AI Sector Debates GPU Depreciation Amid $1 Trillion Buildout

Executives and investors in the artificial intelligence sector are debating the depreciation rates of graphics processing units as companies invest nearly $1 trillion in data center buildouts. The dispute centers on whether AI servers remain useful for six years or face obsolescence in as little as two to three years due to rapid technological advancement.

Nvidia Corporation has accelerated the pace of hardware updates by shifting to an annual chip release cycle. CEO Jensen Huang suggested that the launch of Blackwell chips in volume could quickly render previous generations, such as the Hopper architecture, obsolete.

Cloud providers have taken varying approaches to these assets. Amazon recently reduced the estimated useful life of some servers from six to five years. Meanwhile, infrastructure giants like Google, Oracle, and Microsoft have suggested a lifespan of up to six years. CoreWeave CEO Michael Intrator maintains that older A100 and H100 chips retain significant value and remain fully booked. Conversely, short seller Michael Burry argues that companies are overstating the useful life of AI chips to artificially inflate their earnings.