U.S. data centres face widening power shortfall as AI demand rises
Rapid growth in artificial intelligence computing is intensifying pressure on U.S. electricity supply for data centres. Morgan Stanley estimates the gap between available power and projected AI chip demand reaches 30 to 40 percent by 2028, highlighting a major constraint on sector expansion.
Highlights
- Morgan Stanley estimates U.S. data centres face a power shortfall equivalent to the baseload demand of six New York Cities as AI chip output accelerates.
- AI chip production is projected to rise 40–50 percent annually, while power supply lags at 10–20 percent growth, driving a widening energy gap.
- The data centre electricity deficit could double to 107 GW by 2029, threatening technology-sector growth and demanding major U.S. infrastructure investment.
Morgan Stanley outlines scale of electricity gap
According to Financial Times, Morgan Stanley says U.S. data centres are running into a deepening electricity shortage as AI chip production expands faster than power capacity. The analysts estimate the shortfall is equivalent to the baseload demand of about six New York Cities, with New York City itself requiring roughly 5.5 GW to 6 GW of power.Elon Musk, speaking at the G20 Innovation Ministerial Summit, also points to mounting strain on supply. He says there is a power crisis and that the consensus is for a significant shortfall next year.
The report says AI chip output is expected to increase by 40 percent to 50 percent annually, while available power rises by only 10 percent to 20 percent. That mismatch is driving the projected gap between U.S. energy capacity and data centre demand through the next several years.
Energy constraints threaten technology buildout
Morgan Stanley says the imbalance could worsen without new approaches to power generation and grid support. The report warns the energy gap could grow to the equivalent of ten New York Cities if effective solutions are not put in place.The analysts also say gains in chip efficiency do not necessarily ease the problem, because lower-cost computing can spur even more demand. By 2029, the projected shortfall could double to 107 GW, underscoring the scale of investment needed in power infrastructure and generation.
For the U.S. technology and utilities sectors, the findings point to electricity availability becoming a central factor in data centre expansion. Without faster development of energy resources, the expected growth in AI processing capacity risks being limited by power supply rather than chip demand.
Our earlier report on Ontario’s Pickering Nuclear Generating Station refurbishment covered the start of construction to extend the life of four reactors and bolster long-term electricity supply. We noted the major contract awards, the projected economic and jobs impact, and that the upgraded plant is expected to deliver up to 2,200 MW—highlighting how large-scale generation projects are being positioned as a response to tightening power needs.