Morgan Stanley (MS) points out that the AI industry is facing a long-term challenge of insufficient computing power supply relative to demand. Although corporate investment returns are improving and capital continues to pour in, power shortages, labor shortages, and political headwinds stemming from the midterm elections collectively constitute rigid bottlenecks that constrain computing power expansion. The firm expects computing power to remain scarce for the next several years.
Michelle Weaver, U.S. Thematic Research Strategist at Morgan Stanley, recently said in an interview that the market is still in a clear state of insufficient computing power supply relative to demand, and that computing power is evolving into a constrained resource. She noted that the pace of AI adoption by enterprises is accelerating, and the economic benefits brought by AI have become more clearly tangible. Among S&P 500 constituents, approximately 25% of companies can now quantify the actual returns from their AI investments, up notably from 14% a year ago. This shift means that corporate AI spending is gradually moving from the early experimentation and infrastructure-building phase into a new stage that can generate measurable business value.
At the funding level, AI infrastructure construction does not lack support. Weaver cited the partnership between Nvidia (NVDA) and Wall Street financial institutions as an example, noting that related plans are seeking up to $500 billion in financing for AI infrastructure, indicating that capital continues to flow on a large scale toward data centers and computing power construction. However, ample funding does not mean that infrastructure can expand at the same rapid pace.
Weaver believes that the two core factors currently constraining computing power supply are the shortage of labor required for data center construction and the shortage of electricity needed to power those data centers. AI data centers consume enormous amounts of electricity, while the construction cycles for new power supply, transmission networks, and related supporting facilities are relatively long. Even when considering innovative power solutions such as converting Bitcoin mining facilities and using fuel cells, Weaver estimates that there remains a supply gap of approximately 10% to 20% for the electricity required by AI infrastructure. This means that even if companies have sufficient funds to purchase AI chips and build data centers, they may still struggle to bring actual computing power online in a timely manner due to an inability to secure adequate electricity or a lack of construction personnel. She expects these structural constraints to keep computing power scarce and highly valuable for the next several years.
Overall, Morgan Stanley believes that corporate AI applications are gradually moving from the investment phase into a period of return realization, but the primary constraints facing the industry are no longer limited to capital or chips. Instead, they have extended to broader dimensions including electricity, labor, and permitting and political resistance. Until these bottlenecks are resolved, computing power supply may continue to lag behind rapidly growing demand.
