AI's Power Requirements Under Exponential Growth

The exponential growth of artificial intelligence (AI) systems is driving unprecedented demands for power that could overwhelm existing infrastructure. If not addressed, U.S. companies may have to relocate AI operations overseas, jeopardizing national competitiveness and security, per commentary from RAND Corporation. 

The issue:  

AI systems are generating immense power requirements, potentially reaching 68 gigawatts (GW) by 2027, which exceeds the total global capacity of only 88 GW in 2022. For instance, a single AI training run could demand up to 1 GW by 2028, leading to significant infrastructure challenges.  

What they recommend:  

Experts recommend modeling future power supply against growing data center demand while exploring efficiency improvements in AI hardware to lessen power needs. They also suggest examining permitting bottlenecks and evaluating new power sources capable of supporting AI workloads.  

Go deeper:  

Recent findings indicate that U.S. data centers face extensive permitting delays, with some projects taking four to seven years for grid connections in critical regions. As U.S. companies seek better power availability abroad, this could enhance the compute capabilities of other nations, presenting economic and military advantages. Without swift action, the U.S. may lag in the global AI race amidst tightening power constraints.  

This is a brief overview of a report from RAND Corporation. For complete insights, we recommend reading the full report.