Google has entered into agreements with two U.S. electric utilities, Indiana Michigan Power and Tennessee Power Authority, to reduce power consumption at its AI data centers during periods of peak electricity demand. This marks Google's first participation in a demand-response program aimed at easing strain on the U.S. electrical grid amid surging energy use driven by AI workloads. Under the agreements, Google will pause non-essential AI workloads during times of grid stress caused by high demand or adverse weather conditions that limit power supply. The move addresses concerns over the growing energy demands of AI data centers, which have increasingly strained utility authorities. Industry observers note that the rapid expansion of AI infrastructure is driving sales and stock prices for equipment manufacturers but also raising questions about the long-term sustainability and capital costs of such build-outs. Analysts warn that the heavy investment in AI data centers, which may cost nearly $3 trillion, could face challenges as newer AI chips render older infrastructure less valuable or obsolete.
What’ll happen if we spend nearly $3tn on data centres no one needs? https://t.co/bWpDlBziJq Remember, AI data centers are not an appreciating asset. As new AI chips improve, old AI chips become less valuable, and maybe obsolete. So, don't lend against them.
Netscape’s Lessons for AI Mania by @andykessler https://t.co/erOrnZeMx5 Again, Capex booms rarely work out well for investors, but they may help the economy eventually.
The build-out of AI infrastructure is costing a fortune, straining companies and capital markets, writes the WSJ’s Greg Ip https://t.co/12lMG19Fpb Capex booms rarely pay off.