Damo Research determines the inflection point of the AI market: from computing power construction to industrial proliferation, barbell strategy has become the main line of configuration

The Zhitong Finance App learned that Morgan Stanley released the research report “Decoding the Pace of AI Transformation: From Computing Power Construction to Industrial Dissemination”, which proposed a core judgment based on the sixth round AI stock mapping database covering about 3,600 individual stocks around the world: AI investment has left the stage of simply betting on upstream computing power empowerment targets and has entered a new era of “barbell” dual-line layout. It is necessary not only to seize the structural opportunities brought about by computing power bottlenecks, but also to increase the allocation weight of AI application companies.

Morgan Stanley says its core research theme in 2026 is technology diffusion. In this round of the industrial cycle in the transformation process, it is expected to present a barbell pattern of excessive earnings, empowering enterprises with core infrastructure, and running out of the market at the same time as early software vendors and AI application adopters.

The deductive rule of the technology cycle is that semiconductors usually reap excess profits first, then the infrastructure sector, then software and services. According to the bank's judgment, the investment logic of the AI industry is entering a new stage: from a high concentration of transactions in the past to a broader barbell opportunity pool, covering selected enabling targets and emerging AI application adopters.

Since the start of this AI cycle, the semiconductor and infrastructure sector has recorded excess earnings of 500-700 points compared to the S&P 500 index. We believe that the time has come to increase configuration exposure to early software-enabled companies and AI application adopters.

At the same time, however, the wave of large-scale capital expenditure construction of AI, combined with key bottlenecks such as electricity, will lengthen the upward window for some high-quality targets. Therefore, the current computer cycle will not replicate the simple linear rotation of market-leading sectors in the previous cycle.

According to the bank, the current AI cycle is special, and manpower, electricity, and policy supervision form the three core constraints. The report estimates that from 2026 to 2028, global data centers will have a power gap of 57 GW, compounded by actual barriers such as local approvals and labor shortages. The pace of computing power supply expansion will be significantly suppressed. The shortage of computing power will continue for many years, and there will be no simple and clear sector handover rotation in history. The capital expenditure growth rate of hyperscale cloud vendors is expected to fall back from 93% in 2026 to 14% in 2028. Hardware capital expenditure boom is expected to be gradually priced by the market, but some bottleneck racetrack conditions will continue.

According to the data, AI-enabled companies (upstream suppliers of chips, computing power, hardware, etc.) expect EPS to double in the past two years and still maintain strong profit momentum. At the same time, companies that have adopted AI, that is, companies in various industries that apply AI technology to their own business and achieve cost reduction and efficiency, expect EPS to increase by about 70% over the next 12 months over the next 2 years, and the inflection point of profit has already been shown.

The market unanimously expects that the EBIT profit margin of enterprises with high real influence of AI will expand by 460 basis points from 2025 to 2026, almost double the MSCI ACWI index. However, the current market has not fully priced the long-term dividends brought about by AI production efficiency, and there is room for improvement in long-term profit forecasts. At the valuation level, the forward price-earnings ratio of adopted companies with high AI influence has fallen back to 18 times, compared to 22 times that of empowered companies. The risk-return ratio has improved markedly after the valuation has been digested.

The bank's proposed barbell investment strategy is split into two ends. One side continues to stick to bottleneck assets: Prioritize the layout of electricity-related circuits, including existing power plant service providers, energy storage, power grid equipment, and new energy; at the same time, select individual stocks within semiconductors and infrastructure to focus on bottlenecks in the second-tier industrial chain other than electricity.

On the other side, expand new configurations: add early software to empower enterprises, prioritize infrastructure software, network security, and select some application software; the layout focuses on two major indicators. One is the actual influence of AI on the company's business, and the other is the pricing power of the enterprise. Only companies that can preserve the dividends of AI cost reduction can truly achieve profit redemption. IT services are a beneficiary sector in the late cycle, and are not yet suitable for key layout at this stage.

From a global perspective, AI interpretation paths in various markets are clearly differentiated. AI dividends in the North American market spread from hardware to software and physical industries; revenue from the Asia-Pacific market is still concentrated in the upstream supply chain, but more and more corporate financial reports reveal revenue growth brought about by AI; AI opportunities in the European market are more concentrated on the application implementation of traditional industries.

The report specifically emphasizes that investment must not only chase the “AI concept”; the real influence of AI on corporate investment logic is the key to victory or loss. AI is the target of core logic, and there are huge excess benefits compared to targets that have only a secondary impact; conversely, companies that pose a threat to their core business will significantly outperform their peers that have only experienced moderate impacts.

At the same time, there is still a reverse risk in the market. If demand for computing power continues to exceed expectations and human power bottlenecks continue, the semiconductor and infrastructure sector boom may continue to far exceed market benchmark expectations. Wall Street has underestimated the development space of the technology industry many times in history. Overall, the AI market has gone from simply being hyped up on the theme of “heaping computing power” to a new stage of verifying commercial ROI, taking into account upstream bottleneck assets and application targets for implementation, and has become a configuration idea more suited to the current situation.