Morgan Stanley: The market is seriously underestimating the explosive power of AI, and the gap between computing power and electricity demand is expected to continue widening.

Gelonghui April 13 ─ Morgan Stanley has released a report stating that the market may be severely underestimating the actual force and depth of the AI revolution. Top large language models (LLMs) are experiencing a nonlinear leap in capabilities, and the explosive growth of AI is encountering systemic supply bottlenecks. The global gap in computing power and electricity demand is expected to continue expanding over the long term.
The report cites data showing that in the first quarter of this year, global weekly token usage surged 250% to 22.7 trillion, and some LLM providers have already set limits on user usage. Morgan Stanley forecasts that future growth in computing power demand will be three times the compound annual supply growth projected by NVIDIA, with computing power shortages persisting and becoming even more intense. With the launch of next-generation chips, AI computing costs will drop significantly, further fuelling demand, meaning chip manufacturers, optical communications, and data center equipment suppliers will see long-term structural benefits.
Additionally, energy is becoming a “time bomb” restricting AI development. The firm estimates that between 2025 and 2028, U.S. data centers will face a power shortage of around 55 GW. Even with alternative solutions such as natural gas, fuel cells, or nuclear power, the net deficit may still reach as high as 18% to 30% of total deployed capacity.