Energy Costs Impose Hidden Tax on AI Computing as Cloud Giants Turn to Nuclear Power and Efficiency to Defend Margins

AI data centers consume vast amounts of electricity, making power expenses a critical factor in computing costs and cloud margins. If higher oil prices drive up generation and grid costs, Amazon, Microsoft and Google will face the equivalent of a “hidden tax” as they expand AI infrastructure. Competition is also shifting beyond chips and data-center scale to computing performance per watt.

Recent reporting indicates that the three cloud providers are seeking to defend margins by restarting nuclear facilities or procuring nuclear power, improving data-center efficiency and developing more efficient models such as GPT-5.4. However, the event data provides no oil prices, electricity rates, investment amounts or specific announcement dates. The confirmed turning point is that the industry has shifted its key metric from total computing capacity to “computing performance per watt.”