Artificial intelligence runs on data centers, and data centers run on electricity. Training and running large AI models needs enormous computing power, which draws vast amounts of electricity for the chips and for cooling them.

Morgan Stanley estimates that US data-center developers face a 34 percent net power shortfall through 2028, equivalent to 32 gigawatts, even after accounting for workarounds such as behind-the-meter generation and fuel cells. Grid connections, permits and labor are all tightening at once.

That bottleneck is changing what matters in the chip business. Morgan Stanley says the choke point is no longer just chips but "AI output per watt" — how much useful work a system produces from a fixed power budget. Nvidia's systems are seen as efficient on that measure, which is why the bank kept the stock at the top of its semiconductor rankings.

Not everyone is protected. If data centers cannot be energized, customers may delay or cancel orders, with makers of memory, optical and power-management components most exposed. The power crunch means the AI buildout's pace is now set less by chip supply and more by the electricity grid.

Reporting based on: www.reuters.com, finimize.com.