As enterprise artificial intelligence applications deepen, a growing number of companies are beginning to see quantifiable returns from their AI investments. However, Morgan Stanley believes that insufficient compute supply is becoming a key bottleneck limiting further expansion of the AI industry, and constraints related to power, labor, and politics could keep this issue in place for several years.

Morgan Stanley US thematic research strategist Michelle Weaver said in an interview Wednesday that the market is still in a clear state of compute supply falling short of demand. "Compute is becoming a constrained resource," and bottlenecks in power, politics, and labor will continue to limit supply expansion in the coming years.

Weaver noted that the pace of enterprise AI adoption is accelerating, and the economic benefits of AI have become more apparent. Among S&P 500 index components, roughly 25% of companies can now quantify the actual returns from their AI investments, a significant increase from 14% one year ago. This suggests that enterprise AI spending is gradually shifting from early-stage experimentation and infrastructure building towards a phase that generates measurable business value.

Meanwhile, AI infrastructure construction is not lacking financial support. Weaver cited NVIDIA Corporation (NVDA.US) as an example, where its collaboration with Wall Street financial institutions to raise funds has sought up to $500 billion in financing for AI infrastructure, showing that capital is still flowing massively into data centers and compute construction.

However, ample capital does not mean that AI infrastructure can expand rapidly in sync. Weaver believes the two core factors currently limiting compute supply are a shortage of labor for data center construction and a shortage of electricity to power these data centers. AI data centers consume enormous amounts of electricity, and building new power supply, transmission networks, and related infrastructure takes a long time. Even when considering innovative solutions like converting Bitcoin mining sites or using fuel cells, Weaver estimates there is still a supply gap of about 10% to 20% for the electricity needed by AI infrastructure. This means that even if companies have enough funds to buy AI chips and build data centers, they may not be able to form actual compute capacity in time due to a lack of sufficient power or construction personnel.

Weaver expects these structural constraints will keep compute scarce and high-value for years to come. Beyond power and labor, political factors are also becoming a new obstacle to AI data center expansion. Weaver said that with the US midterm elections approaching, opposition to data center construction is rising in some regions. Local residents are primarily concerned that large-scale data centers could push up electricity bills, increase water demand, and impact water quality, air quality, and the local environment. Although data center operators can adjust their construction and energy plans to some extent to address consumer concerns about electricity costs and environmental issues, Weaver expects political controversy over data center construction to intensify as the midterm elections enter their final stages.

Overall, Morgan Stanley believes that enterprise AI applications are gradually moving from an investment phase to a phase of generating actual returns. However, the main constraints facing the AI industry today are no longer just capital or chips, but have extended to power, labor, and political resistance in data center construction approvals. Until these bottlenecks are resolved, compute supply may continue to lag behind rapidly growing demand.