCID Research Institute: Global AI investment logic is undergoing new changes

Related topics top start Related topics top end Futures promotion begin Futures promotion end Miaopai begin Miaopai end Video player start Video player end Market chart begin Market chart end

Global AI competition is currently unfolding to a large extent around model capabilities, chip performance and the scale of computing power, but the relationship between massive capital expenditure and commercial returns is not simply one of direct proportionality. How to make infrastructure investment translate more effectively into economic benefits is becoming an important question that the global AI industry needs to answer.

Artificial intelligence (AI) is becoming a "strategic high ground" contested by global technology giants, and investment around chips, data centers and computing infrastructure continues to heat up. However, as the scale of investment continues to expand, the focus of market attention is undergoing an important shift: in the past, companies mainly competed over who could obtain advanced chips faster and build larger-scale data centers; today, investors are paying more attention to whether massive investment can be converted into sustained revenue growth, cash flow and productivity gains.

On October 29, Amazon released its third-quarter financial report. The report presented two sides: on the one hand, AWS (Amazon Web Services) revenue and the company's operating profit margin hit new highs; on the other hand, revenue, earnings per share and the fourth-quarter revenue guidance for the advertising business all fell short of expectations. The company's third-quarter capital expenditure rose sharply, AI costs surged, and a negative cash flow problem emerged. This was the main reason for investor disappointment and was directly reflected in Amazon's share price plunging by as much as 7%, reflecting market concerns about Amazon's massive spending in the AI field. This does not mean that the global AI investment boom is about to fade; on the contrary, the growth of cloud computing, intelligent programming and enterprise-level AI services is providing new demand support for infrastructure construction. But the relationship between massive capital expenditure and commercial returns is not simply one of direct proportionality. How to avoid equating the scale of computing power with industrial competitiveness, and how to make infrastructure investment translate more effectively into technological progress and economic benefits, is becoming an important question that the global AI industry needs to answer.

English text is being prepared.

But when the scale of investment reaches the level of hundreds of billions of dollars, simply expanding infrastructure is no longer enough to constitute a long-term advantage. Data center construction requires bearing ongoing costs such as chip procurement, energy supply, network connectivity and daily operations and maintenance, and equipment also faces pressure from technological iteration and renewal. Even if enterprises have sufficient computing power, if they lack stable paying customers, effective business scenarios and reasonable service pricing, it will be difficult to ensure sustainable returns on investment. Evaluating AI investment cannot rely only on the scale of capital expenditure, revenue growth or chip procurement volume; it also requires a comprehensive examination of business structure, funding sources, operational efficiency and the sustainability of future revenue.

At the same time, AI technology itself is also changing the economic logic of computing power investment. Model compression, inference optimization, specialized chips and software-hardware co-design technologies continue to develop, and are expected to reduce the computing cost per task. On the other hand, as AI applications move from a small number of professional scenarios to office work, manufacturing, scientific research and consumer services, lower barriers to use may also generate more demand. Therefore, the future computing power market may see both cost reductions brought by efficiency improvements and total demand growth brought by application expansion. What truly determines a company's competitiveness is whether it can establish a sustainable business model between technological efficiency and market scale.

Computing power investment expansion faces three tests

Corporate computing power investment mainly faces three tests. First, infrastructure construction needs to be tested by actual demand. Different types of AI tasks have very different requirements for computing power: frontier model training requires high-performance computing resources, while a large number of enterprise-level applications place more emphasis on inference cost, response speed, data security and deployment convenience. If companies blindly pursue the number of high-end chips and the theoretical scale of computing power without fully considering user needs and application structure, problems may arise in which resource allocation does not match actual business needs.

Second, heavy-asset expansion may increase companies' operating and financing pressure. AI infrastructure involves multiple links such as chips, servers, network equipment, cooling systems, data centers and energy security, with long construction cycles, large upfront investment and high operating costs. If the growth of paid demand from end applications falls short of expectations, or if market competition leads to continued declines in service prices, companies may face challenges such as insufficient utilization, narrowing profit margins and extended investment payback periods. For companies that rely on external financing and have high customer concentration, changes in the financing environment may further amplify risks.

Third, technological iteration may change the value of existing investment. As algorithm efficiency improves, model architectures evolve and specialized chips develop, the tasks that can be completed with the same scale of hardware resources may continue to increase. If companies carry out large-scale expansion based only on the current technology route and short-term demand forecasts, they may face uncertainty in the future regarding equipment utilization, renewal cycles and asset recovery. This does not mean that demand for computing power will decline as a result; rather, it shows that infrastructure investment must preserve flexibility for changes in technology routes and avoid simply equating short-term technological advantages with long-term competitive advantages.

Changes in the logic of global AI investment have provided important insights for China's development of artificial intelligence. China has a relatively complete manufacturing system, abundant industrial application scenarios and continuously expanding digital demand. It can turn these advantages into practical momentum for the development of the AI industry, and while maintaining necessary investment, pay more attention to the efficiency of resource allocation and application value.

First, adhere to demand-driven development and promote matching between computing power construction and industrial development. In accordance with the energy conditions, industrial foundations and actual needs of different regions, rationally lay out computing power infrastructure and take coordinated account of the different needs of training, inference and industry applications. For new projects, service targets, expected loads, energy costs, operation and maintenance capabilities and investment payback periods should be comprehensively assessed, and project results should be avoided being measured simply by investment amount, number of devices and theoretical computing power scale.

Second, strengthen software-hardware coordination and enhance the economic value of unit computing power. Competition in computing power is not only competition in chips, but also involves algorithms, models, compilation tools, network architecture and system scheduling. Support enterprises in improving the utilization efficiency of existing computing power resources and reducing the comprehensive cost of AI applications through model optimization, inference acceleration, heterogeneous computing and software-hardware co-design. For the vast number of small and medium-sized enterprises, being able to obtain stable, secure computing power services that fit their own businesses at reasonable prices is often more practically meaningful than directly building large computing power facilities.

Third, test technological investment with real applications and promote the deep integration of AI and the real economy. Around scenarios such as manufacturing R&D and design, production control, quality inspection, supply chain management and equipment operation and maintenance, promote AI from pilot demonstrations to large-scale applications. For enterprises, it is not only necessary to examine model performance and technological advancement, but also to assess application costs, frequency of use, the degree of improvement in production efficiency and security compliance costs. By establishing a quantifiable input-output evaluation mechanism, AI projects can truly solve specific problems in production and operations and form business models that can be replicated and promoted.

In addition, it is also necessary to improve differentiated investment and financing mechanisms and enhance the long-term resilience of industrial development. Basic research in artificial intelligence, breakthroughs in key technologies and the construction of common platforms are long-term undertakings and require stable support. In the future, the competitive advantage of the AI industry will depend not only on how much computing power can be built, but also on whether computing power can be transformed into new momentum for technological innovation, enterprise productivity and industrial upgrading with lower costs, higher efficiency and more reliable services. For China, the key to grasping changes in the logic of global AI investment lies in adhering to equal emphasis on technological innovation and industrial application, using actual demand to guide resource allocation, and using system efficiency to enhance industrial competitiveness, so as to form more sustainable advantages in the development of global artificial intelligence. (The author is a researcher at the CCID Research Institute) news_keyword_pub,stock, iframe below the main text iframe below the main text general statement general statement QR code at the end start QR code at the end start

QR code at the end start QR code at the end start editor name and work code editor name and work code