344
Productivity & Workflow355
Automation & Workflow225
Software Development251
Marketing & Growth192
AI Infrastructure & MLOps175
Writing & Content Creation203
Data & Analytics142
Photography & Imaging156
Design & Creative170
Customer Support133
Sales & Outreach125
Voice & Speech135
Education & Learning131
Operations & Admin87
Morgan Stanley estimates US data centers could lack enough electricity as AI chips drive demand, with a gap compared to several New York Cities of power.
In short: Analysts warn that US data centers may not have enough electricity to run the AI computers being built over the next few years.
US data centers, which are large buildings full of computers that run online services, are expected to need much more electricity because of AI. Morgan Stanley analysts updated their estimates and now see a large gap between the power the US grid can supply and the power that new data center equipment is likely to demand.
To make the numbers easier to picture, the analysts compared the shortfall to New York City. They said New York City needs about 5.5 to 6 gigawatts of steady power (a gigawatt is a big unit of electricity, roughly the output of a large power plant). Based on their mid-range estimate, the US could be short by the equivalent of about six New York Cities of power by 2028. Without certain fixes, the gap could look more like 10 New York Cities.
One reason is that newer AI server setups can draw a lot more power per rack, and they produce much more heat. CoreWeave has said a typical AI server rack can put out as much heat as 10 to 25 ovens running at the same time. Even if chips become more efficient, Morgan Stanley expects data center operators to use that efficiency to pack in more computing, not to lower electricity use. This is similar to buying a more fuel-efficient car and then driving more because it is cheaper per mile.
Morgan Stanley points to possible stopgap supplies like fuel cells and placing data centers near nuclear plants, but it still expects a remaining shortfall that would need new gas turbines or off-grid generators. If the power does not show up in time, the rollout of new AI services could slow.
Source: Financial Times