Morgan Stanley Bullish on AI Infrastructure: Pullback Presents Buying Opportunity, Power is the Next Battlefield

Recently, the global AI infrastructure sector has experienced a significant correction, with AI-related stocks averaging nearly a 7% decline over the month ending July 24, sparking market concerns about whether AI demand has peaked and questions about the return on massive capital expenditures (ROI).

However, Morgan Stanley, in its latest research report, presents a starkly optimistic outlook, arguing that the current downturn stems primarily from short-term investor position adjustments, not deteriorating fundamentals. In fact, the firm believes this presents an unusually attractive entry opportunity.

Core Thesis: Computing Demand Will Outstrip Supply for Years

The central view of Morgan Stanley’s research report is that AI infrastructure will evolve over time into a 'smart highway,' delivering substantial net benefits to the global economy.

While the market worries that companies may limit AI usage due to cost, data shows that the average number of tokens consumed per employee remains low, indicating vast room for future growth. The firm forecasts that global demand for AI computing power will significantly exceed supply for many years to come.

Moreover, improvements in AI capabilities are non-linear. Morgan Stanley points out that with the advancement of Recursive Self-Improvement (RSI) technology, frontier models will be able to assist in developing the next generation of systems, accelerating AI progress from 'human development speed' to 'machine iteration speed,' further driving up demand for computing power.

Jevons Paradox: Efficiency Gains Amplify Demand

Regarding the high efficiency and cost competitiveness recently demonstrated by large models in China (e.g., Kimi K3), Morgan Stanley argues this is not a threat but rather validates the 'Jevons Paradox': efficiency gains from technological progress lower costs, ultimately leading to total demand increasing rather than decreasing.

The analysis indicates that advancements in large models confirm that Scaling Laws remain effective. As more powerful models emerge, the global 'pie' of computing power will only grow larger.

Strong Support from Enterprises and Hyperscalers

The report cites a CIO survey from Q2 2026, showing that AI/ML has been the top priority for enterprise spending for multiple consecutive quarters. A high 76% of CIOs expect to deploy AI projects by the end of 2026, indicating that enterprise AI demand is accelerating into the practical implementation phase. Meanwhile, hyperscale cloud providers (Hyperscalers) are demonstrating strong investment confidence.

Morgan Stanley projects that capital expenditures by just the five largest U.S. tech giants will grow from nearly $800 billion this year to $1.2 trillion in 2027 and reach $1.4 trillion in 2028.

Power is the Next Battlefield

Morgan Stanley warns investors that the bottleneck for AI development is shifting from chips to infrastructure, particularly power supply. The firm estimates that U.S. data centers will face a 38GW power deficit by 2028, expanding to 122GW by 2030.

Therefore, the firm recommends investors focus on 'buying opportunities' in the following areas:

- AI Infrastructure Bottleneck Solutions: Fuel cell, natural gas turbine, and energy storage companies, as well as data center REITs with power assets. - Computing Power Manufacturing Ecosystem: Semiconductor manufacturers that continue to benefit from supply-demand imbalances. - Energy Security Assets: Companies supporting reliable power supply and storage. - Hyperscalers: Tech giants with scale advantages capable of generating attractive returns from AI spending.